<?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 Diba on Medium]]></title>
        <description><![CDATA[Stories by Diba on Medium]]></description>
        <link>https://medium.com/@dibakaya?source=rss-8b49ffd8159e------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*T16j_8btdb7LUuJhd9IR9Q.jpeg</url>
            <title>Stories by Diba on Medium</title>
            <link>https://medium.com/@dibakaya?source=rss-8b49ffd8159e------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 12:28:00 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@dibakaya/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[Solving the Hidden Hurdles of AI Product Development]]></title>
            <link>https://medium.com/@dibakaya/solving-the-hidden-hurdles-of-ai-product-development-852f6df14e54?source=rss-8b49ffd8159e------2</link>
            <guid isPermaLink="false">https://medium.com/p/852f6df14e54</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[venture-capital]]></category>
            <category><![CDATA[human-centered-design]]></category>
            <dc:creator><![CDATA[Diba]]></dc:creator>
            <pubDate>Fri, 22 Aug 2025 22:50:48 GMT</pubDate>
            <atom:updated>2025-08-25T23:53:05.527Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/626/1*CKIROfjYUo2KmSa9-bjUZg.png" /></figure><p>A founder said to me recently: <em>“We need someone who can tell our designers and engineers, ‘hey — you’re ignoring this aspect of humanity.’”</em></p><p>It wasn’t a request for a faster model or a smarter pipeline. It was an ask for a bridge between what AI can do and what humans actually need and how we function.</p><p>That moment stuck with me, because it surfaces the real challenge in AI product development: the hardest problems aren’t technical. They’re human. (But isn’t that always the “hard” problem? 😉)</p><p>Here’s why it matters: we’re in an age where AI isn’t just another tool. It’s becoming an <strong>extension of cognition itself.</strong> AI shapes us, and we shape it and every product decision carries two consequences: a <strong>business consequence</strong> (growth, adoption, churn, valuation), and a <strong>human consequence</strong> (agency, trust, skill, identity).</p><p>Both matter. If you only solve for the business side, you risk creating brittle products that don’t last. If you only solve for the human side, you risk losing the users, investors and markets that keep you alive.</p><p>Most of the AI discourse today fixates on technical breakthroughs (…or AGI, but don’t get me started): faster pipelines, better models, smarter algorithms. But sit down with founders in the trenches, and you’ll hear a different story. The toughest problems show up when you’re forced to confront how people actually think, learn, and make meaning.</p><p>That’s what this piece is about. Five hidden hurdles in AI product development, what happens when you don’t solve them, and how to start moving in the right direction.</p><h3>1. Who Does What: AI vs. Human?</h3><p><strong>Problem:</strong> Which tasks should AI automate vs. leave to humans?</p><p><strong>How it shows up: </strong>A founder/PM knows their product needs AI, but they can’t quite define <em>where</em> it fits. Should the AI handle the entire workflow or just pieces? Should it step in early (guidance) or late (execution)? The line between “low-level” and “high-level” work feels blurry, and every decision risks either overwhelming or infantilizing the user.</p><p><strong>Example: </strong>A writing assistant that drafts entire essays for students. For a novice, this creates dependency (eek) students get “answers” but never develop the skill of writing, but more important what the skill of writing offer — putting together an argument. However, for an expert, the same feature might be liberating by saving them time on structure so they can focus on ideas and flow.</p><p><strong>Why it matters:</strong> Automating the <strong>wrong level </strong>of task either overwhelms users with unnecessary work or robs them of growth.</p><p><strong>Human consequence: </strong>Misplaced automation <strong>erodes self-efficacy</strong>; novices feel incompetent, experts feel undervalued.</p><ul><li>Behavior: Novices <strong>avoid using the product without the AI</strong> → engagement drops outside narrow use cases. Experts hack around the product, disable features, or leave → churn among your power users.</li></ul><p><strong>Business fallout:</strong> Instead of a broad, sticky user base, you get <strong>fragmented adoption</strong> and product-market misfit</p><h3>2. If Users Don’t Understand, They Won’t Stay</h3><p><strong>Problem:</strong> AI that “gets the job done” but leaves users disconnected from the <em>why</em>.</p><p><strong>How it shows up: </strong>Founders/PMs celebrate that their AI is faster or more accurate, teams celebrate metrics like speed but then hear feedback like, <em>“I don’t feel like I understand what it’s doing”</em> or <em>“I can’t explain this to my team.”</em> The product works, but trust and adoption stall because people don’t feel oriented.</p><p><strong>Example: </strong>financial planning app that recommends investment strategies without explaining the logic. Users may click “accept” for efficiency but don’t actually learn financial literacy or build confidence in their decisions.</p><p><strong>Why it matters: </strong>Understanding is the foundation of <strong>trust.</strong> Without it, users may comply but <strong>won’t commit.</strong></p><p><strong>Human consequence:</strong> Users operate in <strong>cognitive dissonance</strong> — they accept outputs they can’t justify, undermining their sense of coherence.</p><ul><li>Behavior: Users <strong>stop recommending your tool</strong> (“I can’t explain how it works, so I can’t stake my reputation on it”) and teams rely on it for low-stakes work only → <strong>low expansion into enterprise-critical workflows.</strong></li></ul><p><strong>Business fallout:<em> </em></strong>You get <strong>trial and novelty use,</strong> but the product stalls in growth. <strong>Adoption plateaus </strong>before hitting meaningful revenue.</p><h3>3. Ethical Use of Real vs. Synthetic Users</h3><p><strong>Problem:</strong> Synthetic shortcuts flatten nuance, while real-world testing takes time.</p><p><strong>How it shows up:</strong> Teams under pressure lean on “synthetic users” to simulate research or feedback. It feels efficient, generate 100 interviews instantly! but what you get is statistical mimicry, not lived perspective.</p><p><strong>Example:</strong> A startup building a chat-based product tests prompts with synthetic users. The AI surfaces variations that look engaging on paper, but when real people try them, the language feels awkward, tone-deaf, or culturally off. What looked like “insight” was really just the echo of averages.</p><p><strong>Why it matters:</strong> Synthetic users <strong>can’t construct meaning</strong>. They don’t have bodies, memories, identities, or emotions. They simulate understanding without ever surfacing it.</p><p><strong>Human consequence:</strong> Users experience misrecognition, they feel unseen, stereotyped, or erased. Over time, this creates <strong>alienation and distrust,</strong> especially for marginalized groups.</p><ul><li><strong>Behavior:</strong> Real users <strong>bounce quickly</strong>, spread stories of “it doesn’t work for people like me,” and disengage from research or feedback loops.</li></ul><p><strong>Business fallout:</strong> You end up in a <strong>recursive trap</strong>: designing for a proxy of a proxy. Internal metrics look strong, but the product fails in the market. Missed product-market fit, wasted cycles, <strong>reputational risk,</strong> and stalled enterprise adoption.</p><h3>4. Long-Term Cognitive Impact</h3><p><strong>Problem:</strong> Teams stop at “does this feature work?” instead of asking, <em>“what does this feature do to the way people think over time?”</em></p><p><strong>How it shows up: </strong>Founders/PMs run short pilots or demos, and success looks like “users clicked the button” or “time to complete dropped.” Nobody asks: <em>what skill is being lost? what habit is being formed?</em> Months later, users rely on the product but can’t function well without it.</p><p><strong>Example:</strong> A project management tool that auto-generates all task breakdowns. At first it feels efficient. But over time, users lose the skill of breaking down complex work themselves, weakening problem-solving capacity.</p><p><strong>Why it matters:</strong> AI doesn’t just shape tasks, it reshapes minds. Cognitive offloading, if unchecked, leads to long-term skill erosion.</p><p><strong>Human consequence:</strong> Over-reliance leads to <strong>metacognitive atrophy, </strong>users lose awareness of when and how they’re thinking vs. the AI thinking for them.</p><ul><li><strong>Behavior: </strong>Users stop engaging critically → blind acceptance of outputs and errors compound unnoticed → <strong>trust collapses at scale when one visible failure occurs.</strong></li></ul><p><strong>Business fallout:</strong> You get a <strong>brittle adoption curve</strong>: fast early uptake, but catastrophic drop-off when failures surface because users have no resilience or skill buffer.</p><p><a href="https://medium.com/@dibakaya/second-order-thinking-the-missing-skill-in-ai-product-strategy-1e3ec52031d7">Check out my work on second-order thinking and AI product dev here</a></p><h3>5. Keeping AI Human-Centered</h3><p><strong>Problem:</strong> It’s easy to get dazzled by the tech and lose sight of the human.</p><p><strong>How it shows up:</strong> Founders/PMs describe their product in terms of model performance (“it’s 92% accurate”) but can’t articulate the lived value to a human being. The features look brilliant in demos but don’t connect to real goals, habits, or identities.</p><p><strong>Example: </strong>A language-learning app adds a flashy chatbot that generates perfect grammar drills. Users practice endlessly but never develop conversational confidence because the tool doesn’t simulate real human interaction.</p><p><strong>Why it matters:</strong> Products succeed when they meet people where they are, supporting not just performance but growth, trust, and identity.</p><p><strong>Human consequence:</strong> Users feel <strong>alienated, </strong>tools don’t reflect their goals, habits, or values.</p><ul><li><strong>Behavior: </strong>Users treat the product as a demo toy, not a real workflow tool. Engagement spikes at first use, then drops → <strong>classic “one-and-done” pattern.</strong></li></ul><p><strong>Business fallout:</strong><em> </em>High acquisition costs with <strong>low retention</strong>. Your product burns cash impressing investors but never builds a loyal base.</p><h3>So Who Needs to Care About These Hurdles?</h3><p>It’s tempting to think these are “design problems” or “research problems.” but they’re not. They cut across the table stakes of building, funding, and shipping AI products.</p><ul><li><strong>Founders</strong> need to decide how their product balances automation vs. augmentation, and how to tell a credible story to users and investors alike.</li><li><strong>Investors</strong> need to know if a product is defensible and durable, not just technically shiny, but sticky enough to expand and avoid churn.</li><li><strong>Product Managers</strong> need to translate all of this into real roadmaps: what features to ship, what to measure, and how to course-correct when the market pushes back.</li></ul><h3>That’s why I laid out the five core hurdles alongside concrete actions for each role:</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/934/1*nSjTTkXpgbTRsW8BjXdeSg.jpeg" /><figcaption>Solutions for AI Product Development Hurdles</figcaption></figure><h3>Pulling It All Together</h3><p>None of these hurdles are just design headaches or edge-case risks. They’re the center of the game in AI product development.</p><ul><li>Get task allocation wrong → you fragment your user base.</li><li>Skip user understanding → you cap your own growth.</li><li>Lean too hard on synthetic proxies → you lose market trust.</li><li>Ignore cognitive impact → you create brittle adoption.</li><li>Chase technical dazzle over human goals → you burn cash but don’t stick.</li></ul><p>Remember: <strong>AI isn’t just software. It’s cognitive infrastructure.</strong> It shapes how people think, how they work, and ultimately who they become. That means the <strong>teams who build with this in mind will have the advantage.</strong> Products that not only solve tasks, but also respect and expand human agency, are the ones people trust, recommend, and return to.</p><p>And that’s the real risk most teams miss: it’s not building too little. It’s building without anchoring in how humans actually think, learn, and grow.</p><p>When founders prototype automation levels with care, investors pressure-test durability beyond technical shine, and PMs translate both into real feature choices, you get alignment and it’s where defensible products come from.</p><h3>What’s Next</h3><p>This isn’t a checklist to tick once and forget. It’s a discipline: treating cognition, behavior, and human consequence as <strong>infrastructure</strong> in product strategy. If you’re a:</p><ul><li><strong>Founder:</strong> Ask yourself, “Am I clear on what <em>level</em> of task my AI should own, and where the human stays in the loop?”</li><li><strong>Investor:</strong> Push your portfolio to show how they’re protecting user trust and skill, not just improving model metrics.</li><li><strong>Product Manager:</strong> Run reflection checkpoints in your roadmap. Are users just clicking buttons, or are they learning, trusting, and staying</li></ul><p>If you’re not building for how humans think, you’re not building for the future. Because AI isn’t just shaping workflows, it’s shaping how we think, learn, and decide.</p><p>Happy building ❤</p><p><em>Curious how you’re handling these hurdles in your org — Drop a note I’d love to hear how teams are approaching it.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=852f6df14e54" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Second-Order Thinking Made Useful: A Product Strategy Framework for AI Builders]]></title>
            <link>https://medium.com/@dibakaya/second-order-thinking-the-missing-skill-in-ai-product-strategy-1e3ec52031d7?source=rss-8b49ffd8159e------2</link>
            <guid isPermaLink="false">https://medium.com/p/1e3ec52031d7</guid>
            <category><![CDATA[ai-product-design]]></category>
            <category><![CDATA[cognitive-science]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[product-design]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Diba]]></dc:creator>
            <pubDate>Wed, 23 Jul 2025 20:10:03 GMT</pubDate>
            <atom:updated>2025-08-22T23:02:40.334Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qvfVrUr2Pc24aVt8F-MPXA.png" /></figure><p>Earlier this week I was talking to an executive-in-residence at a VC firm that advises early-stage startups. We were trading notes on how to evaluate teams building AI features and guiding AI product development, and I found myself saying something I’ve been telling my PMs for awhile:</p><blockquote><strong><em>AI product strategy now requires a real understanding of cognition</em></strong><em> — something product development hasn’t seriously wrestled with until now. That makes understanding the context of use not just helpful, but essential.</em></blockquote><p>I explained concepts like context of use, cognitive offloading, and how generative AI shifts not just workflows, but the <em>shape</em> of thinking itself. He nodded, then asked:</p><blockquote><em>“Is there a framework for this something that helps you decide what AI features or products to build?”</em></blockquote><p>That question stuck with me because… right now, the answer is mostly <em>no.</em> As an in-house staff researcher, I work with my team to embed tribal knowledge about the context of use, mental models and the cognition we’re aiming to support (but not supplant). We have rituals: data jams, study sessions and research artifacts. But I couldn’t point to a single external framework that clearly guides how to <em>decide what to build</em> in this new AI terrain.</p><p>Frankly, what frameworks do exist are either buried in academia (often disconnected from the reality of creating products in industry) or overly technical, meant for engineers. Meanwhile, the pace of AI innovation has far outstripped our collective ability to create shared language or practical mental models to guide strategic product development in this area that keeps in touch with what we’re building for, the human and human cognition.</p><p><strong>The result? We’re building AI tools that reshape cognition without consciously mapping what’s being gained, what’s being lost, and who we’re becoming in the process.</strong></p><h3>Heres the thing, AI Is a Tool of Extended Cognition</h3><p>Every time we ask AI to summarize, autocomplete, or decide, we’re offloading a piece of thinking. That’s not inherently bad. In fact, it can be incredibly useful. (for instance, we don’t have to memorize phone numbers or calculate long division by hand, thank god.)</p><p>But as AI becomes embedded in how we read, write, learn, and decide, we have to be more intentional:<br> → <em>What kinds of thinking are we offloading?</em><br> → <em>When is that helpful, and when is it harmful?</em></p><p>Right now, most teams are focused on first-order wins:</p><ul><li>Does it work?</li><li>Does it reduce time or friction?</li><li>Can we scale it?</li></ul><p>These are important but incomplete in the future we are actively creating.</p><p>We also need to ask:</p><ul><li>What happens <em>because</em> this works?</li><li>What are we training people <em>not</em> to do anymore?</li><li>What mindsets, habits, or forms of cognition might erode over time?</li></ul><p>That’s where <strong>second-order thinking</strong> comes in.</p><h3>What Is Second-Order Thinking?</h3><p><strong>First-order thinking</strong> asks: <em>“Did this solve the problem?”</em><br> <strong>Second-order thinking</strong> asks: <em>“What are the ripple effects of solving it this way?”</em></p><p>It’s a mental model for surfacing downstream consequences, systemic shifts, and long-term tradeoffs, especially in environments where short-term metrics are easy to measure but long-term impact is not.</p><p>In the context of AI, it’s how we avoid building products that quietly erode user agency, creativity, and capacity for complexity which are some of the very things that make us beautifully human❤.</p><h3>A 6-Part Framework for Conscious Cognitive Offloading</h3><p>If you’re a designer, product leader, or founder building AI features, here’s a framework to support your second order thinking to help map cognitive impact and identify what to build.</p><p>It’s a first step in externalizing the internal process I’ve been using to guide teams through these questions. A V1, shared with love ❤</p><h3>1. Identify the Cognitive Task</h3><p>→ What kind of thinking is being outsourced?<br> (e.g., writing, searching, synthesizing, deciding, remembering)</p><h3>2. Assess User Maturity</h3><p>→ Is the user a novice who needs scaffolding or an expert who benefits from acceleration?</p><h3>3. Map Gains vs. Losses</h3><p>→ What’s being gained (speed, ease, clarity)?<br> → What’s being lost (resilience, skill-building, agency)?</p><h3>4. Ask Second-Order Questions!</h3><p><em>(bring in your assumptions, hypotheses and your insights, UXR, designer friends into answering these)</em></p><p><strong>What are we training users <em>not</em> to do?</strong>Are we bypassing interpretation, discernment, or generative struggle? What growth or reflection disappears?</p><p><strong>What happens if this becomes the default? </strong>How do expectations, values, or attention shift over time? What roles or skills might atrophy?</p><p><strong>What feedback loop are we reinforcing? </strong>Does this make future offloading more likely?Are we amplifying speed over depth?</p><p><strong>What kind of human are we shaping?</strong>Are we designing for curiosity and discernment or compliance and convenience?</p><p><strong>Is this scaffolding mastery or substituting for it? </strong>Is it a stepping stone or a crutch?</p><p><strong>What will we not <em>see</em> because this works too well? </strong>What perspectives, mistakes, or complexity might be filtered out?</p><h3>5. Create an Offloading Impact Grid</h3><p>You’ve got feature ideas start to plot them on two axes:</p><ul><li>Cognitive Value (High ↔ Low)</li><li>Offload Benefit (High ↔ Low)</li></ul><p>→ Assess for each one is this smart augmentation or seductive convenience?</p><h3>6. Run a Shotgun Concept Test (yes, I made this up)</h3><p>When you’re exploring multiple product directions especially in ambiguous spaces like AI where the “best” idea isn’t obvious — I use something I call <strong>Shotgun Concept Testing</strong>.</p><p>The idea comes from experimental design: just like we test different levels of an independent variable, here we test different <em>conceptual directions</em>. Not just features, but frames. A <strong>frame</strong> is the dominant mental model or narrative the product evokes. It sets expectations, shapes how users interpret the system, and guides how they engage over time.</p><p>Try this:</p><ul><li>Define 2–3 concept clusters or plausible product directions (e.g., AI as a mentor, AI as a shortcut, AI as a collaborator)</li><li>Bundle a few core features under each direction, you’re not building prototypes, just enough fidelity to convey the concept and cognitive experience.</li><li>Then run some concept studies and ask: What’s the mental model in each? What does each <em>enable</em> vs. <em>erode</em>?</li></ul><p>This is especially helpful in ambiguous spaces like AI, where the “best” idea isn’t always obvious but the framing you choose shapes cognition, expectations, and long-term user behavior so learning about that framing via shot gun testing helps you get closer to what you actually want for your users.</p><h3>Final Thought</h3><p>AI isn’t just a tool, it’s a cognitive layer now so with these new products and features, we’re not just creating products we’re creating thinking patterns (ya, whoa).</p><p>Now, that kind of power demands more than esoteric ethics conversations and lone UX voices.</p><p>Second-order thinking isn’t a mindset, it is a skill. and like any skill, it can be developed. It’s a lens in which to start to see the world and knowing <em>when</em> to use it is part of the craft of product development. This framework is a V1: a starting point for making that skill real, shared and usable.</p><p>I believe that if we want to build futures worth living in, we need more than what ‘works’ or built because ‘we could’. We need what’s appropriate for humans. Build like it matters, because it does and as always, if it’s shit in, it’s shit out.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1e3ec52031d7" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>