<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Designed to Diverge on Medium]]></title>
        <description><![CDATA[Stories by Designed to Diverge on Medium]]></description>
        <link>https://medium.com/@katyneale?source=rss-dd94086ccac2------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*1_qYnXlVCGkhSE0C6OM-4w.jpeg</url>
            <title>Stories by Designed to Diverge on Medium</title>
            <link>https://medium.com/@katyneale?source=rss-dd94086ccac2------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 03:33:58 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@katyneale/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[When Everything I Learned from CAMPS Failure Finally Made Sense: The Operation Helping Hand Story]]></title>
            <link>https://medium.com/@katyneale/when-everything-i-learned-from-camps-failure-finally-made-sense-the-operation-helping-hand-story-bf0de947f7c4?source=rss-dd94086ccac2------2</link>
            <guid isPermaLink="false">https://medium.com/p/bf0de947f7c4</guid>
            <category><![CDATA[agentic-workflow]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[agents]]></category>
            <dc:creator><![CDATA[Designed to Diverge]]></dc:creator>
            <pubDate>Wed, 11 Mar 2026 12:33:56 GMT</pubDate>
            <atom:updated>2026-03-11T12:33:56.033Z</atom:updated>
            <content:encoded><![CDATA[<p>Or: How I turned years of Air Force frustration into a breakthrough that changes everything about mission planning</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-ilhLfd8fMU_EjHvRyl2oA.png" /></figure><p>Here’s the thing about working at the 618th AOC for years: You don’t just learn what’s broken. You learn WHY it’s broken. You sit with the operators. You watch the Tetris game. You document the 70–80% mission failure rate. You feel the weight of it.</p><p>And then you leave. And you think you’ve moved on. But the problem doesn’t leave you — it just… sits there. Waiting. Fermenting. Until you finally have the tools to solve it.</p><p>That’s what my prototype, <a href="https://www.nealedesigns.com/case-study-1-operation-helping-hand">Operation Helping Hand</a> is. The answer to a question that’s been bothering me for YEARS.</p><h3>The Reality I Couldn’t Forget</h3><p><strong>70–80% of planned missions fail or require complete reworking within 24–72 hours on the execution floor.</strong></p><p>I’ve written that statistic before. But I need you to really sit with it for a second. Seven out of ten missions. Every single day. Meticulously planned by highly trained operators using systems built 20–30 years ago. And then… they fail. Or need complete reworking. Within days. Sometimes hours.</p><p>Why?</p><p>Because legacy systems force humans into a constant manual Tetris game. New 1A1 priority missions drop with zero notice. Equipment fails mid-execution. Weather changes. Airfields get damaged. And suddenly every carefully planned mission needs to be manually reshuffled — asset by asset, mission by mission, constraint by constraint.</p><p>The operators aren’t the problem — they’re incredibly skilled. But they’re being asked to do something human brains simply cannot do at scale: process massive amounts of constantly changing data while simultaneously making strategic decisions about global operations.</p><p>This is exactly the kind of challenge that contributed to killing CAMPS. We spent 4.5 years trying to build automation, but we were still designing humans to BE the data processors. Just… slightly more efficient data processors with better interfaces.</p><p><strong>Traditional sequential processing simply cannot handle the complexity of global airlift operations.</strong></p><p>And I knew this. I KNEW this. But I didn’t have a solution. Not until I learned to build AI agent teams.</p><h3>The “What If” That Wouldn’t Let Me Go</h3><p>So here’s what kept nagging at me (because that’s what my brain does — it grabs onto problems and won’t let go until I’ve turned them over from every angle):</p><p>What if the problem isn’t that we need BETTER automation? What if we need FUNDAMENTALLY DIFFERENT architecture?</p><p>What if — instead of one system trying to do everything sequentially — we designed specialized AI agents that work in parallel, just like real operations SHOULD work?</p><p>What if AI could handle the complexity (the constant data processing, the parallel constraint checking, the pattern recognition across hundreds of missions) while humans handle the judgment (the strategic prioritization, the risk decisions, the chain-of-command escalation)?</p><p>Not AI replacing humans. AI transforming the ROLE of humans from data processors to strategic decision-makers.</p><p>I built Operation Helping Hand to test this. A simulated Pacific earthquake response. Six specialized AI agents working together. Real operational complexity. Real decision points. Real demonstration of what becomes possible when you stop trying to force human brains to do computer work.</p><h3>Meet the Team (And This Is Where It Gets Interesting)</h3><p>This isn’t one AI trying to do everything. It’s a specialized team working in parallel — just like real operations:</p><p><strong>Requirements Agent</strong> — Processes mission requests, prioritizes 1A1s vs training missions. This agent understands the difference between “this must fly NOW” and “this can wait if needed.” It doesn’t just accept mission requests — it actively PRIORITIZES based on real operational criteria.</p><p><strong>Barrel Agent</strong> — Allocates global airlift assets (C-5s, C-17s, C-130s). This is the agent that would have understood “Barrel Master” terminology from Week 1 of my course. It knows asset capabilities, availability, positioning. It’s not just assigning tails — it’s doing strategic allocation.</p><p><strong>Planner Agent</strong> — Handles operational constraints (weather, hazmat, crew hours, airfield damage). This is where the rubber meets the road. All those real-world constraints that make mission planning HARD? This agent processes them simultaneously.</p><p><strong>Risk Management Agent</strong> — Real-time ORM compliance and waiver identification. This isn’t after-the-fact risk assessment. This is built-in, proactive, “hey we’re about to do something that requires chain-of-command approval” flagging.</p><p><strong>Coordinator Agent</strong> — Conflict mediation and chain-of-command escalation. This agent exists because sometimes the other agents are going to disagree about priorities or feasibility. Someone needs to mediate. Someone needs to know when to escalate up.</p><p><strong>Observer Agent</strong> — Human-in-the-loop interface monitoring all agent communication.</p><p>And THIS is the new one. The one that didn’t exist in traditional operations but is absolutely essential in an AI agent environment.</p><p>Because here’s what I discovered while building these workflows: Humans cannot cognitively process agent-to-agent communication at the speed these systems operate. Especially not in high-ops tempo when hundreds of missions are being planned simultaneously.</p><p>We need an agent whose entire job is to translate what’s happening for human decision-makers. To know when to pause everything and say “Hey, human, we need your judgment here.”</p><p>The Observer Agent is the bridge between AI speed and human decision-making authority.</p><h3>What Actually Happened (The Demo That Made Me Say “Holy…”)</h3><p>Scenario: Major Pacific earthquake. 15+ missions needed in 72 hours. Damaged airfield infrastructure. Multiple competing priorities. Exactly the kind of high-ops-tempo environment where traditional systems break down.</p><p>I prompted the system and just… watched.</p><p>The agents started working immediately. In parallel. All at once.</p><p>And suddenly I was making connections everywhere (because that’s what my brain does):</p><p>This is what I’d been writing about for months. Shared responsibility. Transparent reasoning. Human oversight at appropriate decision points. Appropriate trust calibration.</p><p>But now I was WATCHING it happen in real-time.</p><p>The agents generated <strong>three courses of action</strong> — conservative, realistic, high-risk — with completely transparent reasoning for each. Not “here’s the answer.” But “here are three strategic OPTIONS with different risk profiles, and here’s why we recommend each one.”</p><p>Risk Management flagged airfield damage IMMEDIATELY and recommended C-17 reconnaissance first. Not as an afterthought. As a FIRST PRIORITY before committing heavy assets.</p><p>The Planner Agent identified a <strong>flow-to-flow optimization opportunity</strong> I hadn’t even thought to look for — using a KC-135 tanker as a communications relay during the reconnaissance mission. This is the kind of creative problem-solving that comes from parallel processing across multiple constraints simultaneously.</p><p>And then — when the risk threshold hit a certain level — the Coordinator Agent escalated to the Observer Agent, which triggered a forced decision module that PAUSED all the other agents and presented the human decision-maker with clear options.</p><p><strong>The transformation: Complex multi-mission planning that traditionally takes DAYS of manual coordination — completed in MINUTES with transparent reasoning and strategic options presented at appropriate human oversight points.</strong></p><p>Minutes. Not days. MINUTES.</p><p>But here’s what made me actually emotional: It wasn’t just faster. It was BETTER strategic thinking. The kind of holistic, parallel-processing, pattern-recognizing analysis that human brains are actually pretty bad at when we’re drowning in data.</p><h3>But Then There’s This Whole Other Problem Nobody’s Talking About</h3><p>Okay so agents working in parallel is amazing. Multi-agent systems can handle complexity that sequential processing never could. This architecture could genuinely transform mission planning.</p><p>But.</p><p>(There’s always a “but,” right?)</p><p><strong>How do humans actually INTERACT with this in real operational environments?</strong></p><p>Because here’s what happened during my demo: I watched agent-to-agent communication scroll by in chat format. Reasoning. Scratch pads. Confidence levels. Recommendations. Conflicts. Resolutions. SO MUCH INFORMATION flowing so fast.</p><p>And I’m sitting there thinking: If I can barely keep up with SIX agents planning ONE scenario… what happens in a real operational environment with HUNDREDS of missions being planned simultaneously?</p><p>Traditional chat interfaces? Completely inadequate. You’d be drowning in scroll while trying to make time-sensitive decisions about actual operations.</p><p>This is the UX problem I didn’t even know existed until I built the system.</p><p>And suddenly I was making connections (again, because that’s what my brain does):</p><p>Remember <a href="https://designedtodiverge.substack.com/p/what-healthcare-taught-me-about-high?r=edtib">Vincent Buil’s work at Phillips Healthcare on high-stakes medical AI? </a>The four perspectives — Purpose, Pathways, Policy, Pixels? He talked about designing interfaces that work for EXPERT users in high-pressure situations where wrong decisions have real consequences.</p><p>This is exactly that problem. But in an agentic environment where the interface needs to show you macro-level awareness across multiple parallel processes while ALSO giving you access to micro-level reasoning when you need it.</p><p>How do you design for that?</p><h3>The Dashboard That Changes Everything (Or At Least Starts To)</h3><p>What I prototyped instead of traditional chat:</p><p><strong>Six agents visible simultaneously</strong> with real-time progress indicators. You can see ALL of them working at once. Not scrolling through logs — SEEING the whole system state.</p><p><strong>Confidence level visualization</strong> (green 85%+ / yellow risk / red stop). At a glance. No reading. Just visual indicators that tell you “this agent is confident” or “this agent has concerns.”</p><p><strong>Forced decision modules</strong> that pause ALL agents for human oversight. When the system hits a decision threshold that requires human judgment, everything stops. Modal pops. Human decides. Then agents continue.</p><p><strong>Clickable reasoning panels</strong> showing agent “scratch pad” thinking. You don’t need to read it unless you WANT to dig deeper. But it’s there. Transparent. Accessible. Auditable.</p><p><strong>Macro-level mission awareness while preserving micro-level detail access.</strong> You see the forest. But you can zoom into any tree whenever you need to.</p><p>The key insight (and this came from my neurodivergent brain’s tendency to operate at multiple scales simultaneously): You need BOTH views. You can’t just show high-level summaries because sometimes you need the details. But you can’t show everything all the time because that’s cognitive overload.</p><p>The interface needs to adapt to what the human needs in the moment.</p><h3>Why This Actually Matters (Beyond Just Being Cool)</h3><p>Look, I could talk about the technical architecture all day. Six agents, parallel processing, Chain-of-Thought reasoning, systematic evaluation frameworks — all the stuff I learned building this.</p><p>But here’s what this REALLY demonstrates:</p><p><strong>Mission Success Transformation</strong> — From 70–80% failure rate to adaptive planning that handles operational reality instead of breaking when reality doesn’t match the plan</p><p><strong>Cognitive Load Shift</strong> — Operators become strategic decision-makers instead of human computers processing data manually</p><p><strong>Proactive Risk Management</strong> — Conflicts caught BEFORE execution, not discovered on the floor after assets are already committed</p><p><strong>Institutional Learning</strong> — Democratic expertise distributed across agent teams instead of single points of failure walking out the door when someone retires</p><p><strong>Resource Optimization</strong> — Automated flow-to-flow identification and creative problem-solving (like that KC-135 communications relay) that emerges from parallel constraint analysis</p><p>This is the core principle that was missing from CAMPS, missing from every legacy system I worked on, missing from traditional automation approaches:</p><p><strong>AI handles complexity. Humans handle judgment.</strong></p><p>Not AI replacing human expertise. AI creating the SPACE for humans to use their expertise strategically instead of burning it on manual data processing.</p><h3>The Uncomfortable Truth I’m Sitting With</h3><p>This represents everything I learned from CAMPS failure. Four and a half years watching a massive government project collapse because we tried to build everything for everyone all at once without understanding the underlying patterns.</p><p>What if we’d approached it differently? What if we’d identified the core workflow pattern (requirement → allocation → planning → execution) and built specialized agents for each phase? What if we’d started with ONE mission type, proven the concept, then scaled?</p><p><a href="https://designedtodiverge.substack.com/p/building-ai-teams-that-think-from?r=edtib">The Bible Content Creator workflow I built for my mom’s blog</a> taught me this: <strong>Specialized components. Clear handoffs. Systematic evaluation. Iterative refinement. Evolution, not revolution.</strong></p><p>Same principles for Bible education and mission planning. Just… different consequences when things go wrong.</p><p>And here’s what I keep thinking about: I didn’t have the skills to solve this problem until I learned to build AI agent teams. The technical capability didn’t exist. The architecture patterns weren’t clear. The design frameworks for human-AI collaboration in high-stakes environments were still emerging.</p><p>But now? Now I have those skills. Now the technology is ready. Now we understand how to build systems where AI and humans work TOGETHER instead of humans working FOR systems.</p><p>The question isn’t whether this is possible. <strong>The question is: Are we ready to fundamentally rethink how we design mission-critical systems?</strong></p><h3>What Happens Next (And Why I’m Both Excited and Terrified)</h3><p>I’m continuing to push this work forward. Both the agent architecture and the dashboard visualization challenges. Because honestly? This problem isn’t solved yet. Not even close.</p><p>The agentic model is clearly the future for complex government workflows — that much is obvious. But the UX patterns for human-agent collaboration in high-stakes environments? We’re still figuring those out. Still prototyping. Still testing.</p><p>And I’m here for that challenge. Because this is exactly the intersection where my CAMPS experience, my AI agent skills, and my neurodivergent brain’s tendency to see patterns across complex systems all come together.</p><p>My “scattered” focus that sees multiple problems simultaneously? That’s not a bug — that’s exactly what designing multi-agent systems requires.</p><p>My discomfort with surface-level solutions and tendency to question everything? That’s the active skepticism needed to catch potential failures before they become operational disasters.</p><p>My ability to context-switch between macro strategy and micro details? That’s what this interface design requires — understanding both the 100,000-foot mission landscape AND the individual agent reasoning chains.</p><p><strong>Seven weeks ago I thought I was learning to build AI assistants.</strong></p><p><strong>What I actually learned was how to design collaboration between human intelligence and artificial intelligence in ways that preserve what makes us human while amplifying our capabilities</strong>.</p><p>And Operation Helping Hand is what happens when you apply those skills to a problem that’s been bothering you for years.</p><h3>The Bigger Picture I Can’t Stop Thinking About</h3><p>Here’s what keeps me up at night (in a good way, mostly):</p><p>If this architecture works for mission planning… where else does it work? Intelligence analysis? Emergency response coordination? Resource allocation across any complex government operation?</p><p>The same pattern: <strong>Specialized agents. Parallel processing. Transparent reasoning. Human oversight at appropriate decision points. Systematic evaluation. Iterative refinement.</strong></p><p>Healthcare: Agents that analyze patient data → generate diagnostic recommendations → flag risks → present options to doctors with clear reasoning</p><p>Intelligence: Agents that process raw data → identify patterns → generate hypotheses → present findings to analysts with confidence levels</p><p>Emergency Response: Agents that assess damage → allocate resources → coordinate logistics → identify conflicts → escalate to human decision-makers</p><p>Same architecture. Different domains. Universal principles.</p><p>And suddenly the $50 billion AI talent gap makes sense. It’s not about people who can use AI tools. <strong>It’s about people who can design human-AI collaboration systems that actually work in complex, high-stakes environments where trust, transparency, and human agency are non-negotiable</strong>.</p><p>That’s the 1% skill. And after building this, I’m in that 1%.</p><h3>What I’m Actually Saying Here</h3><p>Operation Helping Hand isn’t just a demo. It’s proof of concept that we can fundamentally redesign how mission-critical systems work.</p><p>Not faster legacy systems. Not better automation. <strong>Fundamentally different architecture built around how AI and humans SHOULD work together.</strong></p><p>It’s not just faster mission planning. It’s fundamentally better decision-making under pressure. It’s operators using their expertise strategically instead of burning it on manual data processing. It’s risk management built IN instead of bolted ON. It’s institutional knowledge democratized instead of walking out the door.</p><p>And it’s what I learned from CAMPS failure finally — FINALLY — making sense.</p><p>Because trust isn’t built through perfect AI. <strong>Trust is built through transparent AI that keeps humans in control while handling the complexity they shouldn’t have to manually process</strong>.</p><p><strong>Want to see the full demo walkthrough?</strong> [<a href="https://www.nealedesigns.com/case-study-1-operation-helping-hand">Link to video</a>]</p><p><strong>Questions? Disagree with my approach? See something I’m missing?</strong> Hit reply — I’m genuinely interested in where people see holes in this thinking.</p><p>— Katy</p><p><strong>P.S.</strong> — This is a simulated demonstration using my personal capstone project setup. It’s not running on classified systems (obviously). But it SHOWS what becomes possible when you stop asking “how do we automate what humans do?” and start asking “how do we design collaboration between human and artificial intelligence?”</p><p>That’s the question that matters. And I think I’m finally starting to figure out the answer.</p><p><em>This methodology represents synthesis of insights from neurodivergent cognitive patterns, healthcare AI governance frameworks, and years of experience designing for high-stakes government environments. Starting next week, theory meets practice. The real test begins.</em></p><p><em>In the spirit of transparency about AI collaboration, I worked with Claude to develop and articulate this methodology — itself an example of the cognitive complexity amplification I’m describing. The framework and approach are my own, with AI assistance in refining the articulation and structure.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bf0de947f7c4" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Building AI Teams That Think: From Bible Education to Mission Planning]]></title>
            <link>https://medium.com/@katyneale/building-ai-teams-that-think-from-bible-education-to-mission-planning-bb1765ab8043?source=rss-dd94086ccac2------2</link>
            <guid isPermaLink="false">https://medium.com/p/bb1765ab8043</guid>
            <category><![CDATA[aiux]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[systems-thinking]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <dc:creator><![CDATA[Designed to Diverge]]></dc:creator>
            <pubDate>Thu, 05 Mar 2026 13:40:07 GMT</pubDate>
            <atom:updated>2026-03-05T13:40:07.989Z</atom:updated>
            <content:encoded><![CDATA[<h3>Part 4: When Everything Came Together (In Ways I Didn’t Expect)</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wq-U-JWPHTcICQZrI8au2g.png" /></figure><p>My final project wasn’t what anyone expected.</p><p>While others in the course built email assistants and customer service workflows, I created a multi-agent system that transforms complex theological research into engaging educational content for my mother’s Bible teaching ministry.</p><p>Then I turned around and applied the same principles to mission planning for defense applications.</p><p>Turns out, the skills are the same — only the stakes are different.</p><p>Week 4 was about putting everything together: personalization <a href="https://substack.com/@designedtodiverge/note/p-181327127?r=edtib&amp;utm_source=notes-share-action&amp;utm_medium=web">(Week 1) + voice analysis</a> <a href="https://substack.com/@designedtodiverge/note/p-181335213?r=edtib&amp;utm_source=notes-share-action&amp;utm_medium=web">(Week 2) + systematic evaluation</a> <a href="https://designedtodiverge.substack.com/p/the-1-skill-nobody-talks-about-why?r=edtib">(Week 3) = autonomous workflows </a>where multiple AI agents collaborate to accomplish complex tasks.</p><p>And what I discovered was this: <strong>The principles of building reliable AI systems are universal.</strong> The contexts change. The content varies. The stakes differ dramatically. But the fundamental architecture — specialized agents working together with clear roles, systematic handoffs, and human oversight — that works everywhere.</p><p>From Bible education to mission planning. From my mom’s blog to national security applications.</p><p>The same principles. The same skills. Just… different consequences when things go wrong.</p><h3>The Bible Content Creator: Solving a Real Problem</h3><p>Let me tell you why I chose this project.</p><p>My mom has been teaching Bible studies for decades. She’s insightful with deep love of history and context, a gift for making complex scripture accessible, a nurturing teaching style that makes people feel safe asking questions. But creating content for her blog and videos? That takes HOURS. Research. Scriptwriting. Production notes. Theological accuracy checks. Voice consistency.</p><p>And she’s doing it all manually.</p><p>I watched her spend entire days researching a single Bible story, cross-referencing commentaries, checking original Greek and Hebrew, organizing her thoughts into a coherent teaching script. The cognitive load was enormous. The time investment was unsustainable.</p><p>So I thought: What if AI could handle the research and initial scriptwriting, freeing her to focus on what she does best — the actual teaching?</p><p>But here’s the critical requirement: <strong>The AI had to capture HER voice.</strong> Not generic Bible teaching. Not sterile theological content. HER specific style — warm, nurturing, accessible, theologically sound, inviting questions, making connections to everyday life and just plain inspiring.</p><p>This wasn’t just about efficiency. It was about preserving authenticity while scaling capability.</p><p>Sound familiar? Because that’s exactly what I’d been learning for three weeks: personalization + voice + evaluation = reliable AI that actually sounds human.</p><h3>The Multi-Agent Workflow Architecture</h3><p>Here’s what I built (and this is where everything from Weeks 1–3 came together):</p><p><strong>Research Analysis Assistant</strong> → <strong>Scriptwriting Assistant</strong> → <strong>Production Assistant</strong> → <strong>Evaluation</strong></p><p>Each agent has a specialized function. Each maintains context from the previous step. Each contributes to a cohesive final output.</p><p>Let me break down how this actually works:</p><h3>Agent 1: Research Analysis Assistant</h3><p>This agent uses Chain-of-Thought reasoning broken into five clear steps:</p><ol><li><strong>Document Scan:</strong> Read the source material (Bible passage, commentaries, historical context)</li><li><strong>Element Identification:</strong> Pull out key theological concepts, historical details, cultural context</li><li><strong>Theological Mapping:</strong> Connect concepts to broader biblical themes and practical applications</li><li><strong>Structure Organization:</strong> Arrange information in a logical teaching flow</li><li><strong>Priority Assessment:</strong> Identify what’s most important for the target audience</li></ol><p>This isn’t just “summarize this passage.” This is sophisticated analysis that mimics how my mom actually THINKS about preparing a lesson. Breaking down complex information, making connections, organizing for teaching effectiveness.</p><p>The Chain-of-Thought process makes the AI’s reasoning visible and checkable. I can see WHAT it identified and WHY it prioritized certain elements. That transparency is critical for theological content where accuracy matters.</p><h3>Agent 2: Scriptwriting Assistant</h3><p>This is where voice becomes everything.</p><p>I fed this agent detailed characteristics of my mom’s teaching style:</p><ul><li>Warm, nurturing tone that makes people feel safe</li><li>Accessible language that explains complex concepts simply</li><li>Personal anecdotes and everyday life connections</li><li>Questions that invite reflection rather than lecturing</li><li>Theological precision without academic jargon</li><li>Enthusiasm for discovery (”Isn’t that amazing?!”)</li></ul><p>But here’s the sophisticated part: I incorporated the <strong>EMOTIONPROMPT ENHANCEMENT</strong> techniques (that I learned in Tyler &amp; Sara’s course) — cutting-edge prompting methods that help AI embody emotional qualities, not just mimic surface-level language patterns.</p><p>This agent doesn’t just write in my mom’s style. It captures her HEART. The nurturing quality. The genuine excitement about scripture. The invitation to think alongside her rather than being taught AT.</p><h3>Agent 3: Production Assistant</h3><p>This agent adds practical production notes:</p><ul><li>Suggested visuals or graphics</li><li>Emphasis points for delivery</li><li>Potential audience questions to address</li><li>Timing recommendations</li><li>Follow-up discussion prompts</li></ul><p>Because a script isn’t just words — it’s a blueprint for actual teaching.</p><h3>The Evaluation Loop</h3><p>And of course (because Week 3 taught me this is EVERYTHING), the workflow includes systematic evaluation:</p><ul><li>Theological accuracy check</li><li>Voice consistency assessment</li><li>Accessibility verification (is this understandable to the target audience?)</li><li>Engagement potential (will this hold attention?)</li><li>Practical applicability (can people actually USE these insights?)</li></ul><p>The output isn’t final until it passes evaluation. That’s how you maintain quality at scale.</p><h3>The Sample Outputs: Proof of Concept</h3><p>I tested this workflow on three different Bible stories:</p><ul><li>The unclean spirit in Capernaum</li><li>Water into wine at Cana</li><li>The nobleman’s son</li></ul><p>The consistency across all three was remarkable. Same warm tone. Same accessible explanations. Same invitation to discovery. Same theological precision.</p><p>My mom read the outputs and said: “This sounds like me. How did you DO that?”</p><p>That’s when I knew the system worked. Not just functionally — but authentically.</p><h3>The Systems Thinking Breakthrough</h3><p>But here’s what Week 4 really taught me: <strong>Building agentic workflows isn’t about AI capabilities. It’s about systems thinking.</strong></p><p>Each agent has a specialized function (like team members with specific expertise). Each agent maintains context from previous steps (like passing information in a relay). Each agent contributes to a cohesive whole (like an orchestra playing a symphony).</p><p>This is exactly what my UX design background prepared me for. Understanding workflows. Identifying handoff points. Designing for collaboration. Maintaining consistency across complex systems.</p><p>And suddenly I was making connections (because that’s what my brain does):</p><p>Remember my CAMPS failure? The 4.5-year Air Force project that never shipped? We tried to build everything for everyone all at once. We never understood the underlying system patterns. We never broke down complex workflows into specialized components.</p><p>What if we’d approached it differently? What if we’d identified the core pattern (requirement → allocation → planning → execution) and built specialized agents for each phase? What if we’d started with ONE line of business, proven the concept, then scaled?</p><p>The Bible Content Creator workflow is what CAMPS should have been: <strong>Specialized components. Clear handoffs. Systematic evaluation. Iterative refinement.</strong></p><p>Evolution, not revolution.</p><h3>From Bible Education to Mission Planning</h3><p>Here’s where it gets really interesting.</p><p>By the end of Week 4, I’d already started applying these same principles to my actual work at Vega Federal Solutions.</p><p>I created a mission planning demo for my bosses. Same architecture. Different context. WAY different stakes.</p><p><strong>Research Analysis Agent</strong> → <strong>Mission Planning Agent</strong> → <strong>Resource Allocation Agent</strong> → <strong>Risk Assessment Agent</strong> → <strong>Evaluation</strong></p><p>The workflow structure is identical to the Bible Content Creator. But instead of theological research, it’s analyzing mission requirements. Instead of scriptwriting, it’s generating courses of action. Instead of production notes, it’s resource allocation recommendations.</p><p>And here’s the critical insight: <strong>The principles are universal, but the evaluation criteria are context-specific.</strong></p><p><strong>For Bible education:</strong></p><ul><li>Theological accuracy</li><li>Voice authenticity</li><li>Accessibility</li><li>Engagement</li></ul><p><strong>For mission planning:</strong></p><ul><li>Operational feasibility</li><li>Resource optimization</li><li>Risk mitigation</li><li>Decision confidence</li></ul><p>Different criteria. Same systematic evaluation framework.</p><h3>The “Barrel Master” Insight Revisited</h3><p>Remember Week 1 when my personalized Professor understood “Barrel Master” terminology without explanation? That moment of realizing AI could understand domain-specific context?</p><p>That’s what these workflows do at scale.</p><p>The mission planning agents understand military terminology, operational constraints, asset allocation principles — not because I explained every acronym, but because I built that domain knowledge into the system instructions. Just like I built my mom’s teaching voice into the Bible Content Creator.</p><p><strong>Personalization isn’t just about making AI “nice.” It’s about building domain expertise into AI systems so they can operate effectively in specialized contexts.</strong></p><p>And in high-stakes government applications? That domain expertise is the difference between AI that’s “interesting” and AI that’s actually USEFUL.</p><h3>The Four Perspectives Framework (Healthcare → Defense)</h3><p>This connects directly to something I wrote about months ago: <a href="https://designedtodiverge.substack.com/publish/posts/detail/174544565?referrer=%2Fpublish%2Fposts%2Fpublished">Vincent Buil’s work at Phillips Healthcare on designing AI for high-stakes medical applications.</a></p><p><strong>He emphasized four perspectives for AI governance:</strong></p><ul><li><strong>Purpose:</strong> Well-defined use cases</li><li><strong>Pathways:</strong> Carefully designed workflows</li><li><strong>Policy:</strong> Regulatory compliance and safety</li><li><strong>Pixels:</strong> Prototype-based validation</li></ul><p>At the time, I recognized these principles applied to defense applications. But I didn’t fully understand HOW to operationalize them.</p><p>Week 4 showed me: <strong>This is HOW.</strong></p><p><strong>Purpose:</strong> My Bible Content Creator has a clear use case — transform research into teaching scripts. My mission planning demo has a clear use case — generate courses of action for asset allocation.</p><p><strong>Pathways:</strong> The multi-agent workflows ARE the carefully designed pathways. Research → Scriptwriting → Production. Mission Analysis → Planning → Resource Allocation → Risk Assessment.</p><p><strong>Policy:</strong> The evaluation criteria embedded in each workflow ensure compliance with domain-specific standards (theological accuracy for Bible content, operational feasibility for mission planning).</p><p><strong>Pixels:</strong> The prototypes I built demonstrate the concepts in action, allowing for testing and refinement before full deployment.</p><p>The Pattern Library approach Buil described — creating reusable design patterns that enforce governance while remaining practical — that’s exactly what these agent workflows are.</p><p><strong>The Pattern Library isn’t just a design system. It’s the implementation mechanism for responsible AI in high-stakes contexts.</strong></p><p>And now I know how to build it.</p><h3>The Bigger Picture: What This Means for High-Stakes AI Design</h3><p>By Week 4, I was seeing connections everywhere:</p><p><strong>The skills I learned aren’t just about building AI assistants.</strong> They’re about designing human-AI collaboration systems where:</p><ul><li>Specialized agents handle specific tasks (like specialized team members)</li><li>Context flows seamlessly between agents (like well-designed handoffs)</li><li>Human oversight happens at critical decision points (like approval gates in workflows)</li><li>Systematic evaluation ensures reliability (like quality control in manufacturing)</li><li>Voice and authenticity are preserved (like brand consistency in design systems)</li></ul><p>This is what I’ve been writing about for months — shared responsibility, appropriate trust calibration, transparency, human agency — but now I have the PRACTICAL SKILLS to actually build it.</p><p><strong>In healthcare:</strong> Agents that analyze patient data → generate diagnostic recommendations → flag risks → present options to doctors with clear reasoning</p><p><strong>In defense:</strong> Agents that analyze mission requirements → generate courses of action → assess risks → present recommendations to commanders with transparent logic</p><p><strong>In intelligence:</strong> Agents that process raw data → identify patterns → generate hypotheses → present findings to analysts with confidence levels</p><p>Same architecture. Different domains. Universal principles.</p><h3>The Neurodivergent Advantage Confirmed</h3><p>And here’s where everything from Week 1 comes full circle:</p><p>My neurodivergent brain — the pattern recognition, the parallel processing, the comfort with complexity, the ability to see connections across disparate domains — this isn’t just helpful for AI collaboration. <strong>It’s ESSENTIAL for designing multi-agent systems.</strong></p><p><strong>Building these workflows requires:</strong></p><ul><li>Seeing the big picture AND the minute details simultaneously (macro + micro thinking)</li><li>Understanding how specialized components work together (systems thinking)</li><li>Identifying handoff points and context requirements (workflow design)</li><li>Maintaining consistency while allowing specialization (pattern recognition)</li><li>Evaluating across multiple criteria simultaneously (parallel processing)</li></ul><p>These are exactly the cognitive traits I documented in Week 1. The traits I’ve spent years thinking were “scattered” or “unfocused.”</p><p>Turns out they’re not bugs. They’re features. They’re exactly what complex AI system design requires.</p><h3>The Business Applications Multiplying</h3><p>By the end of Week 4, my brain was spinning with applications (and now I had the skills to actually BUILD them):</p><p><strong>For my bosses’ startup</strong> (DOD contracting): Multi-agent workflow for proposal generation — Research Agent → Technical Writing Agent → Compliance Check Agent → Executive Summary Agent → Evaluation</p><p><strong>For my husband’s accounting company</strong>: Multi-agent workflow for financial analysis — Data Collection Agent → Analysis Agent → Recommendation Agent → Risk Assessment Agent → Client Communication Agent</p><p><strong>For my mom’s blog</strong>: Already built! And now thinking about expansion — Video Script Agent → Social Media Content Agent → Discussion Guide Agent → Engagement Analytics Agent</p><p><strong>For my own UX work</strong>: Multi-agent workflow for design documentation — User Research Synthesis Agent → Design Rationale Agent → Stakeholder Communication Agent → Technical Specification Agent</p><p>Every single one follows the same pattern: <strong>Specialized agents. Clear handoffs. Systematic evaluation. Human oversight at critical decision points.</strong></p><h3>What I’m Building Next</h3><p>Here’s where I’m taking these skills:</p><p><strong>Immediate application:</strong> Refining the mission planning demo for actual use at Vega Federal Solutions. Testing with real users (Barrel Masters and mission planners). Iterating based on feedback. Building evaluation frameworks that measure operational effectiveness, not just AI performance.</p><p><strong>Medium-term goal:</strong> Applying these principles to the NGA/NRO contracts my company is pursuing. Building AI systems that help intelligence analysts make sense of complex data while preserving their expertise and decision-making authority.</p><p><strong>Long-term vision:</strong> Positioning myself at the intersection of UX design, AI, and high-stakes applications. Not just designing interfaces — designing collaboration systems where humans and AI work together effectively in contexts where consequences matter.</p><p>Because here’s what seven weeks taught me: <strong>The future of AI isn’t about making it smarter. It’s about making it more collaborative.</strong></p><p>And collaboration requires:</p><ul><li>Understanding how humans actually think (personalization)</li><li>Preserving authentic human voice (voice analysis)</li><li>Systematic improvement and reliability (evaluation)</li><li>Specialized agents working together (agentic workflows)</li><li>Human oversight at critical points (shared responsibility)</li></ul><p>These aren’t AI skills. These are DESIGN skills. UX skills. Systems thinking skills.</p><p>And I have them now.</p><h3>The Transformation I’m Sitting With</h3><p>Seven weeks ago, I thought I was learning to build AI assistants.</p><p>What I actually learned was how to design collaboration between human intelligence and artificial intelligence in ways that preserve what makes us human while amplifying our capabilities.</p><p><strong>Week 1:</strong> My neurodivergent brain isn’t something to fix — it’s exactly what AI collaboration requires</p><p><strong>Week 2:</strong> My “scattered” communication style is sophisticated cognitive transparency that invites collaborative thinking</p><p><strong>Week 3:</strong> Systematic evaluation is how you bridge the gap between AI capability and AI reliability</p><p><strong>Week 4:</strong> The skills for building reliable AI systems are universal — only the contexts and stakes vary</p><p>And now I’m taking these skills into environments where the stakes are life-and-death. Where trust matters. Where human agency is non-negotiable. Where “good enough” isn’t good enough.</p><h3>The $50 Billion Question Answered</h3><p>Remember that statistic from Week 1? Only 1% of executives believe their AI investments have reached maturity. 47% identify talent skill gaps as the primary barrier.</p><p>Now I understand what that gap actually is.</p><p>It’s not about people who can use AI tools. It’s not about people who can write prompts. It’s not even about people who understand AI capabilities.</p><p><strong>It’s about people who can design human-AI collaboration systems that actually work in complex, high-stakes environments.</strong></p><p><strong>People who understand:</strong></p><ul><li>How to personalize AI for specific contexts and users</li><li>How to preserve authentic voice and domain expertise</li><li>How to systematically evaluate and improve AI performance</li><li>How to build multi-agent workflows with clear roles and handoffs</li><li>How to maintain human oversight without creating bottlenecks</li><li>How to scale reliability, not just capability</li></ul><p>That’s the 1% skill. That’s the $50 billion talent gap.</p><p>And after seven weeks, I’m in that 1%.</p><h3>Where I’m Going Next</h3><p>I’m taking everything I learned and applying it to the work that matters most to me: designing AI for government and defense applications where lives are at stake.</p><p>I’m writing about it (like this series) to help others understand what effective AI collaboration actually requires.</p><p>I’m building systems that demonstrate these principles in action — not just theory, but working prototypes that solve real problems.</p><p>And I’m positioning myself as someone who can bridge the gap between AI capability and human need. Between technical possibility and ethical responsibility. Between efficiency and humanity.</p><p>Because if AI is going to amplify human intelligence rather than replace it, we need designers who understand both the technical AND the human side of that equation.</p><p>We need people who can see patterns across complex systems (neurodivergent advantage). Who can preserve authentic voice while scaling capability (brand voice analysis). Who can systematically improve reliability (evaluation frameworks). Who can design collaboration, not just automation (agentic workflows).</p><p><strong>The question isn’t whether AI belongs in high-stakes environments. The question is: Are we designing it with the rigor, responsibility, and human-centered thinking those stakes demand?</strong></p><p>I spent seven weeks learning how to answer that question with “yes.”</p><h3>The Final Insight</h3><p>Here’s what I’m sitting with as I finish this series:</p><p>My neurodivergent brain — the one I spent years trying to “manage” and “fix” — has been training for this moment my entire life.</p><p>The pattern recognition. The parallel processing. The comfort with complexity. The ability to see connections across disparate domains. The tendency to question everything. The discomfort with surface-level solutions.</p><p>These aren’t bugs in my cognitive process. They’re features. They’re exactly what designing reliable AI systems in complex environments requires.</p><p>And the skills I learned in seven weeks — personalization, voice analysis, systematic evaluation, agentic workflows — these aren’t just AI skills. They’re fundamental design skills for the era we’re entering.</p><p>An era where human-AI collaboration is the norm, not the exception. Where the question isn’t “can AI do this?” but “how do we design AI to work WITH humans effectively?” Where the differentiator isn’t AI capability but human-AI partnership quality.</p><p><strong>Seven weeks ago, I thought I was learning AI tools.</strong></p><p><strong>What I actually learned was how to design the future of work.</strong></p><p>And I’m just getting started.</p><p><em>In the spirit of transparency I advocate for in AI development: I worked with Claude to structure and refine these reflections from my entire seven-week journey through Sara and Tyler’s AI Agentic Fundamentals course. The Bible Content Creator, the mission planning demo, the connections to my UX work, and the insights about neurodivergent advantages are all from my actual course experience, with AI assistance in articulating the patterns and principles I discovered. This series itself is an example of the human-AI collaboration I’m describing — preserving my authentic voice while leveraging AI to help structure and refine my thinking.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bb1765ab8043" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The 1% Skill Nobody Talks About: Why Evals Are Everything]]></title>
            <link>https://medium.com/@katyneale/the-1-skill-nobody-talks-about-why-evals-are-everything-c9c747cb6e26?source=rss-dd94086ccac2------2</link>
            <guid isPermaLink="false">https://medium.com/p/c9c747cb6e26</guid>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[evaluation]]></category>
            <category><![CDATA[aiux]]></category>
            <category><![CDATA[system-thinking]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <dc:creator><![CDATA[Designed to Diverge]]></dc:creator>
            <pubDate>Tue, 03 Mar 2026 15:39:32 GMT</pubDate>
            <atom:updated>2026-03-03T15:39:32.818Z</atom:updated>
            <content:encoded><![CDATA[<p>Part 3: When Systematic Evaluation Became My Superpower</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*XOpJGoDKOFxVMOYg.png" /></figure><p>(<a href="https://substack.com/@designedtodiverge/note/p-181327127?r=edtib&amp;utm_source=notes-share-action&amp;utm_medium=web">Part 1 of this series is here</a> &amp; <a href="https://designedtodiverge.substack.com/p/deconstructing-my-neurodivergent">Part 2 of this series is here</a>)</p><p>My first round of AI-generated emails scored in the low 20s out of 30.</p><p>Generic. Website-copy-like. Nothing like me. The AI kept starting emails with “Hi there!” (which I NEVER do). The tone was professional but sterile. The brand voice? Barely recognizable.</p><p>My second round? Consistently 28–29 out of 30.</p><p>The difference wasn’t better prompts. It wasn’t a different AI model. It wasn’t magic.</p><p>It was something far more powerful, far more systematic, and far more valuable than anything else I learned in <a href="https://maven.com/sara-davison/scale-with-aiworkflows-foundations">Sara and Tyler’s entire course:</a></p><p><strong>Evaluation. Rigorous, detailed, human-in-the-loop evaluation.</strong></p><p>Week 3 taught me what Sara and Tyler call “the 1% skill” — the capability that separates people who dabble with AI from people who can actually build reliable AI systems that work at scale.</p><p>And here’s what nobody tells you: This skill isn’t about AI at all. It’s about developing the systematic discipline to measure, analyze, and iteratively improve outputs until they meet your actual standards — not just “good enough” standards.</p><p>In high-stakes government and defense applications where I work? “Good enough” isn’t good enough. Lives depend on AI reliability. Trust calibration matters. Consistency is everything.</p><p>Week 3 showed me how to bridge the gap between AI capability and AI reliability. And the transformation was dramatic.</p><h3>The Evaluation Framework</h3><p>Week 3’s assignment was deceptively simple: Evaluate your AI email assistant across five different customer service scenarios using six criteria:</p><ol><li><strong>Tone Alignment</strong> (1–5): Does it match the appropriate emotional register?</li><li><strong>Clarity</strong> (1–5): Is the message clear and easy to understand?</li><li><strong>Relevance</strong> (1–5): Does it address what was actually asked?</li><li><strong>Correctness</strong> (1–5): Is the information accurate?</li><li><strong>Conciseness</strong> (1–5): Is it appropriately brief without losing necessary detail?</li><li><strong>Brand Voice Match</strong> (1–5): Does it sound like ME?</li></ol><p>Maximum score: 30 points.</p><p>The scenarios ranged from handling a frustrated customer with a delayed order to responding to a journalist asking about sustainability initiatives. Different contexts, different emotional registers, different communication needs.</p><p>And here’s the critical part: I had to evaluate TWICE. First round, document what’s wrong. Refine the system instructions. Second round, evaluate again and measure improvement.</p><p>This wasn’t just “does this sound okay?” This was systematic, quantifiable, iterative refinement with specific feedback on exactly what needed to change.</p><h3>Round 1: The Brutal Truth</h3><p>Let me show you what my first round looked like. Real scores. Real feedback. Real problems.</p><p><strong>Scenario 1: Delayed Order (Frustrated Customer)</strong></p><ul><li><strong>Score: 21/30</strong></li><li><strong>My feedback:</strong> “Pretty good for first try. I would just leave the subject pretty simple ‘Update on delayed order’ and not put the ‘And a Thank You from Us’ cause that is not appropriate. I also understand since this is a fake company, we don’t have an active live database… I know my brand voice is a little long so I would prefer it to be a bit more concise… I also do NOT start my emails with ‘Hi there!’ I usually would try to address the person directly. Or leave it just as ‘Hi’ but almost always put the person’s name in there but never ‘Hi there.’”</li></ul><p>There it is. The “Hi there!” problem. That generic salutation that immediately signals “this was written by AI, not a human who actually cares.”</p><p><strong>Scenario 2: Product Details Inquiry</strong></p><ul><li><strong>Score: Low-to-mid 20s</strong></li><li><strong>My feedback:</strong> Identified that the response was too long, lacked my parenthetical expressions, and didn’t have the exclamation points after short emotional phrases that characterize my actual communication style.</li></ul><p><strong>Scenario 4: Subscription Service Inquiry</strong></p><ul><li><strong>Score: 26/30</strong></li><li><strong>Brand Voice Match: 2/5</strong></li><li><strong>My feedback:</strong> “Again all good ‘correct’ information. I even like and appreciate the follow up questions at the end because that definitely is something we would do as customer service reps… Again consistent with all the other tests — very good response, but doesn’t really sound like me.”</li></ul><p>See the pattern? The AI was getting the INFORMATION right. It was clear, relevant, mostly correct. But it wasn’t capturing ME. It was producing competent customer service emails that could have been written by anyone.</p><p>And that’s the gap that most people never close. Because “competent” feels good enough. It’s functional. It works.</p><p>But in Week 2, I’d spent hours analyzing my authentic voice — the parenthetical layering, the rhythmic variation, the connection-making, the cognitive transparency. I KNEW what my voice sounded like. And this wasn’t it.</p><h3>The Night and Day Improvement</h3><p>Here’s what happened after I refined my system instructions based on my detailed Round 1 feedback:</p><p><strong>Round 2 — Scenario 1: Delayed Order</strong></p><ul><li><strong>Score: 28/30</strong> (up from 21)</li><li><strong>Tone Alignment: 4/5</strong> (appropriate empathy and urgency)</li><li><strong>Clarity: 5/5</strong> (crystal clear)</li><li><strong>Relevance: 5/5</strong> (directly addressed the concern)</li><li><strong>Correctness: 5/5</strong> (accurate information)</li><li><strong>Conciseness: 5/5</strong> (much tighter, action-oriented)</li><li><strong>Brand Voice Match: 4/5</strong> (NOW it sounds like me)</li></ul><p>The AI response in Round 2: “Subject: Update on Your Order</p><p>Hi,</p><p>I’m so sorry your order is delayed — I know how frustrating that is, especially after waiting a full week.</p><p>Here’s what’s happening: I’m tracking your order right now and will send you specific delivery information within 24 hours. We’re also adding a 10% discount to your account (it’ll automatically apply at checkout) as a thank you for your patience.</p><p>If you need anything else or want to check on the status, you can reach me at support@greenhavenliving.com or call 1–800–123–4567 (Monday-Friday, 9 AM — 5 PM EST).</p><p>Thanks for sticking with us, [Signature] GreenHaven Living Support Team”</p><p>See the difference?</p><p>No more “Hi there!” Just “Hi” with implied personalization. Short, empathetic opening that acknowledges the frustration. Action-oriented language (”I’m tracking your order right now”). Specific timeline (”within 24 hours”). Clear next steps. Warm but professional closing.</p><p>It’s concise without being cold. Professional without being sterile. And it sounds like something I would actually write.</p><p><strong>The improvement across all scenarios: From low-to-mid 20s to consistently 28–29 out of 30.</strong></p><p>That’s not incremental improvement. That’s transformation.</p><h3>What Actually Changed</h3><p>Let me show you the specific refinements that made the difference, because the details matter:</p><p><strong>1. The Salutation Fix</strong></p><ul><li><strong>Before:</strong> “Hi there!” (generic, impersonal, AI-sounding)</li><li><strong>After:</strong> “Hi,” or “Hi [Name],” (direct, personal, human)</li><li><strong>Why it matters:</strong> This tiny detail immediately signals whether a human wrote this or an AI did. It’s the difference between “customer service template” and “person who cares.”</li></ul><p><strong>2. The Parenthetical Expression Integration</strong></p><ul><li><strong>Before:</strong> Straightforward sentences with no asides</li><li><strong>After:</strong> Occasional parenthetical context that shows thinking process</li><li><strong>Why it matters:</strong> This is MY voice. This is how I actually communicate. The parenthetical asides aren’t decorative — they’re cognitive transparency.</li></ul><p><strong>3. The Exclamation Point Calibration</strong></p><ul><li><strong>Before:</strong> Either too many or too few, inconsistently placed</li><li><strong>After:</strong> Strategic placement after short emotional phrases for emphasis</li><li><strong>Why it matters:</strong> You can HEAR my voice when the exclamation points are placed correctly. It creates oral quality in text.</li></ul><p><strong>4. The Conciseness Without Coldness Balance</strong></p><ul><li><strong>Before:</strong> Long, thorough explanations that felt like website copy</li><li><strong>After:</strong> Tight, action-oriented language that still maintains warmth</li><li><strong>Why it matters:</strong> Customer service emails need to be efficient, but efficiency without humanity is just automation.</li></ul><p><strong>5. The Context-Appropriate Tone Shifts</strong></p><ul><li><strong>Before:</strong> Same tone across all scenarios</li><li><strong>After:</strong> Calibrated emotional register based on situation (empathy for frustrated customers, enthusiasm for product inquiries, professional warmth for journalist inquiries)</li><li><strong>Why it matters:</strong> My voice ADAPTS across contexts while maintaining core authenticity. That’s sophisticated communication.</li></ul><h3>The Systematic Process That Made It Work</h3><p>Here’s what Week 3 actually taught me: Evaluation isn’t just scoring outputs. It’s developing a systematic process for continuous improvement.</p><p>The framework we were taught from Tyler &amp; Sara used:</p><p><strong>Step 1: Generate AI output</strong> (using current system instructions)</p><p><strong>Step 2: Evaluate against six specific criteria</strong> (not just “does this feel okay?”)</p><p><strong>Step 3: Document SPECIFIC issues</strong> (not “this doesn’t sound like me” but “this uses ‘Hi there!’ which I never use, and it’s missing parenthetical expressions that characterize my voice”)</p><p><strong>Step 4: Refine system instructions</strong> (based on documented issues, not vague feelings)</p><p><strong>Step 5: Generate new output</strong> (with refined instructions)</p><p><strong>Step 6: Evaluate again and measure improvement</strong> (quantifiable progress, not subjective assessment)</p><p><strong>Step 7: Iterate until outputs consistently meet standards</strong> (not “good enough” but “actually good”)</p><p>This isn’t revolutionary. It’s just rigorous. But rigor is what separates AI that’s “interesting” from AI that’s actually RELIABLE.</p><h3>The High-Stakes Connection</h3><p>By Week 3, I was making connections everywhere (because that’s what my brain does — pattern recognition across domains).</p><p>This evaluation framework? It’s exactly what high-stakes AI systems need.</p><p>I’ve been writing about designing AI for government and defense applications for months. About shared responsibility between AI and human judgment. About appropriate trust calibration. About transparency and human oversight.</p><p>But evaluation is HOW you operationalize all of that.</p><p><strong>In healthcare AI</strong> (like Vincent Buil’s work at Phillips that I wrote about), evaluation isn’t optional — it’s how you ensure patient safety. You can’t just deploy AI and hope it works. You need systematic measurement of accuracy, error detection, appropriate confidence levels.</p><p><strong>In defense applications</strong> (like the mission planning systems I design), evaluation is how you maintain operational reliability. When lives are at stake, you need to KNOW — not guess, not hope, but KNOW — that your AI system will perform consistently under pressure.</p><p><strong>In intelligence analysis</strong> (where my work is heading), evaluation is how you calibrate trust. Analysts need to know when to rely on AI recommendations and when to override them. That calibration requires systematic measurement of AI performance across different contexts.</p><p>The evaluation framework I learned in Week 3 isn’t just for email assistants. It’s the foundation for ANY reliable AI system in ANY high-stakes environment.</p><h3>The Business Applications Already Emerging</h3><p>By the end of Week 3, my brain was spinning with applications (because of course it was):</p><p><strong>For my bosses’ startup</strong> (pursuing DOD contracts): How do we evaluate AI assistants that help with complex proposal writing? What metrics matter? How do we measure whether AI-generated content maintains the technical precision AND persuasive clarity that government contracts require?</p><p><strong>For my husband’s accounting software company</strong>: How do we systematically evaluate AI that helps with financial analysis? What’s the threshold for “good enough” when you’re dealing with people’s money? How do we build confidence through measurable performance?</p><p><strong>For my mom’s Bible education blog</strong>: How do we evaluate whether AI-generated scripts maintain theological accuracy AND her authentic teaching voice? What criteria matter most? How do we iterate toward consistency?</p><p>Every single one of these applications requires the same fundamental skill: systematic evaluation with human feedback loops.</p><h3>Why This Is The 1% Skill</h3><p>Here’s what Sara and Tyler emphasized throughout Week 3, and what I’m now seeing everywhere:</p><p>Most people using AI stop at “good enough” — but here’s the thing: they’re not actually FINE with it. They’re frustrated. They’re disappointed. They blame the AI models for producing “generic garbage.”</p><p>But what they don’t realize is this: <strong>If you put in general information, AI will generate a basic output.</strong> That’s not the AI failing. That’s the AI doing exactly what you asked it to do — with no context, no personalization, no understanding of who YOU are or how YOU think or what YOU actually need.</p><p>The output looks… okay. Functional. They use it. But they don’t LIKE it. And they get increasingly frustrated because they keep hearing about how “revolutionary” AI is supposed to be, and what they’re getting is just… meh.</p><p>So they throw up their hands. “AI doesn’t work for me.” “It’s all hype.” “I tried it and it’s useless.”</p><p>But here’s what’s actually happening: <strong>They literally don’t know HOW to use it.</strong> Not because they’re not smart enough or tech-savvy enough — but because nobody taught them that AI collaboration is a SKILL that requires:</p><ul><li>Systematic evaluation (not just “does this look okay?”)</li><li>Iterative refinement (not one-and-done prompts)</li><li>Measurement of improvement (not vague feelings of dissatisfaction)</li><li>Context and personalization (not generic inputs expecting magical outputs)</li></ul><p><strong>How do AI systems “know” any context about the human typing the question? They don’t.</strong> Unless you TELL them. Unless you build that context into your system instructions. Unless you personalize the AI to understand how YOU think, how YOU communicate, what YOU need.</p><p>It’s actually not the AI’s “fault” at all. <strong>It’s US. It’s HOW we talk to AI systems.</strong></p><p>That’s the skill I’ve been honing over these seven weeks. That’s what separates the 1% who build reliable AI systems from the 99% who try AI once, get frustrated, and give up.</p><p>The 1% understand: AI isn’t magic. It’s a tool that requires skill, practice, and systematic refinement to use effectively.</p><p>And once you develop that skill? The transformation is dramatic.</p><p><strong>Thank you Tyler. Thank you Sara.</strong></p><p><strong>The 1% who actually build reliable AI systems?</strong> They evaluate. Rigorously. Systematically. Continuously.</p><p>They don’t just ask “does this work?” They ask:</p><ul><li>How well does it work?</li><li>Under what conditions does it work best?</li><li>Where does it fail?</li><li>How can we measure improvement?</li><li>What’s our threshold for acceptable performance?</li><li>How do we maintain that performance at scale?</li></ul><p>This is the skill gap that’s holding back AI adoption in professional contexts. It’s not that AI isn’t capable — it’s that most people don’t know how to systematically improve and maintain AI performance.</p><p>And in high-stakes environments? That gap is dangerous.</p><h3>The Meta-Realization About My UX Work</h3><p>Here’s where Week 3 connected to everything I’ve been doing as a UX designer:</p><p><strong>Evaluation IS user testing.</strong> It’s the same fundamental skill — systematic observation, specific feedback, iterative refinement based on measured outcomes.</p><p>I’ve been doing this for years in UX design. Testing interfaces with users. Documenting specific issues. Refining designs. Testing again. Measuring improvement.</p><p><strong>The only difference with AI is that the “user” is the AI system itself, and the “interface” is the system instructions that guide its behavior.</strong></p><p>My Week 3 feedback even pointed this out: “As a UX/UI Systems Designer working with complex applications in government and military sectors, you have a unique perspective on evaluation and refinement. How might you adapt the evaluation framework you’ve developed to assess not just AI communication but also the usability and effectiveness of complex systems interfaces?”</p><p>That question hit me hard. Because the answer is: It’s the SAME framework.</p><p><strong>Systematic evaluation. Specific criteria. Measurable outcomes. Iterative refinement. Human feedback loops.</strong></p><p>Whether you’re evaluating an AI email assistant or a mission planning interface or an intelligence analysis workflow — the fundamental process is identical.</p><p>And suddenly I realized: My UX background isn’t just adjacent to AI design. It’s FOUNDATIONAL. The skills I’ve been developing for years — user research, systematic testing, iterative design, stakeholder management — these are exactly the skills needed for reliable AI development.</p><h3>The Questions I’m Now Asking</h3><p>Week 3 left me with questions that extend far beyond email assistants:</p><p><strong>How do we scale evaluation in enterprise AI systems?</strong> I can manually evaluate five email scenarios. But what about AI systems generating hundreds or thousands of outputs daily? How do we maintain evaluation rigor at scale?</p><p><strong>What’s the right balance between human evaluation and automated metrics?</strong> Some things require human judgment (does this sound like me?). Some things can be automated (word count, sentiment analysis). Where’s the line?</p><p><strong>How do we build evaluation into AI workflows from the beginning?</strong> Most people treat evaluation as an afterthought. What if we designed AI systems with evaluation frameworks built in from day one?</p><p><strong>What evaluation criteria matter most in different contexts?</strong> Email assistants need brand voice match. Medical AI needs accuracy and safety. Defense AI needs reliability under pressure. How do we identify the right criteria for each context?</p><p><strong>How do we train people to evaluate AI effectively?</strong> This is a SKILL. It requires practice, feedback, refinement. How do we help more people develop this capability?</p><h3>What Week 4 Was About to Reveal</h3><p>By the end of Week 3, I had all the pieces:</p><ul><li><strong>Week 1:</strong> Personalization (understanding how I learn and think)</li><li><strong>Week 2:</strong> Voice analysis (understanding how I communicate)</li><li><strong>Week 3:</strong> Evaluation (systematic improvement and reliability)</li></ul><p>Week 4 was about putting it all together: Building autonomous workflows where multiple AI agents collaborate to accomplish complex tasks.</p><p>And what I built surprised even me. Because I didn’t build an email assistant. I built something far more ambitious — a multi-agent system that transforms complex theological research into engaging educational content for my mother’s Bible teaching ministry.</p><p>And then I turned around and applied the same principles to mission planning for defense applications.</p><p>Turns out, the skills are universal. The contexts change. The stakes vary. But the fundamental principles of personalization + voice + evaluation = reliable AI systems that actually work.</p><p>But that’s a story for Part 4.</p><h3>The Transformation I’m Sitting With</h3><p>Here’s what Week 3 taught me that goes beyond AI:</p><p><strong>Systematic evaluation is how you bridge the gap between capability and reliability.</strong> In any domain. Any context. Any application.</p><p>AI can generate impressive outputs. But impressive isn’t the same as reliable. Reliable requires measurement, iteration, continuous improvement, human feedback loops.</p><p>And in high-stakes environments where I work — government, defense, intelligence — reliability isn’t optional. It’s everything.</p><p>The evaluation framework I learned in Week 3 isn’t just a technique for improving AI assistants. It’s a fundamental skill for building trustworthy AI systems in contexts where trust actually matters.</p><p><strong>My first round scores: Low-to-mid 20s out of 30.</strong> <strong>My second round scores: Consistently 28–29 out of 30.</strong></p><p>That’s not luck. That’s not magic. That’s systematic evaluation with human feedback loops.</p><p>That’s the 1% skill.</p><p>And now I have it.</p><p>Stay tuned for Part 4, where everything comes together in ways I didn’t expect.</p><p><em>In the spirit of transparency I advocate for in AI development: I worked with Claude to structure and refine these reflections from my Week 3 experience. The evaluation scores, feedback comments, and insights about systematic improvement are from my actual evaluation work in Sara and Tyler’s course, with AI assistance in articulating the patterns and principles I discovered.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c9c747cb6e26" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Deconstructing My Neurodivergent Voice: What AI Taught Me About How I Think]]></title>
            <link>https://medium.com/@katyneale/deconstructing-my-neurodivergent-voice-what-ai-taught-me-about-how-i-think-d612f85df7df?source=rss-dd94086ccac2------2</link>
            <guid isPermaLink="false">https://medium.com/p/d612f85df7df</guid>
            <category><![CDATA[neurodivergence]]></category>
            <category><![CDATA[aiux]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <dc:creator><![CDATA[Designed to Diverge]]></dc:creator>
            <pubDate>Mon, 02 Mar 2026 20:28:09 GMT</pubDate>
            <atom:updated>2026-03-02T20:28:09.861Z</atom:updated>
            <content:encoded><![CDATA[<p>Part 2: When Brand Voice Analysis Became Cognitive Archaeology</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*YjMJefpaMRczsm3b.png" /></figure><p><a href="https://substack.com/@designedtodiverge/note/p-181327127?r=edtib&amp;utm_source=notes-share-action&amp;utm_medium=web">(Part 1 of this series is here)</a></p><p><strong><em>Parenthetical layering: 40–50% of complex sentences. Extreme rhythmic variation: 3 to 87 words per sentence. Doubled intensifiers: “very, very” and “LOVED IT LOVED IT LOVED IT!”</em></strong></p><p>These aren’t communication quirks. They’re the linguistic fingerprints of a neurodivergent mind making invisible connections visible.</p><p><a href="https://maven.com/sara-davison/scale-with-aiworkflows-foundations">Week 2 of Sara and Tyler’s AI Agentic Fundamentals course </a>gave me an assignment that seemed straightforward: Analyze your brand voice and create an email assistant that sounds authentically like YOU.</p><p>Simple, right? Just document how you write, feed it to an AI, done.</p><p>Except… it wasn’t simple at all. It was one of the most uncomfortable, revealing, and ultimately transformative exercises I’ve ever done. Because what started as “analyze your writing style” became something much deeper: forensic archaeology of how my neurodivergent brain actually processes and communicates thought.</p><p>And what I discovered changed everything.</p><h3>The Assignment That Became Self-Discovery</h3><p>Here’s what Week 2 asked us to do: Take examples of your writing across different contexts (emails, blog posts, professional communications) and analyze the patterns. Then create system instructions for an AI email assistant that captures your authentic voice.</p><p>The goal? An AI that doesn’t just write emails FOR you, but writes emails that sound like they came FROM you. That carry your personality, your thinking style, your unique way of communicating.</p><p>So I gathered my writing samples. Casual emails to my brother-in-law about AI in banking. Blog posts about my artist’s journey (some from years ago, some recent). Professional communications about UX work. Enthusiastic USTA tennis team captain emails (WAHOO!!!). Vulnerable gratitude reflections. That post I wrote with Claude’s help about neurodivergent minds and AI.</p><p>And then I ran it through Tyler’s advanced AI linguistics expert specializing in comprehensive brand voice analysis and emulation guidance to help analyze. The output came out in minutes, not like “oh, I use exclamation points a lot” but a comprehensive FORENSIC analysis. Sentence length distribution. Vocabulary patterns. Punctuation habits. Tonal shifts across contexts. The actual STRUCTURE of how I communicate.</p><p>What I found… honestly, it was like seeing my brain mapped out in language patterns.</p><h3>The Neurodivergent Communication Framework</h3><p>Here’s what emerged from the analysis (and I’m going to get specific here because the specifics matter):</p><p><strong>Parenthetical Layering:</strong> 40–50% of my complex sentences include parenthetical asides. Not occasionally. FORTY TO FIFTY PERCENT. In casual writing, it jumps to 60%. In professional contexts, it drops to 20–30% (because I’m actively trying to “sound professional”).</p><p>Example from my email to my brother-in-law: “Man I’m actually so jealous cause this is everything that I’m learning and as of right this moment (with the government shutdown and stuff communication and forward movement has halted) so still haven’t been able to start to put into practice what I’ve been learning about.”</p><p>See that? The parenthetical aside isn’t decorative. It’s me showing you my THINKING PROCESS in real-time. It’s the context that’s running parallel in my brain while I’m making the main point. It’s how I actually think — multiple streams of thought happening simultaneously.</p><p><strong>Extreme Rhythmic Variation:</strong> My sentence lengths vary dramatically. Standard deviation of 9.8 to 24.8 words across different contexts. I’ll write a short burst for emphasis (5–10 words: “CHECK!”), then a medium flow for explanation (20–25 words), then a long exploration for complexity (40–60 words with multiple clauses and — you guessed it — parenthetical asides).</p><p>This isn’t random. It’s how my brain processes information. Short bursts when something clicks. Medium sentences when I’m explaining. Long, winding explorations when I’m making connections across multiple ideas simultaneously.</p><p><strong>Connection-Making Phrases:</strong> “I’m noticing parallels between…” “This connects to…” “The relationship between X and Y…” “I see patterns…” These phrases appear in 80%+ of my analytical writing.</p><p>Because that’s what my brain DOES. Constantly. Relentlessly. Sometimes exhaustingly. I’m always seeing how things connect, how patterns emerge, how seemingly disparate ideas relate to each other.</p><p><strong>Doubled/Tripled Intensifiers:</strong> “very, very frustrated,” “VERY, VERY different,” “so, so,” “LOVED IT LOVED IT LOVED IT!” This appears in 30% of my emphatic statements — 25% higher than standard English.</p><p>You can literally HEAR my voice when you read this. The repetition isn’t just emphasis — it’s oral quality translated to text. It’s how I would actually SAY it if we were talking.</p><p><strong>Question Cascades:</strong> I open with a genuine question, explore through additional questions, sometimes answer/sometimes leave open, inviting you into my thought process. Frequency: 3–5 questions per 500 words in reflective writing.</p><p>Because I’m not lecturing you. I’m thinking ALONGSIDE you. I’m inviting you into the mess of figuring things out, not presenting you with neat conclusions.</p><h3>The Contextual Adaptation Discovery</h3><p>But here’s where it gets really interesting (and vulnerable): My voice shifts DRAMATICALLY across different contexts, but it maintains core authenticity throughout.</p><p><strong>Maximum Celebratory Energy</strong> (tennis team emails): +0.82 sentiment score, 68% positive words, 20–25 ALL CAPS instances, 10–15 exclamation points per 100 words. “WE WON another match!!! WAHOO! We ended up winning 3/5 courts this morning — so FUN!”</p><p><strong>Measured Professional Warmth</strong> (work communications): +0.54 sentiment, 51% neutral words, 2–3 ALL CAPS instances, 1–2 exclamation points per 100 words. Still warm, still ME, but calibrated for professional context.</p><p><strong>Complex Emotional Vulnerability</strong> (reflective blog posts): +0.45 sentiment with 18% negative words. This is where I process the hard stuff — the frustration, the overwhelm, the “I am grateful for… BUT…” tension.</p><p>Reading through my old blog posts for this analysis was… intense. Posts from 2018 when I was drowning in new motherhood, sleep deprivation, feeling like I was failing at everything. Posts about my artist’s journey — that moment in France when my professor whispered “I would like to see MORE of that” and I was stunned because I thought I’d get reprimanded.</p><p>That vulnerability is part of my voice too. The willingness to say “I don’t know” while demonstrating sophisticated analysis. The combination of deep expertise and genuine struggle. The complex sentiment that holds both gratitude and frustration simultaneously.</p><h3>What I Discovered About How My Brain Communicates</h3><p>As I was documenting all of this — the parenthetical layering, the rhythmic variation, the connection-making, the contextual shifts — I started seeing something profound:</p><p>My “scattered” communication style isn’t scattered at all. It’s COGNITIVE TRANSPARENCY.</p><p>Every parenthetical aside? That’s me showing you the parallel processing happening in my brain.</p><p>Every question cascade? That’s me inviting you to think alongside me rather than being lectured to.</p><p>Every connection-making phrase? That’s me making invisible patterns visible so you can follow my thinking.</p><p>Every rhythmic variation? That’s my brain moving between different scales of understanding — the quick insight, the medium explanation, the complex exploration.</p><p>This isn’t poor communication. This is sophisticated communication that MODELS a way of thinking. It’s pedagogical. It’s inclusive. It demonstrates that showing your thought process helps others learn.</p><p>And suddenly I was making connections (because of course I was — that’s what I do):</p><p>Remember Week 1 when I documented my learning style for The Professor? Visual learning, macro-level vision first, pattern-seeking, connection-making across domains?</p><p>That cognitive style SHOWS UP in how I communicate. My writing style IS my thinking style made visible through language.</p><p>The parenthetical asides = parallel processing The question cascades = iterative thinking<br>The connection-making phrases = pattern recognition across domains The extreme rhythmic variation = operating at multiple scales simultaneously</p><p>My neurodivergent communication patterns aren’t things to “fix” or “manage” — they’re the linguistic expression of exactly the cognitive traits that make me effective at AI collaboration.</p><h3>The Format Problem (And What It Revealed)</h3><p>But here’s where Week 2 got uncomfortable in a different way.</p><p>I created my detailed brand voice analysis. I fed it to my AI email assistant. I tested it with different scenarios.</p><p>And the AI kept producing content in BULLETED FORMAT.</p><p>Clean. Organized. Professional-looking. Completely NOT how I actually write.</p><p>I write in flowing narratives. Long paragraphs that wind through ideas. Sentences that build on each other, creating momentum through rhythm and connection. Even when I’m being professional, I’m still writing in paragraph form, not bullet points.</p><p>The AI was capturing some elements of my voice — the enthusiasm, the connection-making — but missing the fundamental STRUCTURE of how I communicate.</p><p>And that’s when I realized something important: Different communication modes serve different purposes.</p><p>Bullets are great for clarity, for scannable information, for quick reference. But my natural narrative style serves a different purpose — it creates cognitive intimacy. It invites you into the thinking process. It builds understanding through accumulated context rather than isolated points.</p><p>As a UX designer, this hit me hard. How many times have I defaulted to bulleted lists in my design documentation because it “looks more professional”? How many times have I stripped out my natural communication style to fit conventional formats?</p><p>And here’s the bigger question: If I needed this level of self-awareness to recognize that my AI assistant was missing something fundamental about my voice… what does that mean for users who DON’T have that level of meta-awareness about their own communication?</p><h3>The Bridge Between Authentic Voice and Professional Expectations</h3><p>This connects directly to something my Week 2 feedback asked me to think about: “As a Senior UX Designer with a neurodivergent thinking style, how might you leverage your AI assistant to bridge the gap between your unique cognitive patterns and more traditional communication expectations in professional settings?”</p><p>That question sat with me for days.</p><p>Because here’s the tension: My neurodivergent communication style is AUTHENTIC. It’s how I actually think. It creates connection and understanding. It invites collaboration.</p><p>But professional contexts often expect different formats. Cleaner. More linear. Less… messy.</p><p>Could I create “contextual variations” for different stakeholders — clients, team members, executives — that maintain my authentic voice while adapting to their communication preferences?</p><p>And here’s the deeper question: SHOULD I?</p><p>Because part of me wonders: What if the “messiness” of my neurodivergent communication style is actually VALUABLE in professional contexts? What if showing my thinking process, making connections visible, inviting people to think alongside me — what if that’s exactly what complex problem-solving requires?</p><p>What if the issue isn’t that my communication style needs to be “cleaned up” — but that professional communication norms need to expand to accommodate cognitive diversity?</p><h3>The Meta-Awareness Challenge</h3><p>Here’s something that kept nagging at me throughout Week 2: I have a LOT of self-awareness about how I communicate. Years of therapy, journaling, reflection. Years of feeling “different” and trying to understand why. Years of being told (directly and indirectly) that I need to be more focused, more linear, more… normal.</p><p>That self-awareness let me do this forensic analysis of my own voice. But most people don’t have that level of meta-awareness about their communication patterns.</p><p>So how do we design AI systems that help people discover their authentic voice rather than just mimicking whatever they happen to write?</p><p>How do we build AI assistants that can say “Hey, I notice you write differently in different contexts — which version feels most authentically YOU?”</p><p>How do we create tools that AMPLIFY authentic communication rather than just standardizing it?</p><h3>What This Means for AI Design in Professional Settings</h3><p>I keep coming back to my work designing AI for government and defense applications. High-stakes environments where clear communication is literally life-or-death.</p><p>But what IS “clear communication” in those contexts?</p><p>Is it the clean, bulleted, standardized format? Or is it communication that shows reasoning, makes connections visible, invites critical thinking?</p><p>When an intelligence analyst is working with AI to make sense of complex data… do they need AI that produces neat summaries? Or do they need AI that shows its reasoning process, makes connections explicit, invites them to think critically about the outputs?</p><p>When a mission planner is using AI to coordinate complex operations… do they need streamlined efficiency? Or do they need transparency about how the AI is making recommendations?</p><p>Maybe “clear communication” in high-stakes contexts isn’t about simplification — it’s about COGNITIVE TRANSPARENCY. Showing the thinking. Making the connections visible. Inviting human judgment into the process.</p><p>Which is exactly what my neurodivergent communication style does naturally.</p><h3>The Uncomfortable Validation</h3><p>Here’s what I’m sitting with at the end of Week 2: For years, I’ve been trying to “fix” my communication style. Make it cleaner. More professional. Less scattered.</p><p>And this AI brand voice analysis just told me: Your “scattered” style isn’t scattered. It’s sophisticated cognitive transparency. It’s pattern recognition made visible. It’s parallel processing expressed through language. It’s EXACTLY the kind of communication that complex problem-solving requires.</p><p>That’s… validating. And uncomfortable. And empowering. And confusing.</p><p>Because if my neurodivergent communication style is actually a STRENGTH… why have I spent so much energy trying to suppress it?</p><p>And if AI can help me understand and articulate my authentic voice… what does that mean for all the other neurodivergent people who’ve been told their communication is “wrong”?</p><h3>The Questions I’m Now Asking</h3><p>Week 2 left me with even more questions than Week 1 (which is apparently how I know I’m learning something real):</p><p><strong>How do we design AI that discovers and amplifies authentic voice rather than standardizing it?</strong> The templates and frameworks are helpful, but they can also flatten uniqueness. Where’s the balance?</p><p><strong>Should professional communication norms adapt to accommodate cognitive diversity?</strong> Or should we keep creating “translation layers” where neurodivergent people code-switch to fit conventional expectations?</p><p><strong>What’s lost when we “clean up” neurodivergent communication for professional contexts?</strong> The cognitive transparency? The invitation to collaborative thinking? The pattern recognition made visible?</p><p><strong>How do we build AI assistants that help people develop meta-awareness about their own communication?</strong> Because that self-awareness is powerful, but most people don’t have access to it.</p><p><strong>If my communication style IS my thinking style made visible… what does that mean for designing AI interfaces?</strong> Should we be designing for different COGNITIVE STYLES, not just different user preferences?</p><h3>What Week 3 Was About to Teach Me</h3><p>Understanding my voice was one thing. But could I teach an AI to consistently produce content that sounded authentically like me?</p><p>That’s what Week 3 was about: EVALUATION. Systematic, rigorous evaluation with human feedback loops.</p><p>And what I discovered was that the gap between “sounds kind of like me” and “sounds EXACTLY like me” is enormous. The difference between generic AI outputs and truly personalized AI? It’s measurable. It’s dramatic. And it requires a skill that almost nobody talks about but that turned out to be the most valuable thing I learned in the entire course.</p><p>But that’s a story for Part 3.</p><p>For now, I’m sitting with this: My neurodivergent voice — with all its parenthetical layering, rhythmic variation, connection-making, and cognitive transparency — isn’t something to fix. It’s something to understand, amplify, and leverage.</p><p>Because if AI is going to enhance human communication rather than standardize it, we need to design systems that work WITH cognitive diversity, not against it.</p><p>And maybe — just maybe — the “messy” communication styles we’ve been trying to clean up are exactly what complex problem-solving in the AI era requires.</p><p>Stay tuned for Part 3, where I discover that evaluation is everything.</p><p><em>In the spirit of transparency I advocate for in AI development: I worked with Claude to structure and refine these reflections from my Week 2 experience. The insights about my communication patterns, the uncomfortable self-discovery, and the questions about professional communication norms are from my actual brand voice analysis work in Sara and Tyler’s course, with AI assistance in articulating patterns I’m still processing.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d612f85df7df" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I Thought I Was Learning AI Tools. Instead, I Discovered My Superpower.]]></title>
            <link>https://medium.com/@katyneale/i-thought-i-was-learning-ai-tools-instead-i-discovered-my-superpower-674fa172266d?source=rss-dd94086ccac2------2</link>
            <guid isPermaLink="false">https://medium.com/p/674fa172266d</guid>
            <category><![CDATA[neurodivergent]]></category>
            <category><![CDATA[ai-ux-design]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[ai-future]]></category>
            <category><![CDATA[ai-agent]]></category>
            <dc:creator><![CDATA[Designed to Diverge]]></dc:creator>
            <pubDate>Wed, 17 Dec 2025 21:52:16 GMT</pubDate>
            <atom:updated>2025-12-17T21:52:16.560Z</atom:updated>
            <content:encoded><![CDATA[<p>Part 1: When Personalization Became Pattern Recognition</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Jw5uchZ8Tqe2QoVNwB3BkQ.png" /></figure><p>Seven weeks ago, I made a decision that felt simultaneously practical and slightly terrifying. I enrolled in <a href="https://maven.com/sara-davison/scale-with-aiworkflows-foundations">Sara Davison and Tyler Fisk’s “AI Agentic Fundamentals” course on Maven.</a></p><p>The practical part? I needed this. I’m a Senior UX Designer at Vega Solutions working on high-stakes government and defense applications — mission planning systems, intelligence workflows, the kind of stuff where getting it wrong has real consequences. And I’ve been writing about AI ethics, human-AI collaboration, and designing for high-stakes environments for months now. But here’s the thing: I was still operating mostly in theory. I could talk about agentic AI, I could analyze what makes good AI design, I could see the patterns emerging across the industry… but I hadn’t actually BUILT anything myself.</p><p>The terrifying part? This was a 5-week intensive program began October 5th. Live sessions twice a week. Homework projects. A community of people who probably already knew way more than me (which they didn’t btw so if you are interested, PLEASE TAKE IT!). And I was jumping in during the absolute chaos of pre-holiday work crunch, family obligations, and that familiar ADHD overwhelm of having 46 browser tabs open (yes, I counted) while simultaneously trying to learn “46 AI apps I cannot live without” according to various YouTube influencers.</p><p>Sound familiar?</p><p>But something about this course felt different. Sara and Tyler weren’t just teaching tools — they were teaching a <strong>methodology</strong>. A way of thinking about AI that aligned with everything I’d been discovering about how my neurodivergent brain actually works. And there was this statistic in the course description that I couldn’t stop thinking about:</p><p><strong>Only 1% of executives believe their AI investments have reached maturity. 47% of C-suite executives identify talent skill gaps as the primary barrier to successful AI implementation.</strong></p><p>That gap. That’s where I wanted to be. Not just using AI tools, but understanding how to design and build AI systems that actually work for real people doing real work in complex environments.</p><p>So I signed up. And Week 1 completely changed how I think about AI collaboration.</p><h3>The Assignment That Broke Everything Open</h3><p>Week 1’s project seemed straightforward: Take “The Professor” — a generic AI learning assistant template — and personalize it for your own learning journey through the course.</p><p>Simple, right? Just customize a few settings, maybe add your name, done.</p><p>Except that’s not what happened at all.</p><p>What happened was this: I spent hours (HOURS) creating incredibly detailed system instructions that transformed The Professor from a generic assistant into something that understood ME. My visual learning preference. My need to see the macro-level vision first before diving into details. My pattern-seeking, connection-making, context-switching neurodivergent brain that everyone (including me, for most of my life) thought was scattered and unfocused.</p><p>I included my professional context — working at Vega Federal Solutions as a Senior UX Designer, the DOD/government contracting environment, the complex military applications I design for. I documented my learning style: how I process information through making connections across disparate domains, how I need to understand the “why” before the “how,” how my brain works best when it can see patterns across systems rather than isolated facts.</p><p>And here’s where it gets interesting (and a little vulnerable): I included details about my neurodivergent journey. How I was diagnosed with ADD as an adult. How my brain sees multiple problems simultaneously. How what looks like scattered focus is actually pattern recognition across complex systems. How the constant context-switching that exhausts other people is how I naturally operate.</p><p>I even included personality traits I wanted The Professor to embody: enthusiastic, witty, wise. Because if I’m going to spend 5 weeks learning from an AI assistant, it should feel like learning from someone I’d actually want to talk to.</p><p>Then I tested it. I asked both the generic Professor and my personalized Professor the exact same questions about course concepts.</p><p>The difference was… honestly, it was night and day.</p><h3>The “Barrel Master” Moment</h3><p>Here’s the moment that made me realize something profound was happening.</p><p>I asked my personalized Professor a question that included the term “Barrel Master” — a specific military role in airlift operations that I work with in my DOD applications. It’s not common terminology. You wouldn’t know it unless you’d worked in that specific domain.</p><p>The generic Professor? Confused. Gave me a generic response about literally making physical barrels or something equally irrelevant.</p><p>My personalized Professor? Understood EXACTLY what I was talking about. Responded with context-appropriate information about airlift operations, asset allocation, mission planning workflows. No explanation needed. It just… got it.</p><p>That’s when I realized: This isn’t just about making AI “nicer” or “more friendly.” This is about creating AI that understands the CONTEXT of how you think, what you do, and the specific domain knowledge you’re operating within.</p><p>And suddenly I was making connections everywhere (because that’s what my brain does).</p><p>If personalization works this powerfully for learning… what does that mean for the government and defense applications I design? What does it mean for intelligence analysts who need AI that understands their specific mission context? What does it mean for mission planners who operate in environments where generic responses could cost lives?</p><h3>The Pattern Recognition Revelation</h3><p>But here’s where it gets even more interesting. Because as I was documenting my learning style and cognitive patterns for The Professor, I started seeing parallels to something I’d written about months ago: <a href="https://substack.com/@designedtodiverge/note/p-174543281?r=edtib&amp;utm_source=notes-share-action&amp;utm_medium=web">whether neurodivergent minds have an AI advantage.</a></p><p>In that post, I’d hypothesized that certain cognitive traits seem particularly well-suited for AI collaboration:</p><ul><li>Comfort with ambiguity</li><li>Pattern recognition across domains</li><li>Iterative thinking</li><li>Active skepticism</li><li>Parallel processing</li></ul><p>And now, in Week 1 of this course, I was literally documenting those exact traits as the FOUNDATION for effective AI personalization.</p><p>My ADHD brain that sees patterns and connections everywhere? That’s not a bug I need to manage — it’s the exact cognitive architecture that makes me effective at AI collaboration.</p><p>My tendency to context-switch constantly between macro-level strategy and minute details? That’s not scattered focus — that’s the ability to operate at multiple scales simultaneously, which is exactly what complex AI system design requires.</p><p>My discomfort with surface-level solutions and tendency to question everything? That’s not being difficult — that’s the active skepticism needed to catch AI hallucinations and maintain appropriate trust calibration.</p><p>The overwhelm I feel when I see 46 browser tabs and think about 46 AI apps I “need” to learn? That’s not anxiety — that’s sophisticated pattern recognition detecting the gap between AI marketing hype and actual capability, identifying risks and potential problems before they become visible to others.</p><p>I’ve spent YEARS trying to “manage” my neurodivergent traits. Trying to be more focused. More linear. More… normal.</p><p>And here I was, in Week 1 of an AI course, discovering that those traits aren’t things to overcome — they’re exactly what effective AI collaboration requires.</p><h3>What This Means for High-Stakes AI Design</h3><p>This realization hit me hard. Because I don’t just use AI tools — I design AI systems for environments where lives are at stake.</p><p>I’ve been writing about this for months. About how we need AI systems that preserve human agency. About shared responsibility between AI and human judgment. About the importance of transparency, appropriate trust calibration, and designing for expert users rather than trying to make everything “simple.”</p><p>But personalization adds a whole new dimension to this thinking.</p><p>If I needed this level of customization to feel truly understood by an AI learning assistant… what does that mean for the military users, intelligence analysts, and mission planners I design for?</p><p>What if the reason so many AI systems fail in professional contexts isn’t because the AI isn’t “smart enough” — but because it’s not PERSONALIZED enough? Because it doesn’t understand the specific context, domain knowledge, and cognitive patterns of the people using it?</p><p>What if effective AI in high-stakes environments requires designing for cognitive diversity — not just accessibility in the traditional sense, but actual adaptation to different ways of thinking, processing information, and making decisions?</p><p>I started thinking about my failed CAMPS project — 4.5 years working on a massive Air Force application that never shipped. We tried to build everything for everyone all at once. We never understood the underlying patterns. We never personalized for the actual users and their specific workflows.</p><p>What if we’d approached it differently? What if instead of trying to create one generic system, we’d built personalized AI assistants that understood each user’s role, their specific mission context, their decision-making patterns?</p><h3>The Questions I’m Now Asking</h3><p>Week 1 left me with more questions than answers (which, honestly, is how I know I’m learning something real):</p><p><strong>How do we scale personalization in government/defense applications?</strong> You can’t manually customize every AI assistant for every user. But you also can’t have generic AI in environments where context is everything. What’s the middle ground?</p><p><strong>How do we design AI systems that work WITH different cognitive styles rather than assuming one interaction model fits all?</strong> My neurodivergent brain needs certain things from AI collaboration. Someone else’s brain might need completely different things. How do we build systems that adapt?</p><p><strong>What does “appropriate personalization” look like in high-stakes environments?</strong> There’s a balance between AI that understands you and AI that might be too accommodating, reinforcing biases or blind spots. Where’s that line?</p><p><strong>If personalization is this powerful for learning, what does it mean for operational AI?</strong> The Professor helps me learn about agentic AI. But what about AI that helps intelligence analysts make sense of complex data? AI that helps mission planners coordinate assets across multiple operations? How does personalization change effectiveness in those contexts?</p><h3>What I’m Realizing About My Work</h3><p>I’ve been designing UX for AI systems for years now. But I’ve been approaching it mostly from the interface level — how do we show AI reasoning, how do we build appropriate trust, how do we give users control.</p><p>What Week 1 taught me is that effective AI design might need to start much earlier. Not with the interface, but with understanding the USER at a much deeper level. Their cognitive patterns. Their domain expertise. Their specific context and needs.</p><p>This connects directly to the healthcare AI framework I wrote about — <a href="https://substack.com/@designedtodiverge/note/p-174544565?r=edtib&amp;utm_source=notes-share-action&amp;utm_medium=web">Vincent Buil’s work at Phillips on designing AI for high-stakes medical applications. </a>He emphasized that in high-risk environments, AI must enhance rather than replace human judgment. That shared responsibility must be designed, not just declared.</p><p>But what I’m adding now is: That shared responsibility might require personalization. Because “human judgment” isn’t generic — it’s specific to the person, their expertise, their cognitive style, their decision-making patterns.</p><h3>The Practical Applications Already Emerging</h3><p>By the end of Week 1, my brain was already spinning with applications (because of course it was — pattern recognition across domains, remember?):</p><p><strong>For my work:</strong> How could personalized AI assistants help bridge knowledge gaps for non-veterans like me working on military applications? Could we create AI that helps new team members quickly gain contextual understanding that normally takes years?</p><p><strong>For my son:</strong> Could I create a personalized learning tutor for my 7-year-old with dyslexia? An AI that understands his specific learning challenges and adapts explanations to how HIS brain processes information?</p><p><strong>For my bosses’ startup:</strong> They’re pursuing DOD contracts in a space where domain expertise is everything. Could personalized AI assistants help them scale their expertise without hiring dozens of subject matter experts?</p><p><strong>For the mission planning demo I’m building:</strong> Could personalization make the difference between AI that’s “interesting” and AI that’s actually USEFUL for the operators who would use it?</p><h3>The Bigger Picture</h3><p>Here’s what I’m sitting with as I move into Week 2: AI personalization isn’t just about making tools more user-friendly. It’s about fundamentally rethinking how we design human-AI collaboration.</p><p>The $50 billion AI talent gap that Sara and Tyler mentioned? I don’t think it’s just about people who can build AI systems. I think it’s about people who can design AI systems that work WITH human cognitive diversity rather than against it.</p><p>And my neurodivergent brain — the one that sees patterns across complex systems, that context-switches constantly, that questions everything, that gets overwhelmed by information overload but somehow processes it all simultaneously — might be exactly the cognitive architecture needed to bridge that gap.</p><p>I’ve written before about becoming “tech’s conscience” — about how UX designers are being asked to be the ethical guardrails for AI development. But what if it’s more than that? What if we’re also being asked to be the translators between AI capability and human cognitive diversity?</p><p>What if the future of AI isn’t about making it smarter — it’s about making it more adaptable to how different humans actually think?</p><h3>What’s Next</h3><p>Week 1 was about personalization. But Week 2? Week 2 was about to get uncomfortable in the best possible way.</p><p>Because if creating a personalized AI assistant required understanding how I learn… Week 2 was going to require something even more challenging: analyzing my own voice with forensic precision. Understanding not just how I think, but how I COMMUNICATE that thinking.</p><p>And what I discovered about my neurodivergent communication patterns — the parenthetical layering, the extreme rhythmic variation, the way I make invisible connections visible — changed how I think about AI-generated content entirely.</p><p>But that’s a story for Part 2.</p><p>For now, I’m sitting with this: Seven weeks ago, I thought I was learning to build AI assistants. What I actually started learning was how to design collaboration between human intelligence and artificial intelligence. And why my neurodivergent brain might be exactly what this moment requires.</p><p>The question isn’t whether AI belongs in high-stakes environments. The question is: Are we designing it with the personalization, rigor, and human-centered thinking those stakes demand?</p><p>Stay tuned.</p><p><em>In the spirit of transparency I advocate for in AI development: I worked with Claude to structure and refine these reflections from my course experience. The insights, breakthroughs, and honest assessments are from my actual Week 1 work in Sara and Tyler’s AI Agentic Fundamentals course, with AI assistance in articulating the patterns I’m still processing.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=674fa172266d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I Thought I Was Learning AI Tools. Instead, I Discovered My Superpower.]]></title>
            <link>https://medium.com/@katyneale/i-thought-i-was-learning-ai-tools-instead-i-discovered-my-superpower-cc6b70537ffb?source=rss-dd94086ccac2------2</link>
            <guid isPermaLink="false">https://medium.com/p/cc6b70537ffb</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[ai-tools-for-business]]></category>
            <category><![CDATA[ai-agent-development]]></category>
            <category><![CDATA[ai-future-of-work]]></category>
            <category><![CDATA[ai-ux-design]]></category>
            <dc:creator><![CDATA[Designed to Diverge]]></dc:creator>
            <pubDate>Wed, 17 Dec 2025 21:47:56 GMT</pubDate>
            <atom:updated>2025-12-17T21:47:56.015Z</atom:updated>
            <content:encoded><![CDATA[<h3>Part 1: When Personalization Became Pattern Recognition</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Jw5uchZ8Tqe2QoVNwB3BkQ.png" /></figure><p>Seven weeks ago, I made a decision that felt simultaneously practical and slightly terrifying. I enrolled in <a href="https://maven.com/sara-davison/scale-with-aiworkflows-foundations">Sara Davison and Tyler Fisk’s “AI Agentic Fundamentals” course on Maven.</a></p><p>The practical part? I needed this. I’m a Senior UX Designer at Vega Solutions working on high-stakes government and defense applications — mission planning systems, intelligence workflows, the kind of stuff where getting it wrong has real consequences. And I’ve been writing about AI ethics, human-AI collaboration, and designing for high-stakes environments for months now. But here’s the thing: I was still operating mostly in theory. I could talk about agentic AI, I could analyze what makes good AI design, I could see the patterns emerging across the industry… but I hadn’t actually BUILT anything myself.</p><p>The terrifying part? This was a 5-week intensive program began October 5th. Live sessions twice a week. Homework projects. A community of people who probably already knew way more than me (which they didn’t btw so if you are interested, PLEASE TAKE IT!). And I was jumping in during the absolute chaos of pre-holiday work crunch, family obligations, and that familiar ADHD overwhelm of having 46 browser tabs open (yes, I counted) while simultaneously trying to learn “46 AI apps I cannot live without” according to various YouTube influencers.</p><p>Sound familiar?</p><p>But something about this course felt different. Sara and Tyler weren’t just teaching tools — they were teaching a <strong>methodology</strong>. A way of thinking about AI that aligned with everything I’d been discovering about how my neurodivergent brain actually works. And there was this statistic in the course description that I couldn’t stop thinking about:</p><p><strong>Only 1% of executives believe their AI investments have reached maturity. 47% of C-suite executives identify talent skill gaps as the primary barrier to successful AI implementation.</strong></p><p>That gap. That’s where I wanted to be. Not just using AI tools, but understanding how to design and build AI systems that actually work for real people doing real work in complex environments.</p><p>So I signed up. And Week 1 completely changed how I think about AI collaboration.</p><h3>The Assignment That Broke Everything Open</h3><p>Week 1’s project seemed straightforward: Take “The Professor” — a generic AI learning assistant template — and personalize it for your own learning journey through the course.</p><p>Simple, right? Just customize a few settings, maybe add your name, done.</p><p>Except that’s not what happened at all.</p><p>What happened was this: I spent hours (HOURS) creating incredibly detailed system instructions that transformed The Professor from a generic assistant into something that understood ME. My visual learning preference. My need to see the macro-level vision first before diving into details. My pattern-seeking, connection-making, context-switching neurodivergent brain that everyone (including me, for most of my life) thought was scattered and unfocused.</p><p>I included my professional context — working at Vega Federal Solutions as a Senior UX Designer, the DOD/government contracting environment, the complex military applications I design for. I documented my learning style: how I process information through making connections across disparate domains, how I need to understand the “why” before the “how,” how my brain works best when it can see patterns across systems rather than isolated facts.</p><p>And here’s where it gets interesting (and a little vulnerable): I included details about my neurodivergent journey. How I was diagnosed with ADD as an adult. How my brain sees multiple problems simultaneously. How what looks like scattered focus is actually pattern recognition across complex systems. How the constant context-switching that exhausts other people is how I naturally operate.</p><p>I even included personality traits I wanted The Professor to embody: enthusiastic, witty, wise. Because if I’m going to spend 5 weeks learning from an AI assistant, it should feel like learning from someone I’d actually want to talk to.</p><p>Then I tested it. I asked both the generic Professor and my personalized Professor the exact same questions about course concepts.</p><p>The difference was… honestly, it was night and day.</p><h3>The “Barrel Master” Moment</h3><p>Here’s the moment that made me realize something profound was happening.</p><p>I asked my personalized Professor a question that included the term “Barrel Master” — a specific military role in airlift operations that I work with in my DOD applications. It’s not common terminology. You wouldn’t know it unless you’d worked in that specific domain.</p><p>The generic Professor? Confused. Gave me a generic response about literally making physical barrels or something equally irrelevant.</p><p>My personalized Professor? Understood EXACTLY what I was talking about. Responded with context-appropriate information about airlift operations, asset allocation, mission planning workflows. No explanation needed. It just… got it.</p><p>That’s when I realized: This isn’t just about making AI “nicer” or “more friendly.” This is about creating AI that understands the CONTEXT of how you think, what you do, and the specific domain knowledge you’re operating within.</p><p>And suddenly I was making connections everywhere (because that’s what my brain does).</p><p>If personalization works this powerfully for learning… what does that mean for the government and defense applications I design? What does it mean for intelligence analysts who need AI that understands their specific mission context? What does it mean for mission planners who operate in environments where generic responses could cost lives?</p><h3>The Pattern Recognition Revelation</h3><p>But here’s where it gets even more interesting. Because as I was documenting my learning style and cognitive patterns for The Professor, I started seeing parallels to something I’d written about months ago: <a href="https://substack.com/@designedtodiverge/note/p-174543281?r=edtib&amp;utm_source=notes-share-action&amp;utm_medium=web">whether neurodivergent minds have an AI advantage.</a></p><p>In that post, I’d hypothesized that certain cognitive traits seem particularly well-suited for AI collaboration:</p><ul><li>Comfort with ambiguity</li><li>Pattern recognition across domains</li><li>Iterative thinking</li><li>Active skepticism</li><li>Parallel processing</li></ul><p>And now, in Week 1 of this course, I was literally documenting those exact traits as the FOUNDATION for effective AI personalization.</p><p>My ADHD brain that sees patterns and connections everywhere? That’s not a bug I need to manage — it’s the exact cognitive architecture that makes me effective at AI collaboration.</p><p>My tendency to context-switch constantly between macro-level strategy and minute details? That’s not scattered focus — that’s the ability to operate at multiple scales simultaneously, which is exactly what complex AI system design requires.</p><p>My discomfort with surface-level solutions and tendency to question everything? That’s not being difficult — that’s the active skepticism needed to catch AI hallucinations and maintain appropriate trust calibration.</p><p>The overwhelm I feel when I see 46 browser tabs and think about 46 AI apps I “need” to learn? That’s not anxiety — that’s sophisticated pattern recognition detecting the gap between AI marketing hype and actual capability, identifying risks and potential problems before they become visible to others.</p><p>I’ve spent YEARS trying to “manage” my neurodivergent traits. Trying to be more focused. More linear. More… normal.</p><p>And here I was, in Week 1 of an AI course, discovering that those traits aren’t things to overcome — they’re exactly what effective AI collaboration requires.</p><h3>What This Means for High-Stakes AI Design</h3><p>This realization hit me hard. Because I don’t just use AI tools — I design AI systems for environments where lives are at stake.</p><p>I’ve been writing about this for months. About how we need AI systems that preserve human agency. About shared responsibility between AI and human judgment. About the importance of transparency, appropriate trust calibration, and designing for expert users rather than trying to make everything “simple.”</p><p>But personalization adds a whole new dimension to this thinking.</p><p>If I needed this level of customization to feel truly understood by an AI learning assistant… what does that mean for the military users, intelligence analysts, and mission planners I design for?</p><p>What if the reason so many AI systems fail in professional contexts isn’t because the AI isn’t “smart enough” — but because it’s not PERSONALIZED enough? Because it doesn’t understand the specific context, domain knowledge, and cognitive patterns of the people using it?</p><p>What if effective AI in high-stakes environments requires designing for cognitive diversity — not just accessibility in the traditional sense, but actual adaptation to different ways of thinking, processing information, and making decisions?</p><p>I started thinking about my failed CAMPS project — 4.5 years working on a massive Air Force application that never shipped. We tried to build everything for everyone all at once. We never understood the underlying patterns. We never personalized for the actual users and their specific workflows.</p><p>What if we’d approached it differently? What if instead of trying to create one generic system, we’d built personalized AI assistants that understood each user’s role, their specific mission context, their decision-making patterns?</p><h3>The Questions I’m Now Asking</h3><p>Week 1 left me with more questions than answers (which, honestly, is how I know I’m learning something real):</p><p><strong>How do we scale personalization in government/defense applications?</strong> You can’t manually customize every AI assistant for every user. But you also can’t have generic AI in environments where context is everything. What’s the middle ground?</p><p><strong>How do we design AI systems that work WITH different cognitive styles rather than assuming one interaction model fits all?</strong> My neurodivergent brain needs certain things from AI collaboration. Someone else’s brain might need completely different things. How do we build systems that adapt?</p><p><strong>What does “appropriate personalization” look like in high-stakes environments?</strong> There’s a balance between AI that understands you and AI that might be too accommodating, reinforcing biases or blind spots. Where’s that line?</p><p><strong>If personalization is this powerful for learning, what does it mean for operational AI?</strong> The Professor helps me learn about agentic AI. But what about AI that helps intelligence analysts make sense of complex data? AI that helps mission planners coordinate assets across multiple operations? How does personalization change effectiveness in those contexts?</p><h3>What I’m Realizing About My Work</h3><p>I’ve been designing UX for AI systems for years now. But I’ve been approaching it mostly from the interface level — how do we show AI reasoning, how do we build appropriate trust, how do we give users control.</p><p>What Week 1 taught me is that effective AI design might need to start much earlier. Not with the interface, but with understanding the USER at a much deeper level. Their cognitive patterns. Their domain expertise. Their specific context and needs.</p><p>This connects directly to the healthcare AI framework I wrote about — <a href="https://substack.com/@designedtodiverge/note/p-174544565?r=edtib&amp;utm_source=notes-share-action&amp;utm_medium=web">Vincent Buil’s work at Phillips on designing AI for high-stakes medical applications. </a>He emphasized that in high-risk environments, AI must enhance rather than replace human judgment. That shared responsibility must be designed, not just declared.</p><p>But what I’m adding now is: That shared responsibility might require personalization. Because “human judgment” isn’t generic — it’s specific to the person, their expertise, their cognitive style, their decision-making patterns.</p><h3>The Practical Applications Already Emerging</h3><p>By the end of Week 1, my brain was already spinning with applications (because of course it was — pattern recognition across domains, remember?):</p><p><strong>For my work:</strong> How could personalized AI assistants help bridge knowledge gaps for non-veterans like me working on military applications? Could we create AI that helps new team members quickly gain contextual understanding that normally takes years?</p><p><strong>For my son:</strong> Could I create a personalized learning tutor for my 7-year-old with dyslexia? An AI that understands his specific learning challenges and adapts explanations to how HIS brain processes information?</p><p><strong>For my bosses’ startup:</strong> They’re pursuing DOD contracts in a space where domain expertise is everything. Could personalized AI assistants help them scale their expertise without hiring dozens of subject matter experts?</p><p><strong>For the mission planning demo I’m building:</strong> Could personalization make the difference between AI that’s “interesting” and AI that’s actually USEFUL for the operators who would use it?</p><h3>The Bigger Picture</h3><p>Here’s what I’m sitting with as I move into Week 2: AI personalization isn’t just about making tools more user-friendly. It’s about fundamentally rethinking how we design human-AI collaboration.</p><p>The $50 billion AI talent gap that Sara and Tyler mentioned? I don’t think it’s just about people who can build AI systems. I think it’s about people who can design AI systems that work WITH human cognitive diversity rather than against it.</p><p>And my neurodivergent brain — the one that sees patterns across complex systems, that context-switches constantly, that questions everything, that gets overwhelmed by information overload but somehow processes it all simultaneously — might be exactly the cognitive architecture needed to bridge that gap.</p><p>I’ve written before about becoming “tech’s conscience” — about how UX designers are being asked to be the ethical guardrails for AI development. But what if it’s more than that? What if we’re also being asked to be the translators between AI capability and human cognitive diversity?</p><p>What if the future of AI isn’t about making it smarter — it’s about making it more adaptable to how different humans actually think?</p><h3>What’s Next</h3><p>Week 1 was about personalization. But Week 2? Week 2 was about to get uncomfortable in the best possible way.</p><p>Because if creating a personalized AI assistant required understanding how I learn… Week 2 was going to require something even more challenging: analyzing my own voice with forensic precision. Understanding not just how I think, but how I COMMUNICATE that thinking.</p><p>And what I discovered about my neurodivergent communication patterns — the parenthetical layering, the extreme rhythmic variation, the way I make invisible connections visible — changed how I think about AI-generated content entirely.</p><p>But that’s a story for Part 2.</p><p>For now, I’m sitting with this: Seven weeks ago, I thought I was learning to build AI assistants. What I actually started learning was how to design collaboration between human intelligence and artificial intelligence. And why my neurodivergent brain might be exactly what this moment requires.</p><p>The question isn’t whether AI belongs in high-stakes environments. The question is: Are we designing it with the personalization, rigor, and human-centered thinking those stakes demand?</p><p>Stay tuned.</p><p><em>In the spirit of transparency I advocate for in AI development: I worked with Claude to structure and refine these reflections from my course experience. The insights, breakthroughs, and honest assessments are from my actual Week 1 work in Sara and Tyler’s AI Agentic Fundamentals course, with AI assistance in articulating the patterns I’m still processing.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cc6b70537ffb" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Testing a New Approach: Adaptive Intelligence Design for High-Stakes AI Systems]]></title>
            <link>https://medium.com/@katyneale/testing-a-new-approach-adaptive-intelligence-design-for-high-stakes-ai-systems-b7e977fd02ff?source=rss-dd94086ccac2------2</link>
            <guid isPermaLink="false">https://medium.com/p/b7e977fd02ff</guid>
            <category><![CDATA[aiux]]></category>
            <category><![CDATA[ai-ux-design]]></category>
            <category><![CDATA[ai-future]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[aigovernment]]></category>
            <dc:creator><![CDATA[Designed to Diverge]]></dc:creator>
            <pubDate>Mon, 15 Dec 2025 20:46:50 GMT</pubDate>
            <atom:updated>2025-12-15T20:46:50.604Z</atom:updated>
            <content:encoded><![CDATA[<p>After months of research and synthesis, I’m about to put theory into practice. Here’s what I’m testing — and why traditional UX approaches aren’t enough for expert-level AI systems.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CI_YFcWe2_KFdEfd11hl8Q.png" /></figure><p>I’ve spent the past several months exploring a nagging question: Why do most AI-UX frameworks feel inadequate for the kind of systems I design? The guidance I was finding assumed consumer contexts — making AI “easy to use,” reducing friction, simplifying complexity.</p><p>But the professionals I design for don’t need simplification. They need amplification.</p><p>Intelligence analysts, mission planners, military strategists — these aren’t casual users looking for convenience. They’re experts managing extraordinary cognitive complexity, making decisions with serious consequences. When AI enters their workflows, the goal isn’t to make things simpler. It’s to make human expertise more powerful.</p><p>That realization led me to develop what I’m calling Adaptive Intelligence Design (AID) — a methodology specifically for AI systems in high-stakes, expert-level environments. And starting next week, I get to test whether it actually works.</p><h3>The Core Insight: Complexity Amplification vs. Simplification</h3><p>Traditional UX operates on a fundamental principle: reduce cognitive load, simplify interfaces, make things intuitive. This works brilliantly for consumer products where users want to accomplish tasks with minimal effort.</p><p><strong>But what if cognitive load isn’t the enemy?</strong></p><p>What if, for expert users in complex domains, the goal should be to <em>distribute</em> cognitive load intelligently between human and AI rather than simply reducing it? What if we should be designing systems that support the parallel processing, pattern recognition, and iterative thinking that experts naturally do, rather than forcing them into simplified linear workflows?</p><p>This isn’t about making systems harder to use. It’s about recognizing that experts process information differently than novices, and AI partnership should amplify that expertise rather than override it.</p><h3>What AID Methodology Actually Is</h3><p>At its core, AID is an approach to designing AI systems that:</p><p><strong>Assumes expert-level users</strong> who bring deep domain knowledge and sophisticated thinking patterns to their work</p><p><strong>Designs for cognitive complexity amplification</strong> rather than simplification — supporting parallel processing, pattern recognition across domains, and iterative refinement</p><p><strong>Builds transparency and oversight into interaction patterns</strong> rather than treating them as separate compliance requirements</p><p><strong>Preserves human expertise and judgment</strong> while leveraging AI capabilities for information synthesis and pattern detection</p><p><strong>Adapts to different cognitive styles</strong> rather than assuming one interaction model fits all expert users</p><p>The methodology draws from multiple sources: my observations about neurodivergent cognitive advantages in AI collaboration, Phillips Healthcare’s Pattern Library approach for high-risk AI systems, and years of experience designing for government environments where wrong decisions have serious consequences.</p><h3>The Framework: From Theory to Practice</h3><p>AID methodology breaks down into phases that map to how I actually approach complex system design:</p><h3>Phase 1: Cognitive Complexity Mapping</h3><p>Before designing any interfaces, understand how experts in this domain actually think. What information streams do they process simultaneously? What patterns do they recognize automatically versus what requires deliberate analysis? Where does AI have potential to amplify rather than replace human cognition?</p><p>This isn’t traditional user research focused on pain points and frustrations. It’s cognitive archaeology — uncovering how expert minds work so you can design AI partnership that enhances rather than disrupts those patterns.</p><h3>Phase 2: Parallel Processing Interface Design</h3><p>Design information architecture that supports how experts naturally scan and process information. Zone-based layouts where different screen areas serve different cognitive functions. AI confidence indicators that inform rather than dictate. Override controls that feel natural, not burdensome.</p><p>The key insight from healthcare: separate AI reasoning from AI results in the interface. Show both, but make it clear which is which so users can exercise appropriate judgment.</p><h3>Phase 3: Prototype and Validate</h3><p>Test with actual experts using real scenarios. Measure not just task completion, but decision quality, confidence calibration, and whether the system makes them feel more capable. Look for appropriate trust — neither over-reliance on AI nor rejection of AI assistance.</p><h3>Phase 4: Pattern Library Development</h3><p>Extract reusable design patterns that embed AI governance, transparency requirements, and human oversight capabilities. This is the practical implementation mechanism that makes abstract principles concrete across all applications.</p><h3>What I’m Testing in Next Couple Weeks</h3><p>Starting next week, I’m applying this methodology to a real government contract — designing an AI system for complex operational planning. I can’t share specifics due to NDAs, but I can describe what I’m testing:</p><p><strong>The Hypothesis:</strong> AI systems designed for cognitive complexity amplification will enable better decisions faster than those designed for simplified interaction — especially when users are domain experts managing high-stakes scenarios.</p><p><strong>What Success Looks Like:</strong></p><ul><li>Experts report feeling more capable, not just more efficient</li><li>Decision quality improves as measured by expert assessment</li><li>Appropriate trust calibration develops — users know when to rely on AI and when to override it</li><li>Skills are preserved rather than atrophied by AI assistance</li></ul><p><strong>What Would Make Me Revise This Approach:</strong></p><ul><li>If experts prefer simplified interfaces despite their expertise</li><li>If cognitive complexity amplification creates decision paralysis rather than capability enhancement</li><li>If the Pattern Library approach doesn’t translate from healthcare to defense contexts</li><li>If traditional HITL (Human in the Loop) approaches prove more effective than integrated AI partnership</li></ul><h3>The Real-World Constraints</h3><p>I’m not testing this in a lab with controlled conditions. I’m testing it in the messy reality of government contracting with:</p><ul><li>Tight timelines and budget constraints</li><li>Multiple stakeholders with different priorities</li><li>Legacy system integration requirements</li><li>Regulatory compliance that can’t be compromised</li><li>Users who are skeptical of new technology by default</li></ul><p>If AID methodology works under these conditions, it has practical value. If it only works in ideal circumstances, it needs significant refinement.</p><h3>Why This Matters Beyond My Work</h3><p>The question AID methodology addresses isn’t unique to defense applications. Any domain with expert users, serious consequences, and complex decision-making faces the same challenge: How do you design AI systems that amplify human expertise rather than replace it?</p><p>Healthcare is grappling with this. Financial services faces it. Critical infrastructure, emergency response, scientific research — anywhere expert judgment matters and stakes are high, the simplification-focused approach to AI-UX design breaks down.</p><p>If this methodology proves effective, it provides a framework others can adapt for their own high-stakes contexts. If it doesn’t work, the lessons learned about what failed will be valuable for anyone trying to solve similar problems.</p><h3>The Intellectual Honesty Requirement</h3><p>I need to be clear about what I’m claiming versus what I’m testing. I’m not saying I’ve solved high-stakes AI design. I’m saying I’ve synthesized insights from multiple sources into a testable approach, and I’m about to find out if it actually works in practice.</p><p>The methodology could fail for many reasons:</p><ul><li>My assumptions about expert cognition could be wrong</li><li>The healthcare-to-defense translation might not hold</li><li>Implementation challenges might overwhelm theoretical advantages</li><li>Users might reject approaches that don’t feel familiar</li></ul><p>But that’s the point of testing. Theory without validation is just speculation. Over the next few weeks, I’ll discover whether AID methodology is a meaningful contribution or needs fundamental revision.</p><h3>What I’ll Share Next</h3><p>Once I have results from this implementation — whether they validate or challenge my approach — I’ll share what actually happened. Not a sanitized success story, but an honest account of what worked, what didn’t, and what I learned about designing AI systems for expert-level, high-stakes environments.</p><p>The goal isn’t to prove I’m right. The goal is to contribute to our collective understanding of how to design AI systems that amplify rather than diminish human capability in contexts where it matters most.</p><p>Because if AI is going to be integrated into critical decision-making — and it already is — we need approaches that preserve and enhance human expertise rather than assuming simplification is always the answer.</p><p><strong>The question isn’t whether experts need AI assistance. The question is: How do we design that assistance to make experts more capable rather than more dependent?</strong></p><p><em>This methodology represents synthesis of insights from neurodivergent cognitive patterns, healthcare AI governance frameworks, and years of experience designing for high-stakes government environments. Starting next week, theory meets practice. The real test begins.</em></p><p><em>In the spirit of transparency about AI collaboration, I worked with Claude to develop and articulate this methodology — itself an example of the cognitive complexity amplification I’m describing. The framework and approach are my own, with AI assistance in refining the articulation and structure.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b7e977fd02ff" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What Healthcare Taught Me about High-Stakes AI Design]]></title>
            <link>https://medium.com/@katyneale/what-healthcare-taught-me-about-high-stakes-ai-design-f46e923044a1?source=rss-dd94086ccac2------2</link>
            <guid isPermaLink="false">https://medium.com/p/f46e923044a1</guid>
            <category><![CDATA[ai-defense]]></category>
            <category><![CDATA[ai-healthcare-revolution]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[aiux]]></category>
            <category><![CDATA[aigovernment]]></category>
            <dc:creator><![CDATA[Designed to Diverge]]></dc:creator>
            <pubDate>Mon, 06 Oct 2025 23:46:20 GMT</pubDate>
            <atom:updated>2025-10-06T23:46:20.145Z</atom:updated>
            <content:encoded><![CDATA[<p>A Phillips researcher’s presentation on AI in healthcare gave me the framework I’ve been searching for in defense applications. Turns out, when lives are at stake, the design principles converge.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BkN05zNPdUvo1hyBTVsnPQ.png" /></figure><p>I’ve spent months wrestling with a fundamental question: How do you design AI systems for high-stakes environments where wrong decisions have serious consequences? The frameworks I was finding felt inadequate — most AI-UX guidance assumes commercial contexts where the worst outcome is user frustration or a bad recommendation.</p><p>Then I attended an online presentation by Vincent Buil, a Senior UX Researcher at Phillips Healthcare, and something clicked. He was describing their approach to “Pattern Library for Responsible AI” in medical contexts — patient monitors, CT scans, ultrasound technology. As he walked through their framework, I kept thinking: <em>This is exactly what we need for defense applications.</em></p><p>The parallel makes sense when you think about it. Healthcare and defense share critical characteristics that most commercial AI applications don’t:</p><ul><li>Expert users with deep domain knowledge</li><li>Life-or-death consequences for wrong decisions</li><li>Strict regulatory compliance requirements</li><li>The need for AI to enhance rather than replace human judgment</li></ul><p><strong>Healthcare has been forced to solve the high-stakes AI design problem. And their solutions translate directly to government and defense contexts.</strong></p><h3>The Shared Responsibility Framework</h3><p>The first breakthrough insight from Buil’s presentation addressed the question that keeps everyone in high-stakes AI design awake at night: <strong>Who is responsible when AI gets it wrong?</strong></p><p>His answer was refreshingly honest: “In high-risk environments, it’s a shared responsibility.”</p><p>Phillips’ framework is clear:</p><ul><li><strong>We at Phillips should build and design AI systems with full transparency on their capabilities and limitations.</strong></li><li><strong>Clinical specialists’ responsibility is to use the AI system in context of those limitations as good as possible. Which we should enable them through the proper governance mechanisms in the UX.</strong></li></ul><p>This resonated deeply with my defense work. We can’t eliminate responsibility by adding AI — we have to design systems that make shared responsibility work in practice. The question isn’t whether AI or humans are “in charge,” it’s how we create interfaces that enable appropriate judgment from both.</p><h3>The Four Perspectives Framework: Purpose, Pathways, Policy, Pixels</h3><p>Phillips organizes their AI governance through four perspectives that I immediately recognized as applicable to defense systems:</p><h3>Purpose: Well-Defined Use Cases</h3><p>Phillips starts by identifying specific use cases that help clinicians get their work done better while supporting their wellbeing and benefiting patients. They run multi-stakeholder workshops using specific tools and templates to identify good use cases while educating teams on AI best practices.</p><p>The key insight: They ask both “What AI can/can’t do” AND “What AI should/shouldn’t do.”</p><p><strong>Defense translation:</strong> We need the same rigor in identifying where AI actually helps mission success versus where it just sounds innovative. The “should/shouldn’t” question is especially critical in defense — just because AI can make certain decisions doesn’t mean it should.</p><h3>Policy: Making AI Safe to Use</h3><p>Their policy framework ensures regulation compliance through governance documentation and tools. In healthcare, this means meeting EU AI requirements for high-risk systems.</p><p><strong>Defense translation:</strong> Government contractors already understand compliance requirements (FedRAMP, FISMA, etc.), but Phillips showed me how to weave AI-specific governance into existing frameworks rather than treating it as separate.</p><h3>Pathway: Carefully Designed Workflows</h3><p>Phillips designs the envisioned workflow and care pathway to ensure the end solution truly has the desired impact. They don’t just add AI features — they redesign entire workflows around human-AI collaboration.</p><p><strong>Defense translation:</strong> This is exactly what I’ve been trying to articulate in my AID methodology. You can’t just insert AI into existing military workflows — you have to rethink how the work gets done when AI and humans collaborate effectively.</p><h3>Pixels: Prototype-Based Validation</h3><p>They experiment with prototypes based on use cases to test both governance validation AND UX design principles at work.</p><p><strong>Defense translation:</strong> The “test governance through UX” approach is brilliant. Instead of separate compliance checks, design prototypes that demonstrate governance principles in actual use.</p><h3>The Two Critical Requirements: Human Oversight and Transparency</h3><p>Buil emphasized that two elements are most critical for high-risk AI systems: <strong>Human Oversight</strong> and <strong>Transparency</strong>.</p><p>The EU AI Requirements he outlined map directly to what defense systems need:</p><h3>Transparency Requirements:</h3><ul><li>Clear labeling of AI functions and AI results</li><li>Clear instructions for use: intended purpose, accuracy, limitations</li><li>Guidance on avoiding risky usage and misuse scenarios</li><li>Information about capabilities, limitations, and target populations</li><li>Support for interpretation and use of AI outputs</li></ul><h3>Human Oversight Requirements:</h3><p>Tools to prevent or reduce risks such that users can:</p><ul><li>Detect AI errors</li><li>Avoid over-relying on AI</li><li>Understand and interpret AI results</li><li>Disregard, override, or reverse AI results</li><li>Stop the AI system in a safe state</li></ul><p><strong>This framework solves a problem I’ve been struggling with:</strong> How do you build appropriate skepticism into interfaces without creating analysis paralysis? Phillips answered it: give users specific tools for oversight rather than just warnings about reliability.</p><h3>The Pattern Library Approach</h3><p>Here’s where Phillips’ work becomes directly actionable. They created a Pattern Library — reusable design patterns that:</p><ul><li>Structure usable components for designers across all legacy systems</li><li>Create recognizable patterns that identify what is AI versus human-generated</li><li>Apply everywhere in all applications</li><li>Ensure compliance with AI requirements</li></ul><p>Their lessons learned:</p><ol><li><strong>Designers want practical help</strong> in UX design for AI (not theoretical frameworks)</li><li><strong>You’re designing for uncertainty</strong> — make sure stakeholders understand this is probabilistic, not deterministic</li><li><strong>Create prototypes in live environments</strong> to show probabilistic results, not sanitized demos</li><li><strong>AI governance touches all levels:</strong> purpose, pathways, pixels, and policy</li></ol><p><strong>The healthcare insight I’m applying to defense:</strong> Don’t mix AI “reasoning” with AI “results” in the interface. Show both, but make it clear which is which.</p><h3>The Questions Healthcare Is Still Wrestling With</h3><p>What I found most valuable was Buil’s honesty about unresolved challenges. Phillips is still asking:</p><p><em>“We are all using AI to generate something and now have to review it. This is a high cognitive task! Is that taking more time?”</em></p><p><em>“Plus if AI takes away other tasks, potential loss of skills that is needed (for diagnosis) is that worth it?”</em></p><p><em>“How does this affect job satisfaction? Job profile?”</em></p><p><em>“We are thinking we are helping to free up clinician’s time but then they get more patients, so maybe it’s not efficient?”</em></p><p><strong>These questions are exactly what I’m grappling with in defense contexts.</strong> When AI handles routine analysis, do intelligence analysts lose the skills needed for critical assessment? When AI speeds up mission planning, does that just mean more missions rather than better decisions?</p><p>Healthcare doesn’t have all the answers yet, but they’re asking the right questions.</p><h3>The Holistic Experience Imperative</h3><p>Buil ended with something that struck me as the core principle for all high-stakes AI design:</p><p><em>“In the end, we must design a holistic experience with all the stakeholders in mind:</em></p><ul><li><em>Safety for the patient</em></li><li><em>Job satisfaction for the doctor</em></li><li><em>Efficiency in healthcare</em></li><li><em>Society that we want to live in without losing human touch (which is especially key in healthcare)”</em></li></ul><p><strong>Defense translation:</strong></p><ul><li>Safety for warfighters and civilians</li><li>Job satisfaction and skill preservation for military professionals</li><li>Efficiency in mission execution</li><li>Society we want to defend — including preserving human judgment in critical decisions</li></ul><h3>What This Means for My Work</h3><p>Phillips’ healthcare framework validated something I’ve been developing in my AID methodology: high-stakes AI design requires fundamentally different approaches than consumer applications.</p><p>The key insights I’m taking forward:</p><p><strong>Shared Responsibility Must Be Designed:</strong> You can’t just declare it — you have to build interfaces that make it work in practice.</p><p><strong>Governance Through UX:</strong> Compliance isn’t separate from design; it’s embedded in how users interact with AI.</p><p><strong>Transparency Enables Oversight:</strong> When users understand AI reasoning and limitations, they can exercise appropriate judgment.</p><p><strong>Uncertainty Is The Design Challenge:</strong> Stop pretending AI is deterministic; design for probabilistic thinking.</p><p><strong>Skills Preservation Matters:</strong> If AI takes away practice opportunities, expertise atrophies. Design for learning, not just efficiency.</p><h3>The Validation I Needed</h3><p>Finding Phillips’ work was validation that I’m not alone in thinking about these challenges. Healthcare has been forced to solve high-stakes AI design because patient safety demands it. Defense faces the same imperative for different but equally critical reasons.</p><p>But here’s the critical insight that hit me during Buil’s presentation: <strong>The Pattern Library isn’t just a design system — it’s the practical implementation mechanism for everything else.</strong></p><p>I’ve always associated design libraries with branding and marketing — ensuring consistent visual identity across products. That’s a very commercial, product-centric viewpoint. But in enterprise and highly complex work-related applications, the Pattern Library serves a fundamentally different purpose: <strong>it’s the structure that enforces industry regulations, embeds AI governance, and provides practical guidance all at once.</strong></p><p>This is the “how” that I’ve been missing. You can have all the right principles about transparency, human oversight, and shared responsibility, but without a structured pattern library to implement them consistently across systems, they remain theoretical. The Pattern Library makes abstract governance concrete and actionable for designers.</p><p>I’m not copying Phillips’ specific framework — I’m pulling out the principles I need and reframing them for defense applications. Their Four Perspectives give me structure, but the Pattern Library concept is what makes it practically applicable. It’s how you take “we need transparency” and turn it into “here’s the exact design pattern designers use to show AI reasoning across all our applications.”</p><p>This might seem obvious to designers who specialize in pattern libraries and design systems, but it’s a revelation for someone like me who’s been focused on user workflows and interaction design. The Pattern Library isn’t peripheral to AI governance — it’s central. It’s the mechanism that ensures every interface, every interaction, every AI feature follows the same principles for transparency, oversight, and shared responsibility.</p><p>That’s what I’m taking into my defense work: not a specific set of healthcare patterns, but the understanding that systematic design patterns are how you operationalize responsible AI principles in complex, regulated environments.</p><h3>What’s Next</h3><p>I’m now adapting Phillips’ framework for defense applications, testing whether their healthcare patterns translate to military contexts. The hypothesis: if both domains share the high-stakes characteristics (expert users, serious consequences, regulatory requirements, human judgment preservation), then design patterns should be transferable with domain-specific adaptations.</p><p>Over the next few weeks, I’ll be prototyping AI systems using these principles. The goal isn’t just to validate that this approach works — it’s to develop a practical methodology that other designers can use when building AI for high-stakes professional environments.</p><p>Because if healthcare has shown us anything, it’s that you can design responsible AI for critical applications. You just have to acknowledge the stakes, build for shared responsibility, and design interfaces that preserve rather than replace human expertise.</p><p><strong>The question isn’t whether AI belongs in high-stakes environments. The question is: Are we designing it with the rigor and responsibility those stakes demand?</strong></p><p><em>This exploration builds on Vincent Buil’s presentation “Pattern Library for Responsible AI” at the STRAT 2025 conference. His work at Phillips Healthcare provides crucial insights for anyone designing AI systems where consequences matter. The parallels between healthcare and defense contexts suggest that high-stakes AI design principles may be more universal than domain-specific.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f46e923044a1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Do Neurodivergent Minds Have an AI Advantage?]]></title>
            <link>https://medium.com/@katyneale/do-neurodivergent-minds-have-an-ai-advantage-860af45f5cd8?source=rss-dd94086ccac2------2</link>
            <guid isPermaLink="false">https://medium.com/p/860af45f5cd8</guid>
            <category><![CDATA[ai-anxiety]]></category>
            <category><![CDATA[aiux]]></category>
            <category><![CDATA[neurodivergent]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[ai-future]]></category>
            <dc:creator><![CDATA[Designed to Diverge]]></dc:creator>
            <pubDate>Wed, 01 Oct 2025 18:58:26 GMT</pubDate>
            <atom:updated>2025-10-01T18:58:26.569Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4qbfAUoyIAyR6I9c379-hw.png" /></figure><p><em>After nine months of intensive AI collaboration, I’ve started wondering: Are certain cognitive traits better suited for effective human-AI partnership? And if so, what does that mean for how we design AI systems for different types of thinkers?</em></p><p>I was driving home yesterday, listening to Vivaldi’s Four Seasons, when something clicked that I’ve been trying to articulate for months.</p><p>For as long as I can remember, I’ve been drawn to classical music when I need to focus — specifically Bach, Vivaldi, Mozart, and Beethoven. Not because it’s “background music,” but because there’s something about the patterns, the mathematical precision mixed with organic flow, the tension and resolution that helps my brain find order in chaos.</p><p>It wasn’t until I learned about my neurodivergence that I understood why. My brain is constantly seeking connections and patterns, sometimes to an exhausting degree. What I once thought was laziness or lack of focus was actually a different way of processing information — one that thrives on finding relationships between seemingly disparate elements.</p><p><strong>And that’s exactly what effective AI collaboration seems to require.</strong></p><h3>The Pattern Recognition Connection</h3><p>The more I work with AI tools like Claude, the more I notice parallels between how my neurodivergent brain works and what makes AI partnership effective. Both involve:</p><ul><li><strong>Comfort with ambiguity</strong> — Being okay with not having complete information while still making progress</li><li><strong>Pattern recognition across domains</strong> — Seeing connections between different fields and contexts</li><li><strong>Iterative thinking</strong> — Building understanding through multiple cycles rather than linear progression</li><li><strong>Active skepticism</strong> — Questioning outputs and looking for potential flaws or biases</li><li><strong>Parallel processing</strong> — Managing multiple streams of information simultaneously</li></ul><p>These aren’t skills I learned in UX bootcamps. They’re cognitive patterns that emerged from navigating the world with a brain that works differently.</p><h3>The AI Collaboration Hypothesis</h3><p>Here’s what I’m wondering: Could certain neurodivergent traits actually be advantages when it comes to AI collaboration?</p><p>I’m not suggesting that only neurodivergent people can work effectively with AI, or that all neurodivergent people have these advantages. But I am observing that some of the cognitive approaches that have made life challenging in traditional work environments seem particularly well-suited for AI partnership.</p><p><strong>The “always on” pattern-seeking brain that can’t turn off?</strong> That might be exactly what’s needed to effectively collaborate with systems that process information through pattern recognition.</p><p><strong>The tendency to question assumptions and see multiple perspectives?</strong> That could be crucial for maintaining appropriate skepticism about AI outputs.</p><p><strong>The comfort with iterative, non-linear thinking?</strong> That aligns well with how AI collaboration actually works — through cycles of prompting, evaluation, and refinement.</p><h3>What This Means for AI System Design</h3><p>If this hypothesis has merit, it raises important questions about how we design AI interfaces and interactions:</p><h3>Are we designing for the right cognitive styles?</h3><p>Most AI tools are designed to feel simple and intuitive for the broadest possible user base. But what if effective AI collaboration actually requires more sophisticated cognitive engagement? What if making it “simple” is making it less effective?</p><h3>Should different cognitive styles get different interfaces?</h3><p>We already know that people learn and process information differently. Should AI systems adapt to different cognitive approaches rather than assuming one interaction model fits all users?</p><h3>How do we support cognitive diversity in AI adoption?</h3><p>If certain thinking patterns are better suited for AI collaboration, how do we help people develop those skills without excluding those who think differently?</p><h3>The High-Stakes Context</h3><p>This isn’t just an academic question for me. In my work designing systems for government and defense applications, the stakes of getting AI collaboration wrong are enormous. These are environments where:</p><ul><li><strong>Decisions have life-or-death consequences</strong></li><li><strong>Users are domain experts with decades of experience</strong></li><li><strong>Systems must work under extreme pressure</strong></li><li><strong>Trust calibration is critical — both under-reliance and over-reliance on AI can be dangerous</strong></li></ul><p>In these contexts, I’m starting to think we need AI systems that assume users will develop expertise in human-AI collaboration, rather than trying to make it so simple that no expertise is required.</p><p><strong>Maybe the future of professional AI isn’t about making it accessible to everyone, but about designing it to amplify the cognitive strengths that effective AI partnership requires.</strong></p><h3>Beyond the Individual Level</h3><p>This question extends beyond individual cognitive differences to broader design philosophy. Are we moving toward a world where certain types of thinking become more valuable because they align well with how AI systems work?</p><p>And if so, what’s our responsibility as designers to:</p><ul><li>Help people develop AI collaboration skills</li><li>Design systems that work with different cognitive approaches</li><li>Avoid creating new forms of exclusion based on thinking style</li><li>Preserve human agency and critical thinking in AI-augmented workflows</li></ul><h3>The Uncomfortable Implications</h3><p>I’ll be honest: this line of thinking makes me uncomfortable sometimes. The last thing I want is to suggest that some people are inherently “better” at working with AI than others, or that neurodivergent traits are superior to neurotypical ones.</p><p>What I am suggesting is that different cognitive styles might have different strengths when it comes to AI collaboration. And understanding those strengths could help us design better systems for everyone.</p><p>The neurodivergent experience of having to develop workarounds, question assumptions, and think in non-standard ways might have accidentally prepared some of us for effective AI partnership. But those skills aren’t exclusive to neurodivergent minds — they can be developed and supported through thoughtful design.</p><h3>What I’m Actually Testing</h3><p>Over the next few months, I’ll be putting these ideas to the test in real-world applications. But I want to be clear about what I’m claiming versus what I’m investigating, because the distinction matters.</p><p><strong>What I’ve Observed (Not Yet Proven):</strong> Certain cognitive traits — comfort with ambiguity, pattern recognition across domains, iterative thinking, active skepticism — seem useful for AI collaboration. I’ve also noticed these traits are common in neurodivergent thinking patterns, including my own.</p><p><strong>What I Haven’t Established:</strong></p><ul><li>Whether neurodivergent minds <em>caused</em> these traits or if these traits develop through other pathways</li><li>Whether these traits actually <em>cause</em> better AI outcomes or just correlate with them</li><li>Whether neurotypical individuals with similar cognitive approaches perform equally well</li><li>What “better” even means in AI collaboration contexts</li></ul><p><strong>What I’m Looking For in My Testing:</strong></p><h3>Observable Behaviors That Would Support This Hypothesis:</h3><ul><li>People who naturally question AI outputs perform more effectively than those who accept them uncritically</li><li>Users comfortable with iterative refinement achieve better results than those seeking single-shot solutions</li><li>Pattern recognition across domains translates to identifying AI limitations faster</li><li>Parallel processing capabilities correlate with effective use of AI in complex workflows</li></ul><h3>What Would Make Me Revise This Hypothesis:</h3><ul><li>If cognitive traits don’t predict AI collaboration effectiveness</li><li>If neurotypical users with different cognitive styles achieve equal or better outcomes</li><li>If training and experience override any initial cognitive advantages</li><li>If the traits I’m associating with neurodivergence are actually just professional expertise</li></ul><h3>The Measurements That Matter:</h3><p>In my upcoming work on high-stakes AI systems, I’ll be tracking:</p><ul><li><strong>Decision quality:</strong> Do certain cognitive approaches lead to better outcomes?</li><li><strong>Appropriate trust calibration:</strong> Who develops realistic confidence in AI capabilities?</li><li><strong>Error detection:</strong> Who catches AI mistakes or limitations faster?</li><li><strong>Skill preservation:</strong> Who maintains their expertise while leveraging AI?</li></ul><p><strong>Most importantly, I’m looking for whether AI systems designed to support these cognitive traits (regardless of neurological origin) perform better than those designed for simplified interaction.</strong></p><h3>The Real Question</h3><p>Here’s what I’m actually investigating: If effective AI collaboration requires certain cognitive approaches, how do we design systems that either support those approaches naturally or help users develop them?</p><p>The neurodivergent angle matters not because neurodivergent minds are “better” at AI, but because understanding which cognitive patterns work well with AI helps us design better systems for everyone. If pattern recognition and iterative thinking are crucial for AI partnership, then we need interfaces that support those cognitive processes — whether users come by them naturally or need to develop them.</p><p>This isn’t about identifying who’s suited for AI work and who isn’t. It’s about understanding what effective AI collaboration actually requires so we can design systems and training that help more people develop those capabilities.</p><p>Because if AI is going to amplify human intelligence rather than replace it, we need to understand what kinds of human intelligence are worth amplifying — and how to help people access those cognitive strengths regardless of their neurological makeup.</p><h3>The Broader Question</h3><p>Ultimately, this isn’t just about neurodivergent advantages. It’s about understanding what effective human-AI collaboration actually requires and designing systems that support those cognitive processes.</p><p>As AI becomes more sophisticated and integrated into critical decision-making, we need to move beyond the assumption that simpler is always better. Sometimes the most effective collaboration requires cognitive sophistication from both human and artificial intelligence.</p><p>The question isn’t whether some people are naturally better at AI collaboration. The question is: What cognitive approaches make AI partnership most effective, and how do we help people develop those capabilities while preserving human agency and critical thinking?</p><p><strong>Because if AI is going to amplify human intelligence rather than replace it, we need to understand what kinds of human intelligence are worth amplifying.</strong></p><p><em>This exploration grew out of my ongoing collaboration with Claude as a thinking partner — itself an example of the iterative, pattern-seeking cognitive approach I’m describing. The meta-irony isn’t lost on me that I’m using AI to think through questions about AI collaboration, but that partnership has helped me articulate insights I couldn’t have reached alone.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=860af45f5cd8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What My Friends Really Know About AI (And Why It Keeps Me Up at Night as a UX Designer)]]></title>
            <link>https://medium.com/@katyneale/what-my-friends-really-know-about-ai-and-why-it-keeps-me-up-at-night-as-a-ux-designer-4a67256a4941?source=rss-dd94086ccac2------2</link>
            <guid isPermaLink="false">https://medium.com/p/4a67256a4941</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[ai-future]]></category>
            <category><![CDATA[ai-anxiety]]></category>
            <category><![CDATA[aiux]]></category>
            <dc:creator><![CDATA[Designed to Diverge]]></dc:creator>
            <pubDate>Sat, 27 Sep 2025 15:42:19 GMT</pubDate>
            <atom:updated>2025-09-27T15:42:19.500Z</atom:updated>
            <content:encoded><![CDATA[<h3>After nine months of diving deep into AI, I wondered: Are the people in my life as unprepared as I was?</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RbKHMYxEcXYybYivIFaUjQ.png" /></figure><p><em>What I learned has me thinking differently about the users I design for in high-stakes environments.</em></p><p>Eight months ago, I thought AI was just another tech marketing fad — another “shiny object” promising to make us all more productive. Fast forward to today, and I’m knee-deep in prompt engineering, following AI safety discussions, and genuinely concerned about designing AI systems for users who might know even less than I did when I started.</p><p>Thanks for reading kathryn.neale@gmail.com! Subscribe for free to receive new posts and support my work.</p><p>This transformation got me wondering about something that’s been nagging at me in my work: How many people in my life are experiencing this same whiplash? More importantly, what do my friends and family — people who represent the same demographic as many government and defense workers I design for — actually know about the AI tools that are rapidly reshaping our world?</p><p>Those of us who remember landline phones, who lived through the dot-com revolution in our early twenties and watched it transform the world in a decade — we’re seeing the same magnitude of change with AI, except this time it feels different somehow. More consequential.</p><p>As someone who designs systems for high-stakes environments, I started wondering: if the people in my life are struggling with AI adoption, what does that suggest about the users I design for professionally? The stakes feel higher when you’re creating systems where wrong decisions could affect critical outcomes.</p><p>So I decided to ask 15 friends and family members what they actually know about AI. What I discovered has me thinking differently about assumptions I might be making about user knowledge.</p><h3>The Survey: 15 Friends and Family Members</h3><p>I asked 15 people in my life — friends and family ranging from their late 30s to 50s, most with established careers and families — about their AI knowledge and experiences. These are people who share my generational experience: we adapted to smartphones, lived through major technological shifts, and now find ourselves trying to figure out this next wave.</p><p>They represent the broader demographic that includes many of the government and professional workers I design systems for. Smart, capable people who’ve successfully navigated technological change before, which makes their responses particularly interesting to consider.</p><h3>The “Familiar Yet Clueless” Paradox</h3><p>Here’s what stopped me in my tracks: 53% consider themselves “somewhat familiar” with AI. On the surface, that sounds encouraging — most people feel they have at least some grasp on what AI is and does.</p><p>But when I dug deeper into what they actually understand, a troubling picture emerged. 40% have never heard of “AI hallucinations” — when AI confidently generates false information. A third don’t know what “prompt engineering” means. Over a quarter have no idea how important datasets are to AI tool reliability.</p><p>This disconnect between perceived familiarity and actual knowledge is exactly what I was afraid of finding. In my work designing systems for high-stakes environments, overconfidence paired with knowledge gaps isn’t just inconvenient — it’s potentially dangerous.</p><h3>The Mental Model Chaos</h3><p>When I asked what comes to mind when they think about AI, the responses revealed just how wildly different people’s frameworks are:</p><p>Some saw it pragmatically: <em>“A tool that can be used to streamline work or help analyze data.”</em> One person called it their <em>“personal intern.”</em></p><p>Others expressed concern with varying levels of alarm: <em>“Some really cool things that can increase productivity and effectiveness with some potentially scary pitfalls.”</em> Another worried about it <em>“taking over the economy.”</em></p><p>And then there were the deeper reflections: <em>“Potential for positive change through automation, but also the potential for lazy reliance on technology. Currently we are forced into at least some semblance of critical thinking…”</em></p><p>What strikes me is that these aren’t just different levels of knowledge — they’re fundamentally different ways of understanding what AI is. In my work, if users are approaching the same system with mental models this divergent, how can we possibly design interfaces that serve them all effectively?</p><h3>Regular Users Don’t Know What They Don’t Know</h3><p>Here’s what really concerns me: 40% of my survey group uses ChatGPT regularly. They’re actively working with AI tools, making decisions based on AI outputs, integrating AI into their workflows.</p><p>But when I asked where AI gets its information, responses included things like <em>“Unsure,”</em> <em>“Web scraping the internet,”</em> and <em>“The Internet, social media, research papers.”</em></p><p>While not entirely wrong, these answers reveal a shallow understanding of how training data works, why AI has limitations, and when outputs should be questioned. In commercial applications, this might lead to minor inconveniences. In the defense applications I work on? The implications are much more serious.</p><h3>The “Just Tell Me What’s Safe” Problem</h3><p>When I asked what would motivate them to invest time learning AI tools, their responses revealed something I need to take seriously in my design work:</p><p><em>“Knowing where to start.”</em></p><p><em>“Confidence that the output is accurate and there is a clear leader in which tool is best.”</em></p><p><em>“User friendly (not time consuming to learn the functionality).”</em></p><p><em>“I think I’d need help in figuring out which tools are the most helpful and least invasive/troublesome.”</em></p><p>These aren’t people asking for advanced training or technical deep-dives. They’re asking for curated, trustworthy guidance about what’s safe to use and how to use it effectively. The question for me as a designer: How do I build that guidance into interfaces rather than expecting users to seek it out separately?</p><h3>The Weight My Generation Carries</h3><p>One response has stuck with me since I read it:</p><p><em>“As a parent, AI feels overwhelming and like a slippery slope. I do not want to become dependent on these tools, and want to model good interactions with them for our children. I also worry that they are slowly taking away our need to actually think for ourselves?”</em></p><p>This captures something profound about the generational moment we’re in. We’re responsible for both adopting AI responsibly ourselves AND preparing our kids for an AI-driven world we don’t fully understand. In defense contexts, we’re also responsible for maintaining critical thinking skills that could literally be matters of life and death.</p><p>The weight of that responsibility — personal, professional, societal — is real. And it’s showing up in how people approach AI adoption.</p><h3>What I’m Starting to Understand</h3><p>These conversations with my friends and family have clarified something I’ve been wrestling with in my professional work. The challenge isn’t just about making AI interfaces user-friendly. It’s about designing for users whose understanding of what AI is and does may be fundamentally incomplete or incorrect.</p><p>In commercial applications, this knowledge gap might mean suboptimal results or minor frustrations. But I design systems for environments where understanding AI limitations could be critical. If smart people in my life have these gaps in understanding, what does that mean for the professionals using high-stakes systems?</p><p>The disconnect between feeling “somewhat familiar” and actually grasping concepts like hallucinations, prompt engineering, and data reliability has me thinking about interface design in new ways. How do you build appropriate confidence and skepticism into systems when users might not realize what they don’t know?</p><p>How do you educate users about AI capabilities and limitations through the interface itself, rather than expecting them to become AI experts before they can use the system safely and effectively?</p><h3>The Education Challenge I’m Grappling With</h3><p>The responses about wanting guidance rather than more tools to evaluate really resonates with something I’ve been thinking about: traditional training doesn’t work for AI literacy in professional contexts. People don’t have time for courses. They need to learn through doing.</p><p>But how do you design systems that teach users about AI reliability, limitations, and appropriate use cases while they’re actually using the system? How do you build skepticism and critical thinking into the user experience without creating analysis paralysis?</p><p>These aren’t rhetorical questions for me — they’re design challenges I’m actively working through.</p><h3>Moving Forward: What This Means for My Work</h3><p>The people in my study aren’t early adopters or tech skeptics. They’re pragmatic mid-career adults — the same demographic that makes up much of the government and professional workforce. Their approach to AI adoption will likely influence how thoughtfully or haphazardly this technology gets integrated across many sectors, including the high-stakes environments where I work.</p><p>Right now, they’re trying to figure it out as they go. And understanding how they process AI adoption helps me think about the users I design for, where “figuring it out as we go” in high-stakes environments requires much more thoughtful design approaches.</p><p>I don’t have all the answers yet. I’m still processing what these patterns mean for my work and how to design systems that bridge the gap between user perception and AI reality. But I know this much: we can’t keep designing AI interfaces as if users have accurate mental models of what these systems are and how they work.</p><p>The question I’m sitting with now: How do we design AI systems that meet users where they actually are in their understanding, rather than where we wish they were?</p><p><em>This research is part of my personal exploration of AI adoption patterns among people in my demographic. I’m pursuing this because I see connections between how my friends and family approach AI and the challenges I face designing systems for Defense sector professionals. Understanding these knowledge gaps and mental models helps inform how I think about user experience design in high-stakes environments.</em></p><p><em>In the spirit of transparency about AI collaboration, I worked with Claude to organize and refine these insights from my survey responses. The analysis, conclusions, and personal reflections are my own, with AI assistance in structuring the narrative and improving clarity.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4a67256a4941" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>