<?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 Pete Barber on Medium]]></title>
        <description><![CDATA[Stories by Pete Barber on Medium]]></description>
        <link>https://medium.com/@peter-barber?source=rss-292ca2ed8bb2------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*ocky0jqfX1dNHMW5MahqxA.png</url>
            <title>Stories by Pete Barber on Medium</title>
            <link>https://medium.com/@peter-barber?source=rss-292ca2ed8bb2------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 14:00:21 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@peter-barber/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[Agent reliability is a design problem — not a model problem]]></title>
            <link>https://peter-barber.medium.com/agent-reliability-is-a-design-problem-not-a-model-problem-53045202a3e2?source=rss-292ca2ed8bb2------2</link>
            <guid isPermaLink="false">https://medium.com/p/53045202a3e2</guid>
            <category><![CDATA[writing-prompts]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[design]]></category>
            <category><![CDATA[product-design]]></category>
            <category><![CDATA[product-management]]></category>
            <dc:creator><![CDATA[Pete Barber]]></dc:creator>
            <pubDate>Mon, 09 Mar 2026 02:28:56 GMT</pubDate>
            <atom:updated>2026-03-09T02:28:56.605Z</atom:updated>
            <content:encoded><![CDATA[<p><em>A 5-minute read for designers and PMs building with AI</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*JeRa0WROKe9gkJNS.png" /></figure><p>I ran the same task through two agents.</p><p>Same model. Same tools. Same prompt.</p><p>Both hit a bad source claiming Notion had partnered with Slack. False — someone had conflated an existing integration with a new strategic move.</p><p><strong>Agent A</strong> shipped it as a finding. Confident output. Clean formatting. Fabricated basis.</p><p><strong>Agent B</strong> flagged it. Cross-referenced three sources. No corroboration. Threw it out.</p><p>Same intelligence. One delivered a hallucination dressed as strategy. The other caught it cold.</p><p>The difference wasn’t the model. It was the design of the job.</p><h3>The gap nobody is designing for</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*mRAhRPxei-LxZ2c0.png" /></figure><p>When an agent fails today, the industry reaches for the same fix: better models, more context, faster inference. These are engineering solutions to what is, at its root, a design problem.</p><p>Nobody designed what the agent should <em>do</em> between receiving a task and delivering a result. The goal is described. The tools are described. The process — the sequence of decisions, checks, and judgment calls between intent and outcome — is left to chance.</p><p>This gap is familiar territory for designers. We’ve spent decades between what a system can do and what it should do. That gap is now the central problem in agent development. And almost nobody working on it has a design background.</p><h3>I ran an experiment</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*yHdWMf2h3Yo1-p0P.png" /></figure><p>I built three agents for the same task: research a named competitor, identify their top three strategic moves in the last 90 days, and produce prioritised recommendations.</p><p><strong>Agent A</strong> got the standard treatment: task description, tools, output format. Thirteen lines. No workflow structure. This is how most agents ship.</p><p><strong>Agent B</strong> was structured around Anthony Ulwick’s Jobs-To-Be-Done Job Map — eight phases that describe the shape of any job regardless of who’s doing it: Define, Locate, Prepare, Confirm, Execute, Monitor, Modify, Conclude. Fifty-seven lines.</p><p><strong>Agent C</strong> got a generic checklist with the same number of steps as B, but no JTBD structure — things like “check that you understand the task.” A control to test whether <em>number of steps</em> matters, or <em>content of steps</em>.</p><p>Thirty trials. Blinded scoring across five metrics (0–3 scale, 15 points max): task completion, source quality, recommendation quality, self-correction, confidence calibration.</p><h3>What the data showed</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*hi3NRgF1UrEdoKEb.png" /></figure><p>Agent B won — and not narrowly by a long shot. But the more important number is the standard deviation.</p><p>Agent B’s SD was <strong>0.94</strong>. Agent A’s was <strong>1.94</strong>. The structured agent doesn’t just score higher — it scores <em>consistently</em>. Agent A’s range ran from 9 to 15. Any given run might be excellent or useless. Agent B’s floor was 11.5.</p><p>You can build a workflow around a tool that’s reliably good. You can’t build one around a tool that’s unpredictable.</p><h3>Where the advantage actually lives</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*e7wSK5FPzezKcRGq.png" /></figure><p>Self-correction was the standout metric. Agent B scored 2.83 vs. 2.10 for A and 2.17 for C.</p><p>The Confirm, Monitor, and Modify phases force the agent to verify claims against multiple sources, review its own output for contradictions, and fix errors before finalising.</p><p>Agent C’s self-correction score was barely above Agent A’s — despite having the same number of steps. “Review your work” doesn’t drive self-correction. “Review for contradictions, unsupported claims, logical gaps, recency violations, and whether each recommendation traces to cited evidence” does.</p><p><strong>Specificity is the active ingredient.</strong></p><h3>The finding that broke my hypothesis</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*uGsbES9Poam2RsMm.png" /></figure><p>I expected the framework to matter most on hard tasks — more synthesis, more opportunities for cascading errors.</p><p>The data said the opposite.</p><p>On simple tasks, Agent B led by 3.0 points over A. On hard tasks, the gap compressed to 0.25 — within noise.</p><p>The explanation: hard tasks force thoroughness on their own. When the problem is complex, the agent naturally slows down and checks more carefully. The task itself imposes the discipline that the framework provides.</p><p>On simple tasks, that forcing function disappears — and that’s where the failures hide.</p><p>One trial made this concrete. The task: research a mid-size analytics company, identify recent moves.</p><p>Agent A found a source about a new enterprise pricing tier. Clean write-up. Cited. The recommendation followed logically. The source was a two-year-old blog post about a pricing page update — not a strategic move, not from the last 90 days.</p><p>Agent A never checked the date because the task didn’t feel hard enough to warrant it.</p><p>Agent B’s Prepare phase had already defined “last 90 days” as a filtering criterion before any search ran. The stale finding never made it past the first gate.</p><p><strong>Process design isn’t a safety net for hard problems. It’s a forcing function for the ones that feel easy but aren’t.</strong></p><h3>The streamlined version</h3><p>The full eight-phase JTBD map works, but the data suggests a leaner structure captures most of the benefit. Four phases:</p><p><strong>Scope</strong> — Before doing anything: what does a successful output look like? What constraints apply? What’s in scope and what isn’t?</p><p><strong>Plan</strong> — Before executing: what are the 3–5 specific actions you’ll take, and why? Which sources will you trust? What categories of output are you looking for?</p><p><strong>Execute with verification</strong> — Do the work. For every finding you intend to include, verify it against at least two independent sources. Flag anything that rests on a single source.</p><p><strong>Self-review</strong> — One structured pass over the completed output. Check for contradictions, logical gaps between evidence and recommendations, anything that violates the constraints from Scope. State what you checked and what passed.</p><p>The structural pattern is the constant. The content inside each phase changes per domain. A research agent needs source cross-referencing. A code generation agent needs test coverage. A customer service agent needs policy compliance checks. The verification method changes. The fact that verification happens at all is what the framework guarantees.</p><h3>We shipped it on a production system</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ZfKsXc6cXZGBtk7P.png" /></figure><p>The experiment was competitive research. We wanted to know if it transferred.</p><p>We applied the same 4-phase structure to an agentic reverse engineering system — a completely different domain with different failure modes and verification requirements. We adapted the phase content but kept the structural pattern identical.</p><p><strong>Our benchmark went from 83% to 94.5%.</strong></p><p>That’s an 11.5-point jump on a production system doing real work.</p><h3>Where are the designers?</h3><p>Mostly, they’re not in the room.</p><p>Agent development happens inside engineering teams and ML research groups. Design involvement, where it exists, tends to start at the interface layer — the chat UI, the output format. Not the process layer, where the agent is deciding what to search, what to trust, and when to stop.</p><p>The entry point is closer than it looks. Take an agent prompt your team is using in production. Map it against the four phases above. Ask which ones are covered and which are missing. Almost always, it’s pure Execute — the agent has been told what to produce but not how to plan, verify, or check its own work.</p><p>That gap is a design brief.</p><p>You don’t need to understand transformers. You need to understand where processes fail — where the gap between intent and outcome opens up, and how to make the checks that close it explicit in an instruction set. That’s process design. Designers have been doing it for decades under different names.</p><h3>So what?</h3><p>Agents fail at the same phases humans fail at. They skip validation. They skip self-review. They shortcut on problems that feel easy.</p><p>Stop gold-plating your process for complex tasks. Start designing guardrails for the ones that feel simple.</p><p>That’s where the failures are hiding.</p><p>The model isn’t the moat. <strong>The design is.</strong></p><p><em>-Pete</em></p><p><em>Tags: AI · Design · Product Management · Prompt Engineering · Agentic AI</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=53045202a3e2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Design the Job]]></title>
            <link>https://peter-barber.medium.com/design-the-job-d30304e0a354?source=rss-292ca2ed8bb2------2</link>
            <guid isPermaLink="false">https://medium.com/p/d30304e0a354</guid>
            <category><![CDATA[design]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[prompt-engineering]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Pete Barber]]></dc:creator>
            <pubDate>Sun, 01 Mar 2026 21:02:13 GMT</pubDate>
            <atom:updated>2026-03-01T21:02:13.729Z</atom:updated>
            <content:encoded><![CDATA[<p>Agentic-Centred Design: 30 trials, one finding — agent reliability is a design problem.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*V0LabVqdhFcgbMD0zdy8HA.png" /><figcaption>An introduction to agentic-centred design</figcaption></figure><p>I asked two agents to perform market research.</p><p>Two agents. Same model. Same tools. Same task.</p><p>Both hit a bad source claiming Notion had partnered with Slack. False — someone had conflated an existing integration with a new strategic move.</p><p><strong>Agent A</strong> shipped it as a finding. Built a recommendation on top of it. Confident output. Clean formatting. Fabricated basis. If someone had acted on it, engineering weeks would have burned countering a competitive move that didn’t exist.</p><p><strong>Agent B</strong> flagged the claim. Cross-referenced three sources. No corroboration. Threw it out. Replaced it with something real.</p><p>Same intelligence. Same information. One delivered a hallucination dressed as strategy. The other caught it cold.</p><p>The difference was the design of the Job.</p><p>I spent the next several days testing whether that difference held — and whether I could explain why. The answer was yes, but not for the reasons I expected.</p><h3>Agents Have a Design Vacuum</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YcRCjCqNn2Ono91_ySEiXQ.png" /><figcaption>Blind trust = inherit risk</figcaption></figure><p>Building an AI agent today looks like this: write a prompt, wire up some tools, run it until the demo works. Ship.</p><p>The demo always works.</p><p>Demos are controlled environments with clean inputs and someone who understands the system watching. The product is the thousand runs after — messy inputs, contradictory sources, edge cases nobody mapped because no process required anyone to map them.</p><p>When those runs fail, the industry reaches for the same fix: better models, more context, faster inference. These are engineering solutions to what is, at its root, a design problem. Nobody designed what the agent should <em>do</em> between receiving a task and delivering a result. The goal is described. The tools are described. The process — the sequence of decisions, checks, and judgment calls between intent and outcome — is left to chance.</p><p>This is familiar territory for designers. We’ve spent decades in the gap between what a system can do and what it should do. Between capability and reliability. Between a feature working and a user succeeding.</p><p>That gap is now a central problem in agent development. And almost nobody working in it has a design background.</p><h3>Borrowing a Framework</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*eebdJthr6l8jd02-bK3pWg.png" /><figcaption>Anthony Ulwicks Universal Job Map</figcaption></figure><p>Jobs-To-Be-Done theory has a tool most practitioners skip past: the Universal Job Map. Anthony Ulwick developed it in the 1990s.</p><p>It breaks any job into eight phases — Define, Locate, Prepare, Confirm, Execute, Monitor, Modify, Conclude. His research tested this structure across surgery, tax filing, supply chains, software deployment. The phases held across all of them.</p><p>What makes this relevant to agents: the Job Map <strong>doesn’t care who’s doing the work.</strong> It describes the shape of the job itself. Which means it should describe what an agent does — and the phases the agent skips should predict where it fails.</p><p>I designed an experiment to test if this was the case. The results were more complicated than I wanted, and more useful because of it.</p><h3>The Experiment</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DAPLzaOBg9o7V5RsQGogMg.png" /><figcaption>3 Agents, same tools — different workflow</figcaption></figure><p>I built three agents for the same task: research a named competitor, identify their top three strategic moves in the last 90 days, produce prioritized action recommendations.</p><p><strong>Agent A</strong> got the industry-standard treatment. Task description, tools, output format. Thirteen lines. No workflow structure. This is how most agents ship.</p><p><strong>Agent B</strong> got the same task restructured around all eight Job Map phases. Define success criteria before acting. Qualify sources by reliability. Plan search queries before executing. Cross-validate every finding against at least two independent sources. Synthesise only after validation. Self-review for contradictions and logic gaps. Revise. Then check the output against the criteria defined at the start. Fifty-seven lines.</p><p><strong>Agent C</strong> got a generic checklist with the same number of steps but no JTBD structure — instructions like “check that you understand the task” and “search for relevant information.” This was the control for a specific question: does the number of steps matter, or does the content of those steps matter?</p><p>Thirty trials across three complexity tiers — ten simple (well-known, focused companies), ten medium (moderate information density), ten hard (large enterprises with complex strategic landscapes). All outputs were blinded.</p><p>An AI judge scored each output on five metrics (0–3 scale, 15 points maximum): task completion, source quality, recommendation quality, self-correction, and confidence calibration. Each output was scored in two independent passes and the scores averaged.</p><p>Eight additional ablation trials removed one phase at a time from Agent B’s prompt, holding everything else constant.</p><h3>What the Data Shows</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kDNyA-kMfU_vF6-e2qwNew.png" /><figcaption>Agents following a JTBD Framework delivered more accurate results.</figcaption></figure><p>Agent B beat both controls with a Cohen’s d above 1.0 against each — large by any convention. It also won 75% of its 60 pairwise comparisons against A and C.</p><p>But look at the standard deviations. Agent B’s SD is 0.94 — roughly half of Agent A’s 1.94. The JTBD-structured agent doesn’t just score higher. It scores consistently. Agent A’s range runs from 9 to 15; on any given run, you might get something good or something useless. Agent B’s floor is 11.5. That consistency matters more than the average for anyone trying to put an agent into a real workflow. You can build a process around a tool that’s reliably good. You can’t build one around a tool that’s unpredictable.</p><h3>Where the advantage lives</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1rP3frMUFZyPgfENaASmyQ.png" /></figure><p>Self-correction is the standout. Agent B scores 2.83 versus 2.10 for A and 2.17 for C — a gap of +0.73 and +0.66, far larger than any other metric. The Confirm, Monitor, and Modify phases force the agent to verify claims against multiple sources, review its own output for contradictions, and fix errors before finalising.</p><p>The other four metrics show narrower gaps, which suggests task completion and source quality are more constrained by the underlying model’s capability than by prompt structure. You can’t prompt your way to knowledge the model doesn’t have. But you can prompt the model to check its own work — and that’s where the structure earns its keep.</p><p>Agent C’s self-correction score (<em>2.17</em>) is barely above Agent A’s (<em>2.10</em>), despite having the same number of steps. “Review your work” doesn’t drive self-correction. “Review for contradictions, unsupported claims, logical gaps, recency violations, and whether each recommendation traces to cited evidence” does. <strong>Specificity is the active ingredient.</strong></p><h3>The complexity curve was backwards</h3><p>This is where my hypothesis <strong>broke</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vHabFs7UZhsGMA8v9zQ4LA.png" /><figcaption>Harder problems inherently prescribe more effort.</figcaption></figure><p>I expected the framework to matter <strong>most on hard tasks</strong> — more steps, more synthesis, more opportunities for cascading errors. The data said the opposite.</p><p>On simple tasks, Agent B led by 3.0 points over A and 2.15 over C. On hard tasks, the gap compressed to 0.25 and 0.40 — within noise given the judge’s consistency limitations.</p><p>The explanation, once I sat with it, made sense. Hard tasks force thoroughness on their own. When the problem is complex enough, the agent naturally slows down, checks more sources, hedges more carefully. The task itself imposes the discipline that the framework provides. On hard problems, all three agents converge upward because the problem itself demands rigour.</p><p>One of the simple-tier trials made this concrete.</p><p>The task was straightforward: research a mid-size analytics company, identify their recent moves.</p><p><strong>Agent A </strong>ran the search, found a few sources, and produced a clean three-point analysis. One finding was that the company had launched an enterprise pricing tier. Crisp write-up. Cited a source. The recommendation followed logically: adjust our pricing in response.</p><p>The source was a two-year-old blog post about a pricing page update. Not a new enterprise tier. Not a strategic move. Not from the last 90 days. <strong>Agent A</strong> never checked the date because the task didn’t feel hard enough to warrant checking. <strong>Agent B</strong>’s Prepare phase had already defined “last 90 days” as a filtering criterion before any search ran. The stale finding never made it past the first gate.</p><p>This pattern repeated across simple-tier trials. The failures weren’t dramatic. They were quiet — plausible-sounding, well-formatted outputs built on something that didn’t survive basic verification. The kind of failure that gets acted on precisely because it doesn’t look like a failure.</p><p><strong>Process design isn’t a safety net for hard problems. It’s a forcing function for the ones that feel easy but aren’t.</strong></p><h3>The ablation results</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*K77ptfvz-fqN-E1nF-7VMg.png" /></figure><p>Seven of eight ablations produced no measurable degradation. Only removing Confirm — the phase requiring verification against two or more independent sources — dropped the score, and only by 1 point, concentrated in self-correction and confidence calibration.</p><p>This tells you two things:</p><ol><li><strong>The phases are partially redundant </strong>— the remaining seven compensate when one is removed.</li><li><strong>The framework’s value is holistic</strong> rather than attributable to any single phase. The structure creates a workflow discipline that survives local removal.</li></ol><p>Confirm’s slight edge is consistent with the per-metric data: source verification is the one mechanism that directly prevents the most common failure mode — unsupported claims presented with false confidence.</p><p>I expected to find critical phases. But like all great design, the true value here is <strong>end to end.</strong></p><h3>What This Means for Design</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*X4f9J1_8iGz0eMjTs2U2sg.png" /></figure><p>The data supports three claims and <strong>undermines a fourth.</strong></p><p>✅ <strong>Supported:</strong> Structured agent design dramatically outperforms unstructured. An agent with a designed process beats one without by a margin that changes whether the output is usable.</p><p>✅ <strong>Supported:</strong> The content of the structure matters far more than the presence of structure. A generic checklist is barely better than no checklist. Specific verification requirements, specific self-review criteria, specific planning steps — these move the scores.</p><p>✅ <strong>Supported:</strong> Process design has the highest leverage on tasks that seem easy. The framework’s value is inversely proportional to the task’s inherent complexity. It catches the failures that hide in straightforward-looking work.</p><p>❓<strong>Undermined:</strong> The full eight-phase JTBD map is <strong>not the optimal structure.</strong> It works, but it’s more ceremony than necessary. The data suggests a leaner structure captures the large majority of the benefit with less overhead.</p><h3>The Streamlined Version</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FrRreIhc6UiR1tWqHptWgQ.png" /></figure><p>The data points to three mechanisms that account for the advantage. They collapse into four phases:</p><p><strong>Scope —</strong> define what you’re doing, what good looks like, and what’s out of bounds before you start. This prevents drift and gives the agent criteria to check against at the end. (Collapses Define.)</p><p><strong>Plan —</strong> structure your search before executing it. Decide what sources to trust, what queries to run, what categories of finding you’re looking for. This prevents random searching and forces source quality thinking upfront. (Collapses Locate and Prepare.)</p><p><strong>Execute with verification —</strong> do the work, but validate as you go. Every finding needs corroboration from at least two independent sources. Single-source claims get flagged or discarded. This was the only phase whose removal produced measurable degradation in the ablation study. (Collapses Confirm and Execute.)</p><p><strong>Self-review —</strong> one structured pass over the completed output, checking for contradictions, unsupported claims, logical gaps between evidence and recommendations, and whether the output meets the criteria defined in Scope. (Collapses Monitor, Modify, and Conclude.)</p><p>Here’s what this looks like in practice. The template below is domain-agnostic — fill in the bracketed sections for your use case and it becomes a working system prompt.</p><h4>Phase 1 — Scope</h4><p>Before doing anything, state:</p><ul><li><strong>What does a successful output look like?</strong> Define the specific deliverable and quality bar.</li><li><strong>What constraints apply?</strong> Time window, word count, depth, format.</li><li><strong>What’s in scope and what isn’t?</strong> Explicitly exclude what the agent should ignore.</li><li><strong>Who is this for and what decision will it inform?</strong> Audience and stakes.</li></ul><p>The structural pattern — scope, plan, execute with verification, self-review — is the constant. The content inside each phase changes per domain. A competitive research agent needs source cross-referencing. A code generation agent needs test coverage. A customer service agent needs policy compliance checks. The verification method is different each time. The fact that verification happens at all is what the framework guarantees.</p><h4>Phase 2 — Plan</h4><p>Before executing, structure your approach:</p><ul><li><strong>What are the 3–5 specific actions or queries you’ll take, and why each one?</strong> Force the agent to articulate a strategy, not just start working.</li><li><strong>Which sources or inputs will you trust, and which will you treat with caution?</strong> Define reliability criteria for your domain — authoritative documentation vs. forum posts, verified datasets vs. user-submitted data, primary sources vs. aggregators.</li><li><strong>What categories of output are you looking for?</strong> Name them. The agent can’t find what it hasn’t been told to look for.</li></ul><h4>Phase 3 — Execute with verification</h4><p>Do the work. But for every claim, finding, or output you intend to include:</p><ul><li><strong>Verify it against at least two independent sources or validation methods.</strong> Adapt to domain: for research this means cross-referencing sources; for code it means testing against edge cases; for analysis it means checking the data.</li><li><strong>If something rests on a single source,</strong> flag it explicitly and assess reliability. Discard anything you cannot corroborate.</li><li><strong>Check every element against the constraints defined in Phase 1.</strong></li></ul><h4>Phase 4 — Self-review</h4><p>Review your completed output for:</p><ul><li><strong>Contradictions</strong> between sections.</li><li><strong>Logical gaps</strong> between what you found and what you’re recommending.</li><li><strong>Anything that violates the constraints from Phase 1.</strong></li><li>Then check the full output against the success criteria you defined in Phase 1. If anything fails, fix it. State what you checked and what passed.</li></ul><p>The structural pattern — scope, plan, execute with verification, self-review — is the constant. The content inside each phase changes per domain. A competitive research agent needs source cross-referencing. A code generation agent needs test coverage. A customer service agent needs policy compliance checks. The verification method is different each time. The fact that verification happens at all is what the framework guarantees.</p><h3>What happened when we shipped it?</h3><p>The experiment tested the framework on competitive research. We wanted to know if it transferred.</p><p>We applied the 4-phase structure to our own system — an agentic reverse engineering agent, a completely different domain with different failure modes and different verification requirements. We adapted the phase content (the scope criteria, the verification methods, the self-review checks) but kept the structural pattern identical: scope, plan, execute with verification, self-review.</p><blockquote><strong>Our benchmark went from 83% to 94.5%.</strong></blockquote><p>That’s not a marginal gain on a lab exercise. That’s an 11.5-point jump on a production system doing real work. The framework transferred across domains because the mechanisms transfer — forced planning, inline verification, structured self-review. The specific checks changed. The architecture didn’t.</p><p>This is also the strongest evidence against the concern that the experiment’s results were artefacts of a single task type. Competitive intelligence research and agentic reverse engineering share almost nothing in terms of domain knowledge, source types, or output format. What they share is the structure of the job — and <strong>that’s exactly what the framework designs for.</strong></p><h3>Where are the designers?</h3><blockquote>This piece has been arguing that agentic centered design— is a design discipline. That raises the obvious question: <strong><em>where are the designers?</em></strong></blockquote><p>Mostly, they’re not in the room. Agent development happens inside engineering teams, ML research groups, and founding teams of people who came up through code. Design involvement, where it exists, tends to start at the interface layer — the chat UI, the dashboard, the output format. Not at the process layer, where the agent is deciding what to search, what to trust, and when to stop.</p><p>The entry point is closer than it looks. If you’re a designer, here’s something you could do this week: take an agent prompt your team is already using in production. Map it against the four phases above. Ask which ones are covered and which are missing. Almost always, it’s pure Execute — the agent has been told what to produce but not how to plan, how to verify, or how to check its own work. <strong>That gap is a design brief.</strong></p><p>You don’t need to understand transformers. You don’t need to train models. You need to understand where processes fail — where the gap between intent and outcome opens up, what kinds of checks close it, and how to make those checks explicit in an instruction set. That’s process design. Designers have been doing it for decades under different names: service design, systems design, workflow mapping. The skill transfers. <strong>The domain is new.</strong></p><p>The people who move first will have a head start that compounds. Not because the techniques are secret — everything in this piece is reproducible — but because developing judgment about where agents fail and what process structures prevent those failures takes reps.</p><p>It takes running experiments like this one, seeing the patterns, building intuition for which verification steps matter in which contexts. That intuition is the real asset. It doesn’t come from reading about it.</p><h3>The Open Question</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FrRreIhc6UiR1tWqHptWgQ.png" /></figure><p>Agents fail at the same phases humans fail at. They skip validation. They skip self-review. They shortcut on problems that feel easy.</p><p>What I didn’t expect was where the framework would matter most. Not on the hard problems, where complexity forces its own discipline. On the easy ones. The ones where everything looks fine until someone acts on the output and discovers the foundation was a two-year-old blog post.</p><p><strong>That’s a design insight, not an engineering insight.</strong> And it changes where you invest your effort.</p><p>Stop gold-plating your process for the complex tasks. Start designing guardrails for the ones that feel simple.</p><p>That’s where the failures are hiding.</p><p>The model isn’t the moat. <strong>The design is.</strong></p><p><em>-Pete</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d30304e0a354" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Beyond the Unicorn: The Next Era of Design Leadership]]></title>
            <link>https://peter-barber.medium.com/beyond-the-unicorn-the-next-era-of-design-leadership-9bb252b24c8c?source=rss-292ca2ed8bb2------2</link>
            <guid isPermaLink="false">https://medium.com/p/9bb252b24c8c</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[future-of-work]]></category>
            <category><![CDATA[product-design]]></category>
            <category><![CDATA[design-leadership]]></category>
            <dc:creator><![CDATA[Pete Barber]]></dc:creator>
            <pubDate>Fri, 24 Oct 2025 04:32:34 GMT</pubDate>
            <atom:updated>2025-10-24T04:44:34.661Z</atom:updated>
            <content:encoded><![CDATA[<p>In an age of intelligent tools, our greatest craft is still human.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nh2h3IbaRcJoDD7UEvQFsw.jpeg" /></figure><p>Give me a hammer, and I might build something that looks like a house. It might stand upright, but it wouldn’t be safe, comfortable, or up to code. Hand that same hammer to a builder, and you’d get something entirely different. They see what I can’t. They have a process, an eye trained by experience, and a respect for craft that turns the same resources into something special.</p><p>AI is not so different. It’s another tool — a powerful one — but still just a tool. The real question isn’t whether it replaces us, but how it extends us.</p><p>Tools can replicate capability, but not craft. Craft lies in professional judgment, sensitivity, and the wisdom to discern the definitions of what truly matters.</p><h4>The Difference Between Shelter and a Home</h4><p>In empowered product teams, everyone designs in some way. As Marty Cagan says, we “design the work.” Yet design as a discipline carries a deeper responsibility: to create with meaning. If you want something that lasts, something that connects emotionally and builds trust — give it to a designer.</p><p>If you just need a quick shelter from the rain, you can probably get by without one. Both have value.</p><p><em>The danger lies in confusing the two.</em></p><p>Design has always been a practice of influence rather than ownership. It’s quiet leadership built on persuasion, empathy, and relationships.</p><p>The best work doesn’t live in isolation; it emerges through collaboration. That’s the paradox: design attracts reflective people who lead thoughtfully from behind the scenes, yet our greatest impact depends on how well we navigate the messy, human machinery of teams.</p><p>To bring an idea to life, we must move it through layers of alignment, compromise, and trust. Real leadership in design means helping others see themselves in the vision, <em>not just asking them to believe in ours.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BGLnrtEu77dY5YOGHZkuQw.jpeg" /></figure><h4>Beyond the Unicorn: The Next Era of Design Leadership</h4><p>As the landscape grows more technical, the way we drive change must evolve too. The mythical “unicorn” designer — someone who could code, research, design, and prototype everything alone — was born from scarcity. Today, with the power of AI and shared tooling, that scarcity is gone. We don’t need unicorns anymore; we need something different.</p><p>Think of this new era less as individual mastery and more as collective orchestration. The work ahead isn’t about doing it all ourselves; it’s about designing <em>how</em> work happens. The next wave of impact won’t come from those who can use every tool, but from those who can shape systems of people, processes, and principles that allow great work to emerge.</p><p>We’ve spent years designing products and systems. Now, we have the opportunity to design <em>teams</em> and <em>systems of work.</em> That’s where craft lives next, not in control, <em>but in orchestration</em>. Guiding alignment, nurturing clarity, and creating the conditions where everyone can do their best work. As the tools become more capable, the real leverage shifts from what we make to <strong><em>how we make it together.</em></strong></p><p>Still, at the heart of it all lies something deeply human. We call interfaces “friendly” or “approachable” because they inherently carry emotion. Great design builds trust, and that trust isn’t the product of design alone. Trust is the sum of every hand that touches the work. Craft now isn’t about guarding the process; it’s about nurturing meaning across disciplines.</p><p>AI may change how we create, but not why we create. Our responsibility is to ensure that as our tools grow smarter, our work grows more human.</p><p>Because in the end, anyone can build a house. It takes a designer to make it feel like home. And the future will always need more homes.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9bb252b24c8c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to Design for the World of Agentic AI]]></title>
            <link>https://peter-barber.medium.com/how-to-design-for-the-world-of-agentic-ai-3d981b799852?source=rss-292ca2ed8bb2------2</link>
            <guid isPermaLink="false">https://medium.com/p/3d981b799852</guid>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[design]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[user-experience]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Pete Barber]]></dc:creator>
            <pubDate>Tue, 12 Aug 2025 01:32:45 GMT</pubDate>
            <atom:updated>2025-08-12T01:32:45.786Z</atom:updated>
            <content:encoded><![CDATA[<p>From tasks as function to jobs as progress, toward outcomes, trust, and stewardship.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*w0ZzuY5yqYdMnopvt9ziDA.png" /></figure><h3>The shift in the contract</h3><p>The interface used to be the product. We drew boxes, we argued about steps, we taught users our choreography.</p><p>Agentic AI changes the contract. People state intent, set a few constraints, and review outcomes. Sequence becomes an internal concern, orchestrated by an agent that can plan, branch, and adapt without dragging the user through it.</p><p>Our job shifts from path planning to outcome stewardship.</p><p>Define progress, set a few clear heuristics, and expose just enough visibility and control to earn trust.</p><p>Simple now means less cognitive debt, not fewer screens.</p><h3>The old world, steps and functions</h3><p>Jobs to be done helped teams stop worshipping features, but most adopted it as <strong>function</strong>.</p><p>We decomposed jobs into actions and acceptance criteria, then encoded them into linear flows. Useful for scoping, brittle for reality. Real life is high variance, edge cases are heavy, and users end up doing our exception handling.</p><p>Ulwick’s interpretation of JTBD as function fit the UI era of forms and funnels, where progress was synonymous with step completion. We optimised for path adherence, not meaningful movement. The hidden cost was cognitive load. Users had to learn our internal sequence, remember the state, and babysit, spoon-feeding our systems’ inputs.</p><p>That era taught discipline; it also <strong><em>trapped us in choreography.</em></strong></p><blockquote>How much of good design was additive?</blockquote><p>When variance spiked, we added more steps and knobs, which felt like control but rarely produced better outcomes. We were gatekeepers when the job needed a steward.</p><h3>The new world, agents and progress</h3><p>Christensen’s point was simple and stubborn, people do not hire products for features; they hire them to make progress in a specific circumstance.</p><p>Progress is movement from an unsatisfactory state to a more satisfactory one, under constraints, with real trade-offs. That frame differs from jobs as a function.</p><p><strong>Function</strong> lists actions.</p><p><strong>Progress</strong> names the arc, the struggle, the anxieties that slow movement, and the forces that help or hinder change.</p><p>Design inherited a bias for function because UI required it. We decomposed behaviour into steps, then asked users to carry out the sequence. This worked until variance rose, which it always does in real life.</p><p>Progress thinking absorbs variance. It treats each circumstance as a field of constraints, motivations, and tolerances, and asks, what meaningful state change are we stewarding here, not what steps must be completed.</p><p>Agentic systems finally let us operationalise that idea. An agent can hold a rich representation of circumstance, goals, and constraints, plan across multiple paths, adjust as signals change, and explain trade-offs in plain language.</p><p>Users no longer need to translate progress into the product’s steps. They can state intent, set boundaries, and let the system shoulder choreography.</p><p>The designer’s value shifts to shaping how the agent understands progress, how it balances trade-offs, and how it exposes rationale so the user can steer with confidence.</p><p>This is where agentic AI expands the frontier of ambition. Many outcomes used to be too complicated, not because they lacked value, but because the sequence to get there was brittle, and the cost of explaining each branch was too high. Agents can explore option space quietly, surface a small set of credible paths, and narrate the implications.</p><p><strong>Stewardship, and the emotional residue your product leaves, is now your responsibility.</strong> For context on designing for emotional residue over functional outcomes, see your earlier piece, <a href="https://uxdesign.cc/designing-for-emotional-residue-over-functional-outcomes-b1f5a26ef6bc"><em>Designing for Emotional Residue over Functional Outcomes</em></a>.</p><p>Experiences feel like magic — not because the system shows everything, — but because it hides the mess and shows the meaning tied to progress.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8bcmWbfhQV6ObmLEmEdLLQ.png" /></figure><h3>Hide the mess, show the meaning</h3><p>Agents make sequence an internal concern. That forces a simple question: if the steps disappear, what should remain visible?</p><p>In a progress-first frame, people do not hire products to witness choreography; they hire them to feel movement from not yet to now we are getting somewhere.</p><p>Meaning is the smallest honest narrative about progress, what we are aiming for, what trade-offs we are making, and what has changed.</p><p>When experience centres meaning, users do not need to babysit mechanics. They need to see themselves moving forward, with enough context to steer. That is the core JTBD reframing in the agentic era, from <strong><em>teaching users our sequence to stewarding their progress under constraints.</em></strong></p><p>Hide the mess, it is editorial judgment.</p><p>Explain intent and implications, keep wiring backstage, reveal details only when they change the user’s decision.</p><p>This is the new transaction that exists in experiences today. How interfaces stay quiet while ambitions get bigger.</p><p>The systems carry more operational burden, so that our users have a greater sense of authorship over what matters.</p><h4>Three tenets of agentic design</h4><p>Agentic AI does not change what people hire products for, it changes what the product can carry. When the sequence disappears into the system, judgment and intention move to the foreground. The three tenets that follow are not a checklist, they are a standard for stewardship, a way to keep progress legible while ambition scales.</p><ol><li><strong>Progress first</strong>, jobs are movement under constraint, not a checklist. Design around the user’s state change, let sequence be an internal concern.</li><li><strong>Magic, not mystery</strong>, ambition can grow while interfaces stay quiet. Hide mechanics, show intent, trade-offs, and what changed, the parts that make progress legible.</li><li><strong>Visibility + control = trust. </strong>Match what you show with what the user can do, so autonomy can be safely delegated.</li></ol><p>In a progress-first world, users lend autonomy to a system that acts on their behalf. That loan is underwritten by a simple contract, show me what matters and let me steer when it matters.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CB-Xsv2jGZjJxcqPD1HzwQ.png" /></figure><h3>Outcomes as the currency</h3><p>In an agentic world the steps recede, the outcome becomes the contract.</p><p>People do not hire your product to witness sequence, they hire it to see their world change in ways that matter to them. Progress is the unit of value, not path completion, not API calls, not tokens spent.</p><p>Outcomes are not outputs. Outputs tally what the system did. Outcomes describe what improved for the user, within the constraints they set, at a cost they accept. Treat definition of done as a human promise, clear enough that a person could agree to it before action, and recognise it after action without translating your internal metrics.</p><p>Make outcomes legible first, then evidenced, then contextualised. Start in the user’s language, what changed, what it cost, what trade-offs were made, and how confident we are that it was the right move. Only then attach the agent’s evidence, the signals and rationale that underwrite the result.</p><p>Put the user’s outcome first, then the agent’s evidence, then business impact as a consequence.</p><p>Order signals <strong><em>who you steward.</em></strong></p><p>Learning requires consent and reciprocity. If the system collects outcome telemetry to improve stewardship, the value exchange must be obvious, clearer guidance next time, tighter defaults, fewer questions, better results under the same constraints.</p><p>Let people choose the depth of telemetry, and let that choice shape how boldly the agent acts. Stewardship respects boundaries as it expands capability.</p><p>Finally, tie outcomes to recourse. When the result misses the user’s definition of done, show a credible path to make it right, and let that correction refine future behaviour. Trust grows when progress is visible, controllable, and correctable.</p><h3>TL;DR</h3><p>Agentic AI lets products carry sequence so people can carry intent. The bar for simplicity is lower cognitive debt, not fewer clicks. Design for progress, practise stewardship, and earn trust with visibility and control. Hide the mess, show the meaning, and let outcomes be the currency that proves it.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3d981b799852" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[You Can’t Automate Perspective]]></title>
            <link>https://peter-barber.medium.com/you-cant-automate-perspective-bc7d1393a0dc?source=rss-292ca2ed8bb2------2</link>
            <guid isPermaLink="false">https://medium.com/p/bc7d1393a0dc</guid>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[design]]></category>
            <category><![CDATA[future-of-work]]></category>
            <dc:creator><![CDATA[Pete Barber]]></dc:creator>
            <pubDate>Mon, 28 Apr 2025 23:38:09 GMT</pubDate>
            <atom:updated>2025-04-28T23:38:31.902Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VE9vEK1GfzR4eODBxrNo1A.png" /><figcaption>Why we’ll still matter</figcaption></figure><h4>Why Cross-Functional Teams Will Still Matter</h4><p>There’s a rising fantasy in product teams right now — that AI will finally let us replace each other.</p><p>Product Managers coding. Designers shipping full-stack apps. Engineers making their own UX calls. Everyone becoming a team of one.</p><p>At first, it sounds efficient: fewer handoffs, faster output, total control. But here’s the truth — <strong>the work might move faster, but the outcomes will get worse.</strong></p><blockquote>“You can’t shortcut perspective without sacrificing meaning.”</blockquote><p>Great products don’t come from solo operators. They come from <strong>collaboration, contradiction, and perspective</strong> — from a designer pushing back on a question of user value, from an engineer flagging hidden technical debt, from a PM connecting dots no one else saw.</p><p>That’s not inefficiency.<br><strong>That’s the system working.</strong></p><p>Cross-functional teams weren’t built to slow things down — they were built to <strong>bring quality, relevance, and empathy into the room at the same time</strong>. It’s that convergence of perspectives that keeps products grounded in reality and aligned to what actually matters.</p><p>AI gives individuals more leverage. But leverage without perspective is a liability.</p><p>Collaboration isn’t just a quality check. <strong>It’s the system that creates quality. </strong>It’s how we make products that mean something to the people who use them.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*V1rx4FqA45a8DUz2IWRgmg.png" /></figure><h4>The Roles Aren’t Dead — Just the Handoffs</h4><p>AI isn’t eliminating roles.<br> It’s eliminating the wait.</p><ul><li>“Can you do this when you have time?”</li><li>“Let me know when the design’s done so I can start.”</li><li>“We’re blocked on dev, so everything’s paused.”</li></ul><p>And thank God.</p><p>Those delays weren’t about roles. They were about <strong>processes that couldn’t keep up</strong> — rigid workflows, siloed tools, and invisible dependencies that slowed teams down and frayed trust.</p><p>Now, with AI in the mix, the friction shifts. Designers can explore technical implications without waiting on engineering. Product Managers can simulate workflows without needing wireframes first. Engineers can interrogate UX decisions and generate alternatives.</p><p>Work can overlap, flex, and self-initiate.</p><p>But that doesn’t mean we no longer need each other.<br> It means we need <strong>each other’s judgment more than ever.</strong></p><blockquote>“Speed without shared context is just waste in disguise.”</blockquote><p>The best teams aren’t blending roles into a single person. They’re blending capabilities across a <strong>cohesive, interdependent system</strong> — fewer handoffs, more overlap. Less waiting, more thinking together.</p><p>AI lets us move faster. But it’s up to us to move smarter — and that means keeping the roles, while finally letting go of the walls between them.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kpC1BZCgrhVXntxN4_xMqA.png" /></figure><h4>Designing Software Is Designing a Service</h4><p>If we can accept that a product is more than its features, we also need to accept this:</p><blockquote><strong><em>The process that makes a product is itself a service — not to users, but to the organization.</em></strong></blockquote><p>Designers, Engineers, and Product Managers aren’t just contributors with isolated tasks. Together, we’re a <strong>service team</strong> — delivering clarity, reducing friction, and moving ideas into reality. And just like with any good service, <strong>how we operate matters as much as what we deliver.</strong></p><blockquote>“Design is how it works. But first, design how you work.”</blockquote><p>AI increases surface-level productivity — but it doesn’t replace the messy, human choreography of building together. If anything, it exposes the weak spots.</p><p>Now’s the time to turn our tools inward.</p><p>Designers can apply service blueprints to team collaboration. Engineers can optimize not just code, but the system of how code gets made.<br> Product Managers can treat the team as a user — mapping pain points, resolving blockers, shaping flow.</p><p><strong>The experience of working together is part of the product. </strong>And in an AI-accelerated world, designing that internal experience — that invisible service layer — is more valuable than ever.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SQc1HHEldFTvbDb3-qmZzA.png" /></figure><h3>Vision Becomes the Real Constraint</h3><p>When AI lowers the cost of execution, the challenge doesn’t become how to build — it becomes <strong>what to build.</strong></p><p>It’s no longer “Can we do this?”<br>It’s “<strong>Should</strong> we do this?”</p><p>And more importantly — “<strong>What do we refuse to do?</strong>”</p><blockquote>“When you can build anything, clarity becomes the rarest resource.”</blockquote><p>AI gives us endless options.<br>And endless options without focus is a distraction machine.</p><p>The constraint is no longer resources or talent.<br><strong>It’s vision</strong> — the ability to define what matters, to draw the line around purpose, and to stay inside it.</p><p>This is where strong product leadership matters most.<br>The job isn’t just prioritization — <strong>it’s meaning-making.</strong></p><p>It’s shaping context, naming tradeoffs, and holding the line on why we’re building anything at all.</p><p>AI makes everything easier — except vision.<br> That still takes people.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wG1OKtfn9e7UJgYz-0MzFw.png" /></figure><h3>The Scope of a Role Is Expanding, Not Collapsing</h3><p>There’s a myth that AI will flatten roles into a single, do-it-all function.<br> But we’re not seeing collapse — <strong>we’re seeing expansion.</strong></p><p>Designers aren’t becoming irrelevant — they’re stepping beyond Figma. Engineers aren’t being replaced — they’re stepping into systems thinking. Product Managers aren’t reduced to backlog grooming — they’re designing focus itself.</p><p>Here’s the shift:</p><blockquote><strong><em>The tools we once reserved for users are now being turned inward — onto our own teams.</em></strong></blockquote><p>Designers can run service design sprints on the workflow itself. Product Managers can apply journey thinking to collaboration. Engineers can measure team velocity the same way they measure performance.</p><p>The goal isn’t to do more — it’s to work better. More aligned. More human. More aware of the systems we’re building inside the organization.</p><p>Because when the team functions well, the product inevitably gets better.</p><p>AI didn’t kill your role.<br><strong>It gave you more surface area to practice it with intention.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DeELVuPaMZSrZceJahnUWQ.png" /></figure><h4>Final Thoughts: You Can’t Automate Perspective</h4><p>AI will reshape how we work.<br>That’s inevitable.</p><p>But the idea that it replaces who we need — or why we work together — is a dangerous illusion.</p><blockquote>“You can generate a thousand versions of a feature — but only people can decide what matters.”</blockquote><p>Collaboration isn’t overhead.<br>Perspective isn’t optional.<br>And roles aren’t just lists of tasks — they’re living functions for judgment, care, challenge, and synthesis.</p><p>The future of work isn’t about fewer people doing more.<br>It’s about teams doing different work —<strong> faster, deeper, and with more shared clarity.</strong></p><p>So if you’re building with AI, here’s the real unlock:</p><blockquote><strong><em>Don’t just accelerate output. Elevate interaction.<br>Design how your team works — not just what your product does.</em></strong></blockquote><p>In a world of infinite generation, <strong>perspective is your rarest asset.</strong></p><p>Fortunately for us, you still can’t automate that.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bc7d1393a0dc" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Designing for emotional residue over functional outcomes]]></title>
            <link>https://uxdesign.cc/designing-for-emotional-residue-over-functional-outcomes-b1f5a26ef6bc?source=rss-292ca2ed8bb2------2</link>
            <guid isPermaLink="false">https://medium.com/p/b1f5a26ef6bc</guid>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[editor-picks]]></category>
            <category><![CDATA[ux]]></category>
            <category><![CDATA[product-design]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Pete Barber]]></dc:creator>
            <pubDate>Mon, 14 Apr 2025 03:17:04 GMT</pubDate>
            <atom:updated>2025-04-14T21:30:52.958Z</atom:updated>
            <content:encoded><![CDATA[<h4>Why design’s most human contribution is now its most strategic advantage</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*tb5l0aQ0Y90F_jbp" /><figcaption>Photo by <a href="https://unsplash.com/@vackground?utm_source=medium&amp;utm_medium=referral">vackground.com</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h3>The shift that’s already happening</h3><p>More and more, our tools are designing with us, or for us.</p><p>OpenAI is now building an AI software engineer, capable of doing everything a human developer does — from planning and writing code to testing it and managing pull requests. It’s not a concept. It’s already shipping. <a href="https://officechai.com/ai/openai-is-building-a-swe-an-ai-that-does-everything-a-human-software-engineer-does/">It’s happening now</a>.</p><p>We’re entering the era of agentic software engineering. Autonomous systems can scope, build, and deploy functional products with minimal human input. What used to take months can now happen in hours.</p><p>When execution becomes infinite, it stops being a differentiator. <strong>Functionality becomes a commodity.</strong></p><p>The question becomes clear:</p><p><strong><em>When anyone can build anything, what makes it worth returning to?</em></strong></p><p>It won’t be just what a product <em>does</em>. It’ll be what it <em>leaves behind</em>.</p><p>That’s <strong>emotional residue </strong>— the subtle signal that lingers after the feature is done and the tab is closed. The feeling that you were understood. The sense that something cared.</p><p>It’s not utility. <strong>It’s memory.</strong></p><p>In a future where code becomes cheap and automation is everywhere, emotional residue <strong>may become the most valuable output of design.</strong> And it will be the most human part of every product we make.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ct2KFivRxaoFnnVn" /><figcaption>Photo by <a href="https://unsplash.com/@vackground?utm_source=medium&amp;utm_medium=referral">vackground.com</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h4>Emotional Residue — the hidden layer that lingers</h4><p>Most products are designed to deliver outcomes.</p><p>A task is completed, a button is pressed, a notification arrives. It works. It functions. It passes the test.</p><p>But function isn’t the full story. Not anymore.</p><p>Emotional residue is what remains after the interaction ends. It’s the quiet impression a product leaves behind — the subtle signal that this was made with care.</p><p>It’s the tone of the copy when something goes wrong. The way a system lets you recover with dignity. The rhythm of a transition that breathes, instead of rushing to the next screen.</p><p>We don’t always remember what we tapped or typed. But we remember how it made us feel —<strong> competent, calm, confused, seen.</strong></p><p>That feeling, <strong>that residue</strong>, is more than a nice-to-have. It’s strategic. It builds trust. It drives repeat behaviour. It turns users into advocates.</p><p>And as generative tools take over execution, <strong>emotional residue becomes one of the few things AI can’t generate on its own.</strong> Because it’s not just about what gets built. It’s about<strong> what gets felt.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*uZPO3zWHimOCVj-o" /><figcaption>Photo by <a href="https://unsplash.com/@scheglovgr?utm_source=medium&amp;utm_medium=referral">Grigorii Shcheglov</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h4>The new frontier of design in agentic systems</h4><p>As agentic systems take on more of the building, the structure of product teams is already shifting.</p><p>Engineers are becoming system architects, not line-by-line implementers. PMs are steering outcomes, not grooming backlogs. Designers are moving from layout to logic, collaborating with models to define not just how things look, but how they behave.</p><p>Some argue this evolution will lead to clearer handoffs and tighter lanes. But the opposite is true. The lines are blurring — and that’s where the real opportunity lies.</p><p>Agentic tools don’t streamline handoffs, they collapse them. When a PM can generate a prototype or a designer can prompt a working flow, <em>who owns what</em> becomes<strong><em> less important than how we think together.</em></strong></p><p>This doesn’t reduce the need for collaboration — it intensifies it.<br>It demands shared intuition, shared context, and shared care for the user.</p><p>And the pressure is already here. <a href="https://www.forbes.com/sites/johnwerner/2025/04/10/leaked-shopify-memo-illustrates-bull-approach-to-ai/">Shopify recently told team leads they must justify why a task <em>can’t</em> be done with AI before opening a new headcount.</a> Across the industry, big tech companies are holding back hiring while leaning harder into agentic tooling and automation. The message is clear: the teams that remain must deliver more with less — and work more fluidly than ever.</p><p>That’s where design steps up. Not as a decorator, but as connective tissue.</p><p>The discipline that moves across silos, shaping cohesion where automation fragments it.</p><p>Design becomes less about artefacts, and more about <strong><em>alignment</em>.</strong></p><p>Less about ownership, and more about <strong><em>orchestration</em>.</strong></p><p>In a world where anyone can generate anything, the hardest thing to create is <strong><em>coherence</em></strong> — and that’s something only well-aligned, cross-functional teams can deliver.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*m2WlA_jW00XkvIY9" /><figcaption>Photo by <a href="https://unsplash.com/@googledeepmind?utm_source=medium&amp;utm_medium=referral">Google DeepMind</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h4>Why emotional residue will define great products</h4><p>When functional outcomes become commoditised, <strong>emotional resonance becomes the differentiator.</strong></p><p>This isn’t theory — it’s how people actually experience products. Users don’t return because a button worked. They return because the experience made sense. It respected their time. It gave them confidence.</p><p>Research indicates that users form lasting impressions based on how a product makes them feel, not just on its functionality. <a href="https://www.hci.org.uk/article/the-role-of-emotional-design-in-user-experience-a-comprehensive-analysis">A study highlighted in the <em>Journal of Interactive Design</em> demonstrated that incorporating emotional design elements led to a significant uplift in conversion rates and increased customer satisfaction levels.</a></p><p><strong>That’s emotional residue</strong> — and it drives real business outcomes.</p><p>It builds trust.</p><p>Shapes brand memory.</p><p>Increases retention.</p><p>It turns a moment of use into a lasting impression. And in a crowded market, <strong>those impressions compound.</strong></p><p>We see it in the products people love and advocate for.<br>- <a href="https://www.apple.com/au/">Apple</a> doesn’t just work — it<strong> <em>feels</em> <em>considered</em>.</strong><br>- <a href="https://www.figma.com/">Figma</a> doesn’t just load fast — it makes you<strong> <em>feel fast.</em></strong><br>- <a href="https://linear.app/">Linear</a> doesn’t just manage issues — it gives you a <strong><em>sense of momentum and clarity.</em></strong></p><p>These aren’t just design wins. They’re <strong>emotional signals</strong>, deeply aligned with the <strong>product’s core value.</strong></p><p>And in a world where every competitor can match your features,<strong><em> </em>how your product feels becomes the moat.</strong> The deeper emotional layers are the hardest to replicate.</p><p>They don’t come from prompts. They come from care, from context, from teams that sweat the details most users will never see —<strong> but always feel.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*JZpqoO6DjnHPUkx0" /><figcaption>Photo by <a href="https://unsplash.com/@googledeepmind?utm_source=medium&amp;utm_medium=referral">Google DeepMind</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h4>What execs should do about it</h4><p>If emotional residue is the new frontier, we need to design for it deliberately. That doesn’t mean adding polish at the end. It means rethinking how we prioritise, how we collaborate, and what we reward.</p><p><strong>1. Make emotional quality a first-class product concern.</strong><br> Don’t relegate it to the tail-end of design reviews. Bake it into the brief. Make it part of the definition of done. Treat tone, timing, and clarity as seriously as logic and layout.</p><p><strong>2. Shift from artefact ownership to shared emotional intent.</strong><br> In agentic environments, the boundaries between disciplines blur. Use that to your advantage. Align around how the product should <em>feel</em>, not just what it should <em>do</em>. Intent becomes the new spec.</p><p><strong>3. Invest in cross-functional design fluency.</strong><br> It’s not enough for designers to care about emotion. PMs, engineers, and AI agents all shape experience now. Build shared language and shared standards for emotional quality across roles.</p><p><strong>4. Use AI to compress execution, then spend that time on care.</strong><br> The win isn’t just faster delivery. It’s more space for depth. Let automation handle the repeatable work so humans can focus on the emotional craft — the things AI can’t yet feel.</p><p><strong>5. Measure what lingers, not just what completes.</strong><br> Traditional metrics track conversion and completion. But also look at retention, advocacy, NPS drivers, and qualitative feedback. What do users say when they describe your product to others? That’s your emotional signal.</p><p>The best products of the next decade won’t just be fast or smart — they’ll be the ones that leave people feeling something worth returning to.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*MGXXA39Cd_OC8-Yk" /><figcaption>Photo by <a href="https://unsplash.com/@googledeepmind?utm_source=medium&amp;utm_medium=referral">Google DeepMind</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h4>The opportunity ahead</h4><p>In a world where AI can build anything, it’s easy to think the work is done.</p><p>But what matters most won’t be what gets built. It’ll be what gets felt. The products that endure will be the ones that care about what lingers — not just what launches.</p><p>Design is how we create it that emotional residue. It’s how we signal intent, earn trust, and make technology feel human, even when humans aren’t in the loop.</p><p>As agentic tools accelerate execution, the opportunity isn’t to do more.<br><strong>It’s to go deeper.</strong></p><p>To use the time we save not to ship faster, but to ship better. To move beyond features, and design for the feeling that remains after the feature is done.</p><p>Because the future of the product won’t be defined by speed, scale, or specs.</p><p>It will be defined by the quiet, human moments our products leave behind — and the teams who cared enough to create them.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b1f5a26ef6bc" width="1" height="1" alt=""><hr><p><a href="https://uxdesign.cc/designing-for-emotional-residue-over-functional-outcomes-b1f5a26ef6bc">Designing for emotional residue over functional outcomes</a> was originally published in <a href="https://uxdesign.cc/">UX Collective</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Generative UI: The End of the One-Size-Fits-All Design]]></title>
            <link>https://peter-barber.medium.com/generative-ui-the-end-of-the-one-size-fits-all-design-5fc54a33e3a6?source=rss-292ca2ed8bb2------2</link>
            <guid isPermaLink="false">https://medium.com/p/5fc54a33e3a6</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[research]]></category>
            <category><![CDATA[design]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[product-management]]></category>
            <dc:creator><![CDATA[Pete Barber]]></dc:creator>
            <pubDate>Mon, 24 Mar 2025 03:42:20 GMT</pubDate>
            <atom:updated>2025-03-24T03:42:20.264Z</atom:updated>
            <content:encoded><![CDATA[<p><em>How behaviour-driven AI is revolutionising product design, making personas obsolete.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*suXGqFyPeQb93rm6" /><figcaption>Photo by <a href="https://unsplash.com/@markusspiske?utm_source=medium&amp;utm_medium=referral">Markus Spiske</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>We’ve accepted for far too long that UI design must target averages. Personas, archetypes, customer segments, they’ve been our tools, and we’ve made the best of them. But no matter how detailed, no persona can capture the rich complexity of human behaviour. No user has ever been truly average. It’s time our products stop pretending otherwise. The uncomfortable truth is: <strong>no user is average, and we’re selling our customers short.</strong></p><blockquote><strong><em>“No user has ever been truly average. It’s time our products stop pretending otherwise.</em>”</strong></blockquote><p>Generative UI challenges this entire premise. Imagine interfaces that continuously learn from your past behaviours, preferences, and interactions, recreating themselves dynamically, uniquely tailored to you. <strong>No rigid flows, no generalised layouts.</strong> Instead, every interaction becomes optimised for your exact needs, abilities, and <strong>desired outcomes.</strong></p><p>Consider this: you’re filling out a notoriously cumbersome insurance form online. Generative UI senses your hesitation, <strong>immediately pivoting from a complex form into a conversational AI guide that walks you step-by-step.</strong> But your colleague, who despises chatbots and prefers simplicity, sees a minimalist, streamlined form instead. The same product effortlessly adapts in real-time, <strong>serving wildly different experiences that are equally optimal.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*75xMOKDh4Rbl0jq4" /><figcaption>Photo by <a href="https://unsplash.com/@vjgalaxy?utm_source=medium&amp;utm_medium=referral">Alejandro Piñero Amerio</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>In my career designing <strong>high-risk and complex customer journeys</strong>, I’ve witnessed repeatedly how rigid design systems fail precisely because they <strong>can’t account for the nuances of individual behaviour</strong>. Accessibility, usability, and delight become infinitely achievable when UI becomes generative, shaped by the user,<strong> not predefined constraints.</strong></p><p>Today’s “personalised” experiences barely scratch the surface, often <strong>limited to superficial settings or static preferences.</strong> Generative UI isn’t an incremental improvement, it’s revolutionary. It adapts at the pace of human need, powered by AI-driven insights that <strong>continuously reshape the user’s journey.</strong></p><p>Studies indicate that <a href="https://www.sender.net/blog/personalization-statistics/"><strong>74% of customers feel frustrated when website content is not personalised</strong></a>, signalling a clear demand for something better.</p><p>This shift isn’t just inevitable, it’s essential. Designers and product leaders have an opportunity, a responsibility, to build products that genuinely serve each individual user. The future is generative, adaptive, and profoundly human.</p><p>It’s time to abandon the one-size-fits-all mindset and embrace generative UI as <strong>the new standard for design.</strong></p><p>How would your approach to design change if every user could have a unique, AI-generated interface? I’d love to hear your thoughts in the comments below.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5fc54a33e3a6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Congratulations, You’re Now an Endpoint]]></title>
            <link>https://peter-barber.medium.com/congratulations-youre-now-an-endpoint-3c783caec522?source=rss-292ca2ed8bb2------2</link>
            <guid isPermaLink="false">https://medium.com/p/3c783caec522</guid>
            <category><![CDATA[design]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[ux]]></category>
            <dc:creator><![CDATA[Pete Barber]]></dc:creator>
            <pubDate>Thu, 06 Mar 2025 03:38:54 GMT</pubDate>
            <atom:updated>2025-03-06T03:47:42.194Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*c63vSOBRTHpApQAd" /><figcaption>Photo by <a href="https://unsplash.com/@markusspiske?utm_source=medium&amp;utm_medium=referral">Markus Spiske</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>You are now an endpoint.</p><p>Every time you prompt an AI chat interface, you’re making the purest possible API call. No buttons, no toggles, no dropdowns. Just a direct request, structured in natural language, processed by a system that interprets, executes, and responds. The interaction layer has been flattened to its absolute minimum: you, the human, are now an API client.</p><p>And it works. It’s fast. It’s intuitive. People don’t need documentation to ask for what they want. But this shift exposes a hard truth: interaction design is no longer about enabling features. It’s about surfacing outcomes.</p><h3>Interaction Design Is System Design</h3><p>For decades, interaction design has been about mediating complexity. Buttons, menus, modals — these weren’t just conveniences. They were scaffolding, shaping mental models so users could navigate the complexity of underlying systems. But today, systems like ChatGPT collapse that scaffolding. The interface has become language itself, and suddenly, the complexity is hidden behind a single text box.</p><p>That’s a fundamental shift for UX. When the interface disappears, how do you shape user expectations? How do you guide behavior? What does “<strong>good design</strong>” mean in a world where users don’t browse, but simply ask?</p><p>The answer is <strong>not</strong> a return to traditional UI. The answer is <strong>designing for outcomes.</strong></p><h3>Designing for Outcomes, Not Features</h3><p>Most software interfaces today are still built around feature sets. You want to analyze data? Here’s a dashboard. You want to organize tasks? Here’s a list, a calendar, some filters. But when AI interfaces reduce user interaction to a single request-response loop, those paradigms start to break down. Users aren’t looking for a way to interact with a system; they’re looking for an <strong>outcome.</strong></p><p>That means the design challenge isn’t about presenting functionality — it’s about delivering the right result, as fast as possible, with the least amount of friction. It means shifting from “How do we expose this feature?” to “How do we guarantee this outcome?”</p><h3>What This Means for the Future of UI</h3><ul><li><strong>Interfaces will shift from invisible to dynamic visual storytelling.</strong> If the fastest way to get an answer is a prompt, traditional UIs won’t disappear — they’ll evolve. The best interfaces won’t just be minimal; they’ll be cinematic, immersive, and designed to <strong>show</strong> the system’s thinking in real-time. Think Iron Man’s heads-up display — data appearing, shifting, and responding at light-speed. This is the future of interaction.</li><li><strong>Systems must be predictive and anticipatory.</strong> Users won’t just expect a response — they’ll expect the system to <strong>know</strong> what’s needed before they ask. AI-driven interfaces will proactively surface the right data, reducing the gap between intent and action.</li><li><strong>Personalization will define the next era of UX.</strong> Interaction design will shift from static layouts to <strong>bespokely dynamic</strong> experiences — interfaces that tailor themselves to the user’s context, behavior, and goals. Static design paradigms are dead; the future is real-time, adaptive, and deeply personalized.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*YCQH7DrrWyxVl1Dw" /><figcaption>Photo by <a href="https://unsplash.com/@sortino?utm_source=medium&amp;utm_medium=referral">Joshua Sortino</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h3>Who’s Doing This Today?</h3><p>This isn’t just theory. We’re already seeing early glimpses of this future:</p><ul><li><strong>Apple’s Vision Pro</strong> rethinks spatial computing, seamlessly blending digital elements with real-world interaction. It anticipates user intent rather than waiting for input.</li><li><strong>Tesla’s Auto UI</strong> dynamically adjusts the interface based on driving conditions, presenting only the most relevant controls at any given moment.</li><li><strong>OpenAI’s GPT Agents</strong> are moving toward proactive AI-driven workflows, assembling responses dynamically instead of waiting for explicit user commands.</li><li><strong>Google’s Gemini and DeepMind’s Adaptive UI research</strong> focus on interfaces that evolve based on context, reducing cognitive load and friction.</li></ul><p>These are just the first steps. The companies that figure out how to create <strong>predictive, dynamic, and outcome-driven interfaces</strong> will own the future of human-computer interaction.</p><h3>What Could Go Wrong?</h3><p>Of course, not everyone is on board with this shift. The critics raise valid concerns:</p><ul><li><strong>What happens when AI gets it wrong?</strong> If we remove traditional UI elements, how do users course-correct when the system misinterprets intent?</li><li><strong>Loss of user control.</strong> The more predictive and anticipatory a system becomes, the harder it is for users to feel in control of their experience.</li><li><strong>Trust and transparency.</strong> If a system is dynamically making decisions, how do we ensure users understand why they’re seeing certain results?</li></ul><p>These are design challenges, not deal-breakers. The best interfaces won’t just be predictive — they’ll be explainable, transparent, and give users the right level of control when needed.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*jHL1fkWU9n1Sfj6U" /><figcaption>Photo by <a href="https://unsplash.com/@halacious?utm_source=medium&amp;utm_medium=referral">Hal Gatewood</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h3>How to Design for Outcomes</h3><p>Building interfaces for this new paradigm requires a shift in approach. Instead of designing screens, think about designing <strong>flows of intelligence</strong> — systems that interpret, anticipate, and act.</p><p>Here’s how to get started:</p><ul><li><strong>Shift from presenting data to surfacing insights.</strong> Instead of displaying everything and forcing users to analyze, design systems that extract what’s important and highlight it dynamically.</li><li><strong>Design for adaptability.</strong> Static layouts and fixed workflows won’t cut it. Interfaces should morph based on user context, intent, and urgency.</li><li><strong>Make system thinking visible.</strong> When AI drives an interface, users need to understand why a decision was made. Use visual storytelling, subtle animations, and explanations to build trust.</li><li><strong>Prototype dynamic experiences, not static screens.</strong> Stop thinking in wireframes — think in flows, inputs, and predictive responses. Use AI-driven prototypes to simulate real-time interaction.</li><li><strong>Rethink success metrics.</strong> If your UX team measures success in terms of clicks or time spent on a page, you’re optimizing for the wrong thing. Measure how quickly and accurately users achieve their goals.</li></ul><p>The future belongs to those who stop thinking about software as a set of features and start designing for results. The best interfaces aren’t the ones users love interacting with.</p><p>The best interfaces are the ones users don’t even realize they’re using.</p><p>We’re all now endpoints. <strong>So design for it</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3c783caec522" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Last JTBD Resource You’ll Ever Need]]></title>
            <link>https://peter-barber.medium.com/the-last-jtbd-resource-youll-ever-need-f875d8c7a086?source=rss-292ca2ed8bb2------2</link>
            <guid isPermaLink="false">https://medium.com/p/f875d8c7a086</guid>
            <category><![CDATA[product-design]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[jtbd]]></category>
            <category><![CDATA[workshop-facilitation]]></category>
            <category><![CDATA[product]]></category>
            <dc:creator><![CDATA[Pete Barber]]></dc:creator>
            <pubDate>Thu, 27 Feb 2025 03:20:21 GMT</pubDate>
            <atom:updated>2025-02-27T03:37:33.516Z</atom:updated>
            <content:encoded><![CDATA[<p>JTBD is broken. Here’s how we fix it.</p><p>Most product teams think they’re solving the right problems until they launch — and nobody cares.</p><p>Think of a top product org — now think of their latest AI feature that you never use.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/0*Hm2THoHEoqd8Srp1.gif" /><figcaption>Is Meta AI worth $65b? 👀</figcaption></figure><p>The Jobs to Be Done (JTBD) framework was supposed to fix this.</p><p>It promised a way to understand what users need. But in practice? It’s often too abstract to drive real decisions. It’s a great buzzword in workshops, yet somehow, teams still end up straying from the framework, leading to misplaced investments in sub-par customer solutions.</p><p>That’s why I built the Lean JTBD Canvas — a practical tool to align teams across functions, bridge the gap between theory and execution, and make sure designers aren’t just crafting beautiful interfaces but driving real product impact.</p><h4>The Missing Link: JTBD Meets DVF</h4><p><strong>JTBD is powerful — but most teams don’t know how to apply it.</strong> It’s great for understanding user intent, but in practice, teams still default to feature-building, misaligned priorities, and execution disconnected from real outcomes.</p><p>That’s why I built the <strong>Lean JTBD Canvas</strong> — a <strong>practical framework</strong> that brings together <strong>JTBD + DVF</strong> to ensure every product decision passes three critical tests:<br>✅ <strong>Desirable</strong> → Are we solving a problem users truly care about?<br>✅ <strong>Viable</strong> → Does it drive sustainable business impact?<br>✅ <strong>Feasible</strong> → Can we build and scale it effectively?</p><p><strong>Without this balance, teams fail:</strong><br>🪤 Too much Desirability → You build something users love but can’t monetize.<br>🪤 Too much Viability → Business pressures drive useless features.<br>🪤 Too much Feasibility → Engineering constraints dictate the roadmap instead of user needs.</p><p>The Lean JTBD Canvas is designed to <strong>force alignment across teams</strong> — moving beyond theory to ensure teams don’t just brainstorm needs, but actually <strong>build the right things, the right way.</strong></p><h4>From Christensen to Ulwick</h4><p>Jobs to Be Done (JTBD) changed how we think about customer needs. <strong>Clayton Christensen</strong> reframed product adoption — not as a demographic choice, but as customers <em>hiring</em> a product to make progress. Anthony Ulwick took this further with <strong>Outcome-Driven Innovation (ODI)</strong>, adding rigor to JTBD by quantifying unmet needs and prioritizing solutions based on real user struggles.</p><p>The key shift?</p><blockquote><strong>Customers don’t buy products — they hire them to solve problems.</strong></blockquote><p>But while JTBD clarifies <em>why</em> users make decisions, it doesn’t tell teams <em>what</em> to build or <em>how</em> to prioritize. That’s why integrating <strong>Desirability, Viability, and Feasibility (DVF)</strong> is critical — it transforms JTBD from theory into execution.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FVpzUor6ujnEPhdcPf5J2A.png" /><figcaption>Put your snorkel on, because we’re about to dive in.</figcaption></figure><h4>How to Use the Lean JTBD Canvas: A Step-by-Step Guide</h4><p>The <strong>Lean JTBD Canvas</strong> isn’t just a design tool — it’s a <strong>framework for cross-functional alignment</strong>. The best product decisions don’t come from a single discipline; they emerge when <strong>design, product, engineering, and other key stakeholders collaborate from the start</strong> to ensure every decision is <strong>desirable, viable, and feasible</strong>.</p><p>This canvas provides <strong>a structured way to move teams through problem discovery and validation</strong>, ensuring solutions are deeply rooted in <strong>real user needs</strong>, while staying grounded in <strong>business goals and technical realities</strong>.</p><h4>How It Works in a Cross-Functional Team</h4><p><strong>Design &amp; Product should co-lead the session.</strong></p><ul><li><strong>Design drives problem clarity</strong> → preventing premature solutionizing and keeping focus on user needs.</li><li><strong>Product ensures alignment with strategy &amp; viability</strong> → prioritizing impact and keeping execution tied to business goals.</li></ul><p><strong>Real alignment requires involving your entire cross-functional team</strong> — not just at the end, but from the very start.</p><ul><li><strong>Engineering</strong> brings feasibility into the conversation early, helping teams avoid unrealistic solutions.</li><li><strong>Research</strong> ensures the process is grounded in evidence, not just assumptions.</li><li><strong>Marketing, Sales, and Support</strong> provide critical insights from customer interactions that can uncover hidden needs.</li></ul><p>This balance <strong>prevents JTBD from becoming a wishlist of user needs detached from business realities</strong> or, conversely, a <strong>purely business-driven exercise that overlooks what matters to users</strong>.</p><p>When used effectively, the <strong>Lean JTBD Canvas acts as a forcing function for alignment</strong> — helping teams <strong>uncover insights, challenge assumptions, and make strategic product decisions</strong> that affect impact.</p><h3>1. Job Performer: Who Do You Want to Innovate For?</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rRuSI_NS1VMUkGrM4WzmBQ.png" /></figure><h4>Why This Step Matters</h4><p>Before identifying jobs, you need to <strong>define who is hiring your product.</strong> Rather than traditional personas, the <strong>Job Performer</strong> focuses on <strong>behavioral archetypes</strong> — real patterns of user behavior rather than demographic segments.</p><p>A well-defined Job Performer provides <strong>clarity on context</strong> and ensures the team aligns on <strong>who they are designing for</strong> before diving into problem-solving.</p><h4>How to Write an Effective Job Performer</h4><p>Focus on behaviors, not job titles (e.g., <em>“Busy parents ordering dinner after work,”</em> <strong>not</strong> <em>“Millennials aged 25–40.”</em>).</p><p>Ground it in real-world patterns (e.g., <em>“People who regularly order takeout due to time constraints”</em> rather than <em>“People who like food.”</em>).</p><p>Keep it open-ended enough that it doesn’t box you into a narrow solution space.</p><h4>Example Post-Its from a Team</h4><ul><li>✅ <em>Office workers ordering lunch on a short break</em></li><li>✅ <em>Parents trying to feed their family after a long day</em></li><li>✅ <em>Late-night eaters craving meals outside standard restaurant hours</em></li></ul><h4>Common Mistakes</h4><ul><li>❌ Defining too broad an audience (e.g., <em>“Anyone who eats food.”</em>)</li><li>❌ Using demographic labels instead of behavioral archetypes (<em>“College students aged 18–24.”</em>)</li><li>❌ Assuming too much about their needs — keep it based on observed behaviors, not assumptions.</li></ul><h4>Facilitation Tips</h4><ul><li>💡 <strong>Ask teams:</strong> <em>“If this user disappeared, would this problem still exist?”</em> If yes, refine your archetype.</li><li>💡 Push teams to <strong>define real-world examples</strong> — if they struggle, their definition is likely too vague.</li></ul><h3>2. Desirability: Understanding the User’s Reality</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*oIXSXhPzupPLZSqlO_c25A.png" /><figcaption>Start with Pain points, then start working across</figcaption></figure><p>The <strong>top half of the canvas</strong> is designed to <strong>uncover user needs and frustrations through a JTBD lens</strong>. The goal is to surface real problems, motivations, and success criteria <strong>before jumping into solutions.</strong></p><h4>A. Pain Points — What’s Causing Friction?</h4><p>Before jumping into the job story, we first need to <strong>identify the core frustrations</strong> stopping users from making progress.</p><p>✅ <em>“I don’t have time to cook after work, but I don’t want to wait an hour for delivery.”<br></em>✅ <em>“I hate that I can’t track my order and have no idea when my food will arrive.”</em></p><p>💡 <strong>Facilitation Tip:</strong> Ask your team <strong>“Why is this a problem?” </strong>The trick here is to use as much validated research as possible while highlighting the obvious barriers that exist for your customer.</p><h4>B. Situation — When Does This Problem Occur?</h4><ul><li>Define the <strong>trigger moments</strong> when the pain point is most acute.</li><li>Provides <strong>context</strong> so teams don’t generalize the problem.</li><li>This forms the <strong>first part of the job story</strong>: <em>“When [Situation]…”</em></li></ul><p>✅ <em>“When I get home late from work and I’m too exhausted to cook.”<br></em>✅ <em>“When I have a short lunch break and need something fast.”<br></em>✅ <em>“When I order dinner, but my delivery estimate keeps changing.”</em></p><p>💡 <strong>Facilitation Tip:</strong> Push teams to be <strong>specific</strong> — a vague situation leads to a vague solution.</p><p>💡 <strong>Facilitation Tip:</strong> It’s okay if suggestions arn’t perfect or are made from an educated perspective. We’ll cover how to deal with this later in the canvas.</p><h4>C. Motivation — What is the User Trying to Achieve?</h4><ul><li>This is the <strong>core user intent</strong> behind their actions.</li><li>Helps differentiate <strong>surface-level feature requests</strong> from <strong>underlying jobs</strong>.</li><li>This forms the <strong>second part of the job story</strong>: <em>“…I want to [Motivation]…”</em></li></ul><p>✅ <em>“…I want to get a fresh meal delivered quickly without planning ahead.”<br></em>✅ <em>“…I want to know exactly when my food will arrive so I can time my break.”<br></em>✅ <em>“…I want to trust that my delivery estimate is accurate so I can plan around it.”</em></p><p>💡 <strong>Facilitation Tip:</strong> Ask <em>“Why is the user hiring a tool/process?”</em></p><h4>D. Expected Outcomes — What Does Success Look Like?</h4><ul><li>Defines <strong>how users measure success.</strong></li><li>Ensures teams optimize for <strong>user-defined success</strong> rather than internal metrics.</li><li>This forms the <strong>final part of the job story</strong>: <em>“…so I can [Expected Outcome].”</em></li></ul><p>✅ <em>“…so I can eat a proper meal without wasting time waiting or cooking.”<br></em>✅ <em>“…so I can make the most of my short break without rushing.”<br></em>✅ <em>“…so I can plan my evening without unexpected delays.”</em></p><p>💡 <strong>Facilitation Tip:</strong> If you get stuck, ask your team<em> “What do you think customers would write in a positive review of our product/service”</em></p><h4>The Magic: This Naturally Creates a Job Story</h4><p>When filled out correctly, these sections naturally <strong>form a job story</strong> following this structure:</p><p><strong>When [Situation], I want to [Motivation], so I can [Expected Outcome].</strong></p><blockquote><strong><em>[Situation]</em> — </strong>When I get home late from work and I’m too exhausted to cook</blockquote><blockquote><strong><em>[Motivation]</em> — </strong>I want to get a fresh meal delivered quickly without planning ahead</blockquote><blockquote><strong><em>[Expected Outcome]</em> — </strong>so I can eat a proper meal without wasting time waiting or cooking.</blockquote><h4>Why This Approach Works</h4><p>✅ <strong>It keeps teams focused on real user needs</strong> rather than feature brainstorming.<br>✅ <strong>It aligns stakeholders on a shared understanding</strong> before jumping into solutions.<br>✅ <strong>It scales across different industries</strong> — this method applies equally well to <strong>cybersecurity, healthcare, fintech, or any product category.</strong></p><p>💡 <strong>Facilitation Tip:</strong> At the end of this section, have teams <strong>read their job stories aloud</strong> — if they feel natural and compelling, they’re on the right track. If not, they likely need to refine their Situation, Motivation, or Expected Outcome.</p><h3>3. Next Step: Synthesizing the Core JTBD</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RbbCvXmno9zdJkXERQ-lKA.png" /><figcaption>After documenting the world from our customer’s POV, dial in on why we’re here</figcaption></figure><p>Once these sections are complete, teams can <strong>distill the Core JTBD</strong> from their job stories. The goal is to capture the <strong>essence of what users are hiring the product for.</strong></p><p>✅ <strong>“Help busy people get quick, reliable meals with minimal effort.”<br></strong>✅ <strong>“Make food delivery predictable and seamless for customers with tight schedules.”</strong></p><p>You’ll often find there’ll be multiple related jobs that your teams form during this stage. Feel free to document them all, but try and nail the highest level progress job you can.</p><p>By following this structured approach, teams <strong>arrive at JTBD insights naturally, rather than forcing them.</strong></p><h3>4. Viability &amp; Feasibility: Aligning with Business and Technical Realities</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MdVThcY4hq89FSv_NfDFPg.png" /></figure><p>Once the <strong>Desirability</strong> section is complete, the team shifts focus to <strong>Viability</strong> (business alignment) and <strong>Feasibility</strong> (technical constraints). This ensures teams don’t just build things that users want but that also <strong>make sense for the business and are realistically achievable.</strong></p><p>At this stage, teams move from <strong>understanding the problem space</strong> to <strong>defining measurable success</strong> while setting clear boundaries to maintain focus.</p><h3>A. Goals — What Are We Aiming For?</h3><h4>Why This Step Matters</h4><p>Now that we understand user needs, we need to <strong>define what success looks like for the business</strong>. The key here is to <strong>ground objectives in tangible impact</strong> rather than vague aspirations. This step ensures teams don’t optimize for user desirability in isolation but align their work with company strategy.</p><h4>How to Write an Effective Goal</h4><ul><li>Make it <strong>outcome-driven</strong>, not output-driven.</li><li>Ensure it <strong>balances user and business impact</strong> (user wins = business wins).</li><li>It should <strong>feel ambitious but achievable</strong> within the scope of the project.</li></ul><p>✅ <strong>“Reduce delivery wait times”<br></strong>✅ <strong>“Increase repeat orders”<br></strong>✅ <strong>“Improve estimated delivery accuracy”</strong></p><p>💡 <strong>Facilitation Tip:</strong> Ask: <em>“If we achieve this, would it meaningfully improve both the user experience and business metrics?”</em> If not, refine it.</p><h3>B. Success Metrics — How Do We Know We’ve Achieved the Goal?</h3><p>This step <strong>turns goals into measurable OKRs</strong>, ensuring the team has a <strong>clear definition of success</strong>. A good success metric should:</p><ul><li><strong>Be quantifiable</strong> (no vague “improve experience” statements).</li><li><strong>Have a clear timeframe</strong> for evaluation.</li><li><strong>Tie back to user impact and business viability.</strong></li></ul><p><strong>Objective:</strong> Reduce delivery wait times<br>✅ <strong>Key Result 1:</strong> 80% of orders delivered within 20 minutes for high-density areas.<br>✅ <strong>Key Result 2:</strong> Customer satisfaction scores remain above 4.5/5 despite speed increase.</p><p><strong>Objective:</strong> Improve estimated delivery accuracy.<br>✅ <strong>Key Result 1:</strong> 90% of deliveries arrive within ±5 minutes of predicted time.<br>✅ <strong>Key Result 2:</strong> Reduce customer complaints related to late orders by 40%.</p><p>💡 <strong>Facilitation Tip:</strong> If a success metric isn’t <strong>verifiable within a reasonable timeframe</strong>, push the team to refine it.</p><h3>C. Non-Goals — What’s Out of Scope?</h3><h4>Why This Step Matters</h4><p>To stay <strong>laser-focused</strong>, teams need to <strong>explicitly define what they are NOT doing</strong>. This helps prevent scope creep, unrealistic expectations, and misalignment.</p><p>A strong <strong>Non-Goal</strong> should:</p><ul><li><strong>Clarify boundaries</strong> so teams don’t waste cycles on unnecessary discussions.</li><li><strong>Prevent distraction</strong> by aligning stakeholders on what will NOT be tackled.</li><li><strong>Avoid adding “nice-to-haves” that dilute impact.</strong></li></ul><p>✅ <em>“We are NOT expanding restaurant partnerships in this phase — focus is on delivery optimization.”<br></em>✅ <em>“We are NOT launching a new loyalty program — focus is on repeat orders via better UX.”<br></em>✅ <em>“We are NOT building a new dispatch system — focus is improving existing routing algorithms.</em></p><p>💡 <strong>Facilitation Tip:</strong> Ask: <em>“If a stakeholder asked us to add this, would it completely derail focus?”</em> If yes, it belongs here.</p><h3>5. Constraints, Risks &amp; Assumptions — Challenging What We Think We Know</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NPjYzyPmFm1emb9GgtUJxg.png" /></figure><h4>Why This Step Matters</h4><p>This is the moment where teams <strong>stress test everything they’ve defined so far</strong>. The goal is to surface <strong>hidden assumptions, risks, and gaps</strong> that could derail execution.</p><p>At this stage, teams should ask:<br>🛑 <strong>“What assumptions have we made that, if wrong, would break everything?”<br></strong>⚠️ <strong>“What are the biggest risks to success?”<br></strong>🔍 <strong>“What do we still need to validate?”</strong></p><p><strong>Now is the time to challenge everything.</strong></p><h3>Why are these tools more important than ever?</h3><p><strong>Let’s not sugarcoat it — designers are on borrowed time.</strong></p><p>AI tools are already generating UI layouts, writing UX copy, and even running usability tests. And you can bet that somewhere, right now, an executive is looking at design as a “nice-to-have” and sharpening the budget-cutting axe.</p><p>The <strong>only way to stay relevant</strong> is to prove — right now — that design isn’t just about aesthetics. It’s about <strong>making sure teams don’t waste millions building the wrong shit.</strong> It’s about <strong>identifying the real problems AI products should be solving before another garbage solution gets pushed to market.</strong></p><p><strong>Designers can be the ones to stop the cycle.</strong></p><p>We know how to dig deep into <strong>human motivation, context, and decision-making.</strong> We know how to <strong>translate hype into actual user value.</strong></p><p>The best product teams don’t just execute — they own the problem space.</p><p><strong>Be one of them.</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f875d8c7a086" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Security is Failing Because We’re Designing for Data, Not for Humans]]></title>
            <link>https://peter-barber.medium.com/security-is-failing-because-were-designing-for-data-not-for-humans-7fb4d547039d?source=rss-292ca2ed8bb2------2</link>
            <guid isPermaLink="false">https://medium.com/p/7fb4d547039d</guid>
            <category><![CDATA[security]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[product-design]]></category>
            <category><![CDATA[product-management]]></category>
            <dc:creator><![CDATA[Pete Barber]]></dc:creator>
            <pubDate>Wed, 19 Feb 2025 04:00:09 GMT</pubDate>
            <atom:updated>2025-02-19T04:00:09.181Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*0lDimqsfSf-_739F" /><figcaption>Photo by <a href="https://unsplash.com/@sortino?utm_source=medium&amp;utm_medium=referral">Joshua Sortino</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h3>The Problem No One Wants to Admit</h3><p>For years, cybersecurity has been obsessed with <strong>seeing more</strong> — more vulnerabilities, more logs, more alerts. We’ve been sold the idea that <strong>more visibility means more security.</strong> If we just collect enough data, security teams will be able to protect systems more effectively.</p><p>But that’s not what’s happening.</p><p>Instead of <strong>clarity</strong>, we’ve created <strong>overload</strong>. Instead of <strong>helping defenders</strong>, we’ve built tools that <strong>bury them under a flood of unprioritized, unactionable information</strong>. The promise of “full visibility” has turned into an <strong>endless firehose of risk</strong>, with no intelligent way to separate the meaningful from the meaningless.</p><p>And here’s the kicker — when security tools <strong>aren’t designed for humans, people work around them.</strong></p><ul><li>When security products introduce friction, <strong>users bypass them</strong>.</li><li>When dashboards flood teams with thousands of issues, <strong>critical risks get lost</strong>.</li><li>When security requires expertise just to operate, <strong>it creates gaps instead of closing them</strong>.</li></ul><p>Security isn’t just about <strong>technology</strong>. It’s about <strong>how people interact with that technology</strong> — and that’s where most products are failing.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*c0gm5KNROpsMy2gS" /><figcaption>Photo by <a href="https://unsplash.com/@edge2edgemedia?utm_source=medium&amp;utm_medium=referral">Edge2Edge Media</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h3>More Data, More Confusion, More Risk</h3><p>There’s a fundamental misunderstanding in the way we approach security today. We’ve been trained to think that <strong>more data equals more protection</strong>. But raw data, on its own, doesn’t create security — it creates <strong>decision paralysis</strong>.</p><p>Consider a security analyst logging into their dashboard at the start of the day. They don’t just see <strong>one or two</strong> critical vulnerabilities; they see <strong>hundreds, even thousands</strong>, all flagged as issues that require attention. Some of these are minor misconfigurations. Others are real threats. But without intelligent prioritization, <strong>everything looks urgent</strong>, and as a result, nothing actually gets done.</p><p>This is the visibility paradox: <strong>when everything is important, nothing is.</strong></p><p>Security teams aren’t struggling because they <strong>don’t have enough data</strong> — they’re struggling because they <strong>can’t act on it.</strong> And the problem compounds when tools are overly complex, requiring deep expertise to interpret and operate.</p><p>Take MFA (multi-factor authentication), for example. Security teams roll it out to protect against account takeovers, but when it’s designed <strong>in a frustrating or intrusive way</strong>, users get fatigued. Attackers have learned to <strong>weaponize that fatigue</strong>, bombarding users with MFA requests until they finally accept one out of frustration. The security measure meant to <strong>protect</strong> them ends up being the very thing attackers <strong>exploit</strong> — not because MFA is flawed, but because its <strong>implementation ignored human behavior.</strong></p><p>The lesson? <strong>If security doesn’t account for usability, it backfires.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*1MxVOYnY0HxWo4YR" /><figcaption>Photo by <a href="https://unsplash.com/@christopher__burns?utm_source=medium&amp;utm_medium=referral">Christopher Burns</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h3>The Future of Security: From Visibility to Intelligence</h3><p>Security tools need to evolve beyond <strong>just surfacing more data</strong>. The future of security isn’t about collecting more alerts, logs, and threat feeds — it’s about making <strong>sense of them in real time</strong> and <strong>enabling action instead of creating paralysis.</strong></p><p>The shift from <strong>data overload to decision intelligence</strong> starts with three key principles:</p><h4>1. Prioritize What Matters</h4><p>Not all vulnerabilities are created equal. A security system that treats a <strong>low-severity misconfiguration</strong> the same way it treats an <strong>actively exploited vulnerability</strong> does nothing but <strong>waste time and resources</strong>.</p><p>Attackers are <strong>as strategic as they are opportunistic</strong> — they look for the weakest, easiest-to-exploit entry points first. Security tools need to do the same. Instead of listing every possible issue, systems should <strong>automatically rank threats based on exploitability, real-world risk, and potential business impact.</strong></p><p>Consider two scenarios: one where a security team is given <strong>a list of 5,000 potential issues</strong> with no clear prioritization, and another where they’re given <strong>the 10 most urgent threats that, if exploited, could lead to a breach.</strong> One approach <strong>overwhelms</strong> them; the other <strong>enables</strong> them to act effectively.</p><h4>2. Make Secure Behavior Effortless</h4><p>Security has a design problem. When security tools introduce too many steps, require extensive training, or slow down workflows, users <strong>find ways to work around them</strong>. And when security becomes an inconvenience, <strong>people default to whatever is easiest</strong> — even if that means bypassing critical protections.</p><p>Think about how modern password managers work. Instead of forcing users to remember <strong>dozens of complex passwords</strong>, they <strong>store, autofill, and even generate strong credentials</strong> on demand. <strong>The best security is the kind that doesn’t feel like security at all — it just works.</strong></p><p>This principle applies across the board. Security tools should <strong>integrate seamlessly into workflows</strong>, guiding users toward safer behavior <strong>without them needing to think about it.</strong> If a security feature <strong>requires constant explanation, onboarding, or manual effort</strong>, it’s a sign that something is wrong.</p><h4>3. Automate the Grunt Work</h4><p>The reality is that security teams are <strong>outnumbered and outpaced</strong>. The sheer <strong>volume of threats</strong>, vulnerabilities, and compliance requirements makes it impossible for human analysts to manually keep up.</p><p>This is where <strong>intelligent automation</strong> comes into play. Instead of expecting security professionals to <strong>scan through thousands of logs and alerts</strong>, security tools should be <strong>automating correlation, filtering out noise, and surfacing actionable insights.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*f_JBMCbAZOhaI6S1" /><figcaption>Photo by <a href="https://unsplash.com/@jdent?utm_source=medium&amp;utm_medium=referral">Jason Dent</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h3>AI is Changing the Definition of Safe</h3><p>And here’s where things get even more complex: AI isn’t just <strong>changing security</strong> — it’s <strong>changing the very definition of what “safe” even means.</strong></p><p>AI-powered systems are already <strong>finding and exploiting vulnerabilities at scale</strong> — sometimes faster than humans can patch them. <strong>Autonomous cyber threats are emerging</strong>, with attackers leveraging AI to <strong>map attack surfaces, identify weak points, and execute attacks at machine speed.</strong></p><p>If security remains <strong>a visibility problem</strong>, we’re walking into a future where <strong>AI doesn’t just expose our weaknesses — it exploits them faster than we can react.</strong></p><p>The future of security isn’t just about <strong>seeing more</strong> — it’s about <strong>understanding faster and acting smarter.</strong></p><p>Security isn’t just a technical challenge — it’s a <strong>human challenge.</strong> And in an era where AI is reshaping the battlefield, we can’t afford to keep designing security for <strong>machines to process</strong> instead of <strong>humans to use.</strong></p><p>The only question is: <strong>Are we ready to build security for the world that’s coming next?</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7fb4d547039d" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>