<?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 Meg Kurdziolek on Medium]]></title>
        <description><![CDATA[Stories by Meg Kurdziolek on Medium]]></description>
        <link>https://medium.com/@megak?source=rss-5e4d9b1fe93------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*FQrGCGElldvMHdwezRJDxQ.jpeg</url>
            <title>Stories by Meg Kurdziolek on Medium</title>
            <link>https://medium.com/@megak?source=rss-5e4d9b1fe93------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 07:41:49 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@megak/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[Persistent lessons in human-centered automation]]></title>
            <link>https://uxdesign.cc/persistent-lessons-in-human-centered-automation-45b53412e6a5?source=rss-5e4d9b1fe93------2</link>
            <guid isPermaLink="false">https://medium.com/p/45b53412e6a5</guid>
            <category><![CDATA[design-patterns]]></category>
            <category><![CDATA[ux-research]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[editor-picks]]></category>
            <category><![CDATA[ux]]></category>
            <dc:creator><![CDATA[Meg Kurdziolek]]></dc:creator>
            <pubDate>Tue, 08 Sep 2026 20:34:17 GMT</pubDate>
            <atom:updated>2026-09-08T21:31:27.143Z</atom:updated>
            <content:encoded><![CDATA[<h4>What a 1980s photocopier study reveals about the Agentic UX era</h4><figure><img alt="A real photograph of a Bossa Nova inventory robot in a grocery aisle overlaid with hand-drawn line art sketches. In the foreground, a sketched woman with a frustrated expression holds a container of blueberries up to the robot’s light panel, illustrating a breakdown in human-robot interaction." src="https://cdn-images-1.medium.com/max/1024/1*9TwDqgP0J42XZmP1G_oqiw.jpeg" /><figcaption><em>A mixed-media recreation of a field observation at Bossa Nova Robotics, where a shopper attempts to scan a container of blueberries against an autonomous inventory robot.</em></figcaption></figure><p>During my time as a User Researcher at <a href="https://www.bossanova.com/">Bossa Nova Robotics</a>, I conducted multiple studies in grocery store aisles watching our inventory-assessing robot do its job. My objective was to identify moments when the robot had to pause because of obstructions or human interference. The goal was to find ways for the robot to better navigate the space and indicate what it was doing to nearby humans and, in doing so, optimize its efficiency.</p><p>One afternoon, I watched a woman walk up to the ~300-pound machine. It stopped. She held a container of blueberries up to its light panel. Nothing happened. She then lifted the blueberries to the top of its lidar stack. Still nothing. Frustrated, she muttered, “This thing is so stupid,” and walked away.</p><p>The irony was not lost on me. That robot was filled with high-end cameras and on-board compute running perception algorithms in real-time. The machine was cutting-edge when it came to artificial intelligence. But when it came to coordinating and understanding human intention, the woman was right, it was pretty stupid. The lesson I learned that day was that no matter how much machine intelligence is baked into a system, if it doesn’t account for the unpredictability of human behavior it will still be dumb in human intelligence.</p><p>Today, we are seeing an incredible shift in HCI as LLMs give us new ways of presenting natural-seeming interaction patterns. Looking back to the birth of HCI reveals vital lessons we must apply to modern AI interaction design.</p><h3>The birth of HCI, and the lessons we learned</h3><p>In the 1970s, the Computer Science Laboratory at Xerox PARC was the absolute mecca of innovation. They gave us the graphical user interface, the laser printer, and the first modern personal computers. They were pursuing a radical idea: that computing could transcend massive mainframes and become accessible, responsive tools for individual users.</p><p>Within Xerox PARC, a group affectionately known as the “artificial intelligentsia” drove a prevailing belief. They thought making a machine smarter was the exact same thing as making it useful.</p><figure><img alt="Xerox PARC Computer Science Laboratory researchers gathered on floor beanbags during a 1970s brainstorming session." src="https://cdn-images-1.medium.com/max/600/1*JQEj0wZduFS29AlOmp4QUQ.jpeg" /><figcaption><em>The Xerox PARC Computer Science Laboratory “beanbag” conference room, c. 1970s. Image courtesy of the </em><a href="https://www.computerhistory.org/revolution/input-output/14/348"><em>Computer History Museum</em></a><em>.</em></figcaption></figure><p>However, there was a tension between system-centric and human-centric thought in those days. The system-centric camp believed human action was predictable in some way, and if an AI planning model was smart enough, it would be able to assist humans in accomplishing a task. Researchers Thomas Moran, Stuart Card, and Allen Newell even wrote a foundational book called <a href="https://www.amazon.com/Psychology-Human-Computer-Interaction-Stuart-Card/dp/0898598591">The Psychology of Human-Computer Interaction</a>, which modeled humans as information processors perfectly analogous to computers.</p><p>Anthropologist Lucy Suchman challenged this assumption in her famous “man vs. machine” study. She set up a camera and simply watched two users struggle to make double-sided copies of a large stack of papers with a new Xerox photocopier. When the copier jammed, it instructed the users to reload all the pages in order. The users, however, knew they were nearly done with their task and only needed to copy the last few pages again. They struggled in vain for several minutes, and became so frustrated with the machine’s rigidity that they ultimately cussed out the machine. These users weren’t just any users, however. They were some of the premier experts in computing at the time. The man vs. machine study became cemented in history as the most famous of all usability studies.</p><figure><img alt="Overhead black-and-white video capture of two researchers leaning over a large Xerox photocopier in frustration, with a caption subtitle reading “You gotta be kidding.”" src="https://cdn-images-1.medium.com/max/1024/1*DnMmZ66acBQ_jk-cE0yTMQ.png" /><figcaption><em>A scene from Lucy Suchman’s foundational “man vs. machine” usability study at Xerox PARC. Screenshot from </em><a href="https://www.youtube.com/watch?v=cNJWafS-BA4"><em>YouTube</em></a></figcaption></figure><p>Through this study, Suchman showed us that human work is not a rigid plan. It is moment-to-moment improvisation. Human behavior can only be understood in relation to real-world situations. The <a href="https://en.wikipedia.org/wiki/Plans_and_Situated_Actions">metaphor she used</a> to describe this was that human action is like navigating a canoe through rapids. You start with a plan, but you make moment to moment decisions while you are in the experience.</p><p>A few years later, technologists at PARC decided to tackle a new challenge: designing technologies to facilitate human-to-human collaboration. Researchers like Deborah Tatar (who would later become my academic advisor in graduate school) observed early collaborative systems like <a href="https://doi.org/10.1016/0020-7373(91)90041-5">Colab and Cognoter</a>. These systems relied on a rigid ‘parcel-post’ model of communication. They treated human interaction like discrete packages being mailed back and forth. But in Tatar’s evaluation of the system, she found that real communication is a messy dance of shared gazes, subtle nods, and constant adjustments based on our partner’s reactions. Tatar and others learned that when machines enforce rigid logic, human-to-human collaboration collapses, and they are ultimately unsuccessful, and unsatisfied, with such a tool.</p><p>At the end of this era, we saw the rise of HCI and CSCW as new fields of research within computing. And we walked away with a clear set of requirements for successful human-centric technology design:</p><ul><li><strong>Action is Situated: </strong>Human behavior cannot be predicted by a rigid script. It must always be understood in relation to the messy, real-world context the user is currently experiencing.</li><li><strong>Account for the Machine Sensory Gap:</strong> Machines only “know” the limited slice of reality that their sensors and training data tell them. Humans know the actual world. Systems must be designed to bridge that gap.</li><li><strong>Plans are Resources, Not Blueprints: </strong>A plan is just a rough guide. Real work requires moment-to-moment improvisation to overcome unexpected hurdles.</li><li><strong>Interaction is Improvised:</strong> Communication is a coordinated, back-and-forth dance of shared meaning, not a simple delivery of information bits from point A to point B.</li><li><strong>Person-Centered Design:</strong> Systems must be designed to increase human competence and pride in work, rather than forcing humans to adapt to the limitations of the machine.</li></ul><h3>The power of context gives rise to AIX</h3><p>Back then, traditional systems, AI powered or otherwise, failed those requirements. They lacked the context and flexibility to support situated action or improvised interaction. As a result, for the last 50 years UX Design has developed principles and design patterns that focus on making systems look more like discrete tools in a tool box with obvious affordances rather than monolithic be-all-end-all machines. The systems are rigid, but the design allows you to understand their utility and use them together to accomplish a number of tasks.</p><p>But now, LLMs have us rethinking UX design entirely. Today, large language models are passing some of those requirements with flying colors. AI has moved from rigid, deterministic scripts to probabilistic, flexible reasoning. LLMs possess massive context, memory, and the capacity to repair misunderstandings interactively. Instead of treating communication like a rigid ‘parcel-post’ system, LLMs can engage in the messy dance of human conversation.</p><p>This feels like a reunion between AI and human-centered design. We can coordinate shared meaning with the machine. In fact, researchers are already identifying an entirely new interaction paradigm: the Agent interaction model (AIX). Unlike traditional graphical user interfaces that rely on explicit, step-by-step commands, AIX allows humans to state an intent and let the system proactively reason through the execution. It pushes HCI forward in thrilling new ways, allowing us to collaborate with machines almost like we would with a colleague.</p><h3>The illusion of complete intelligence: where LLMs still stumble</h3><p>The conversational fluency of modern LLMs creates a powerful illusion. When a system communicates in polished, natural prose, our brains naturally assume it possesses general human competence. We project common sense, situational awareness, and real-world understanding onto the machine.</p><p>This assumption is dangerous. While LLMs solve the rigid syntax problem of early computing, they do not automatically solve the five requirements of human-centered design.</p><p>The machine sensory gap has not disappeared; it has simply moved behind a curtain of sophisticated language. In their landmark paper <a href="https://dl.acm.org/doi/10.1145/3442188.3445922">On the Dangers of Stochastic Parrots</a>, Emily Bender, Timnit Gebru, and their co-authors pointed out that large language models are systems for stitching together linguistic forms based on statistical likelihood, without any underlying model of intent, meaning, or physical reality. An LLM operates strictly on statistical associations across textual tokens. It has no physical embodiment, no lived presence in the physical room, and no direct perception of the user’s unstated emotional state, physical distractions, or immediate workspace constraints. When an edge case occurs, the model cannot look around the room to see what went wrong. It can only generate the next most probable sequence of words.</p><p>This creates the problem of jagged capabilities. An AI agent might synthesize hundreds of pages of research in seconds, and in the very next step hallucinate a non-existent corporate policy or misinterpret a safety constraint. Because the output format remains confident and articulate throughout, users struggle to tell where the model’s true competence ends and its blind speculation begins.</p><p>The interface paradigms we use today exacerbate this issue. Traditional graphical user interfaces (GUIs) rely on visible menus, buttons, and state indicators that show users what actions are possible. Pure conversational interfaces (Language User Interfaces, or LUIs) strip away these visible affordances. Users face an empty text box, left to guess what the system can do, what context it actually retains, and where its failure boundaries lie.</p><h3>Trust does not happen in a 1:1 vacuum</h3><p>To build systems people can actually rely on, we have to look closely at how trust operates in the real world. Trust is not a static score or an isolated 1:1 transaction between a human and a computer. It is an ongoing negotiation that happens across networks of people working together.</p><p>When I worked on <a href="https://uxpamagazine.org/explaining-the-unexplainable-explainable-ai-xai-for-ux/">Explainable AI (XAI) services for Google Cloud</a>, we evaluated user needs across high-stakes domains like digital pathology and clinical decision support. We observed that an AI model never operates alone in an exam room. It sits inside a complex web of relationships involving pathologists, oncologists, patients, data scientists, and regulatory auditors.</p><p>Each person in that network approaches the system with entirely different sensemaking goals:</p><ul><li><strong>The Data Scientist</strong> needs technical interpretability: feature attributions, loss curves, and data drift metrics to verify the pipeline.</li><li><strong>The Pathologist</strong> needs clinical rationale: why the model flagged a specific tissue cluster as malignant, which cellular patterns drove the prediction, and what differential diagnoses were considered.</li><li><strong>The Oncologist and Patient </strong>need collaborative clarity: what the findings mean for treatment options, how certain the prognosis is, and what trade-offs exist.</li><li><strong>The Auditor</strong> needs provenance: an unalterable trail showing data lineage, model versioning, and compliance with clinical standards.</li></ul><figure><img alt="Infographic diagram showing multi-stakeholder coordination around an Explainable AI model, illustrating how data scientists, auditors, doctors, and patients have distinct needs while doctors and patients engage in a shared sensemaking loop." src="https://cdn-images-1.medium.com/max/1024/1*MRoQGc8vj_5ThFmhts5Pqw.jpeg" /><figcaption>Different stakeholders have distinct explanatory needs, and users such as doctors and patients will coordinate action and make meaning together through the system.</figcaption></figure><p>These stakeholders do not just consume model outputs individually. They use the tool to coordinate decisions with one another. The oncologist discusses the findings with the patient, pointing to visual evidence on the screen to build a shared treatment plan.</p><p>Current LLM interfaces almost completely ignore this multi-stakeholder reality. They operate as private, siloed chatbots designed for a single user staring at a prompt box. We lack interaction patterns that support group sensemaking, where multiple humans can interrogate, annotate, and co-evaluate AI reasoning together.</p><h3>New interaction patterns for the intervention handshake</h3><p>If we want agentic systems to succeed where previous generations of automation failed, we must design for human sensemaking from the start. We need concrete interaction patterns that help people calibrate their trust, understand system state, and step in to steer the machine without friction.</p><p>Here is a list of key design patterns for the intervention handshake, and I’ll cover each in more detail below:</p><ol><li><strong>Calibrated Uncertainty &amp; Grounded Provenance:</strong> Visually distinguish facts from speculative leaps; make confidence inspectable and anchor claims in source evidence.</li><li><strong>Continuous Mixed-Initiative Steering:</strong> Enable lightweight course correction (steerable scratchpads, inline checkpoints) without resetting state.</li><li><strong>Shared Mental Models:</strong> Provide collaborative surfaces where multiple stakeholders can co-interrogate and guide AI reasoning together.</li><li><strong>Progressive Disclosure: </strong>Expose underlying decision trees, retrieved chunks, and alternative logic on demand during high-stakes moments.</li></ol><h4>1. Calibrated uncertainty and grounded provenance</h4><p>AI systems should never present speculative inferences with the same authoritative tone they use for verified facts. Interfaces must make uncertainty visible. When an agent synthesizes information or suggests an action, it should visually distinguish between high-confidence retrieval and probabilistic extrapolation. Surfacing underlying evidence, inline citations, and inspectable source traces allows users to verify claims quickly without having to double-check every single sentence from scratch.</p><h4>2. The intervention handshake (continuous mixed-initiative steering)</h4><p>Because real work is improvised, human-AI collaboration cannot be an all-or-nothing handoff. In traditional automation, handing control over to a machine often meant losing situational awareness until an emergency forced an abrupt, disorienting takeover.</p><p>Agentic systems need continuous, lightweight steering mechanisms. Users should be able to pause an ongoing workflow, prune an unproductive reasoning branch, tweak a parameter, or supply missing context mid-execution without blowing away the agent’s progress. Interaction patterns like steerable scratchpads, inline checkpoints, and editable thought trails turn the system into a collaborative workspace rather than an uncontrollable black box.</p><h4>3. Shared mental models and multi-user sensemaking</h4><p>We need interfaces that move beyond the single-user prompt box to support collaborative analysis. This means creating shared canvas spaces where multiple team members can view the agent’s intermediate steps, challenge its assumptions, attach domain-specific constraints, and align on decisions together. The AI becomes a shared surface for group sensemaking rather than an opaque intermediary.</p><h4>4. Progressive disclosure of machine logic</h4><p>Different tasks require different levels of inspection. During routine operations, users need concise, actionable outputs that preserve cognitive bandwidth. But when anomalies arise or high-stakes decisions must be made, the interface must allow users to peel back layers of reasoning progressively. Exposing decision trees, retrieved context chunks, and alternative options on demand gives users the exact depth of detail they need at any given moment.</p><h3>Designing for human competence over machine autonomy</h3><p>Fifty years ago, Xerox PARC showed us that treating humans as rigid information processors leads to fragile, frustrating technology. The lesson remains unchanged today. Making a model larger, faster, or more articulate does not eliminate the need for human sensemaking.</p><p>The goal of AI design is not to build autonomous machines that run without human involvement. The goal is to design systems that expand human competence, foster situational awareness, and support genuine collaboration. When we design for situated action, bridge the machine sensory gap, and build clear affordances for human steering, we transform AI from an unpredictable black box into a reliable partner.</p><p><strong>References and Recommended reading</strong></p><ul><li><a href="https://en.wikipedia.org/wiki/Plans_and_Situated_Actions">Plans and Situated Actions</a> by Lucy Suchman (book) — The foundational ethnographic study demonstrating why human action is improvised and situated rather than script-driven.</li><li><a href="https://www.amazon.com/Psychology-Human-Computer-Interaction-Stuart-Card/dp/0898598591">The Psychology of Human-Computer Interaction</a> by Stuart Card, Thomas Moran, and Allen Newell (book) — The seminal text modeling humans as information processors that spurred the debate between system-centric and human-centric design.</li><li><a href="https://doi.org/10.1016/0020-7373(91)90041-5">Design for Conversation: Lessons from Cognoter</a> by Deborah Tatar et al.(paper) — An early Xerox PARC study on the breakdowns that occur when software enforces rigid communication packages on human collaboration.</li><li><a href="https://dl.acm.org/doi/10.1145/3442188.3445922">On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?</a> by Emily M. Bender et al. (paper) — The landmark critique explaining why large language models stitch together statistical forms without genuine comprehension or real-world grounding.</li><li><a href="https://uxpamagazine.org/explaining-the-unexplainable-explainable-ai-xai-for-ux/">Explaining the Unexplainable: Explainable AI (XAI) for UX</a> by Meg Kurdziolek(article) — A practical look at designing multi-stakeholder trust and transparency in complex machine learning workflows.</li><li><a href="https://www.sciencedirect.com/science/article/pii/S209657962500021X">A New Human-Computer Interaction Paradigm: Agent Interaction Model (AIX)</a> (paper) — An analysis of how intent-driven agentic systems are reshaping the boundaries of human-computer interaction.</li><li><a href="https://www.microsoft.com/en-us/research/project/guidelines-for-human-ai-interaction/">Guidelines for Human-AI Interaction</a> by Microsoft Research (framework) — Industry-standard design heuristics for transparency, error recovery, and user control in AI systems.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=45b53412e6a5" width="1" height="1" alt=""><hr><p><a href="https://uxdesign.cc/persistent-lessons-in-human-centered-automation-45b53412e6a5">Persistent lessons in human-centered automation</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[The Interesting Social Physics of My Midlife Crisis Lists]]></title>
            <link>https://medium.com/@megak/the-interesting-social-physics-of-my-midlife-crisis-lists-e6371331a7b7?source=rss-5e4d9b1fe93------2</link>
            <guid isPermaLink="false">https://medium.com/p/e6371331a7b7</guid>
            <category><![CDATA[life-lessons]]></category>
            <category><![CDATA[adventure]]></category>
            <category><![CDATA[friendship]]></category>
            <category><![CDATA[psychology]]></category>
            <category><![CDATA[personal-growth]]></category>
            <dc:creator><![CDATA[Meg Kurdziolek]]></dc:creator>
            <pubDate>Sat, 05 Sep 2026 20:28:52 GMT</pubDate>
            <atom:updated>2026-09-05T20:28:52.468Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="A winding dirt hiking trail descends into a dramatic red-rock canyon gorge surrounded by desert vegetation, with sunlit clouds and layered canyon ridges in the distance." src="https://cdn-images-1.medium.com/max/1024/1*EC5NitMYELuhxmHyxkjMDg.jpeg" /><figcaption>A view to where we were going as the sun rose on the north rim</figcaption></figure><p>In early 2023, my life fell apart in rapid succession. I lost my grandmother, got laid off from a job I loved, and had to say goodbye to my cat of seventeen years. By the time I hit my 40th birthday in May, I realized I had spent my 30s waiting. I was waiting for my kids to grow up, for my body to recover, for the pandemic to end, and for work to get easier. I decided that my 40s would be about doing. I built <a href="https://medium.com/@megak/my-mid-life-crisis-list-aa568c581285?sharedUserId=megak">an aspirational “midlife crisis” list</a> of hard, weird, and exciting things to try.</p><p>I am now in year three of this experiment. When I created my first midlife crisis list, my intention was to pursue all of these things independently. I did not plan on anyone joining me. But when I shared the list with friends, something unexpected happened. People started claiming items to do with me. The list did not isolate me. It helped me build my community.</p><h3>The magnetism of specificity</h3><p>Adult socializing often relies on low-stakes coffee chats or professional networking. We stick to polite surface conversations that rarely yield true intimacy.</p><p>Sharing a highly specific list of personal interests acts as a signal flare. It filters out the noise. When I added “catch a big fish,” “go backpacking,” or “drink a good whiskey” to my list, I was not asking for generic company. I was sharing a specific interest.</p><p>This specificity attracts the right people. Friends and acquaintances who share those exact interests immediately raise their hands. We bypass the awkward phases of getting to know each other and jump straight to common ground.</p><p>When I listed “drink a good whiskey,” I quickly discovered which of my friends were whiskey connoisseurs. Getting a tattoo enabled me to meet my tattoo artist. Sitting together in that unique, highly personal space created a connection that would never have formed in a normal social routine.</p><h3>Sculpting the ideal self through Type 2 fun</h3><p>I have always enjoyed “Type 2 fun.” This is the kind of fun that is miserable while it is happening, but glorious in retrospect. Doing difficult things with other people builds trust faster than any team-building exercise.</p><p>I met one of my best friends, Leslie, while training for my first marathon in 2019. We have been Type 2 fun partners ever since. When I listed “backpacking” on my crisis list, she immediately joined in. Planning and surviving rugged adventures has become our shared rhythm.</p><p>In relationship psychology, the <a href="https://doi.org/10.1037/0022-3514.77.2.293">Michelangelo Phenomenon</a> explains this dynamic. The theory suggests that close partners and friends act as “sculptors” of our ideal selves. Michelangelo described sculpting as chipping away at a block of marble to reveal the statue already hidden inside. Similarly, close friends help us chip away hesitation, waiting, and fear. They do this through <em>perceptual affirmation</em> (seeing who we want to become) and <em>behavioral affirmation</em> (acting in ways that help us get there).</p><p>High-friction activities force this vulnerability. There is no hiding your frustration or fatigue when you are carrying a heavy backpack in the wilderness with a friend, or when you are sitting under a tattoo needle.</p><p>This shared struggle creates deep bonds:</p><ul><li><strong>Backpacking with Leslie: </strong>Attempting long distance hikes and carrying heavy packs up hills together has turned our mutual interest in endurance into a great friendship. We act as each others “sculptors”, affirming our adventurous selves.</li><li><strong>Crying at the finish line:</strong> I have often found myself on the verge of tears as I approach the end of marathons and ultramarathons. It is a raw mixture of exhaustion, relief, and pride. When I went to the Eagle Up Ultra this year with my running club, it was natural to cheer and cry as each one of us crossed the finish line. Rushing to hug a friend who is also crying at a finish line creates a level of intimacy that coffee chats cannot replicate.</li><li><strong>The sisterhood canyon hike:</strong> I recently completed a rim-to-rim hike of the Grand Canyon with my sisters. It was a nearly spiritual experience. A water main break on the North Rim forced us to carry all our own water. Under the 115*F desert sun, real anxiety set in as we drank our final drops just 100 meters from Phantom Ranch. With the ranch water shut off and showers unavailable, we washed off the trail dust in Bright Angel Creek. Starting at 4:00 AM to beat the July heat, watching the morning sun light up the canyon walls after hours of climbing delivered the most breathtaking views of my life. Tackling that grueling canyon together helped my sisters sculpt a daring, resilient version of who I am.</li></ul><figure><img alt="Selfie of three women smiling on a narrow steel suspension bridge spanning the Colorado River, with towering rocky canyon walls rising in the background under a clear blue sky." src="https://cdn-images-1.medium.com/max/1024/1*5mtww9gRZvHohzs11-tuFw.jpeg" /><figcaption>My sisters and I standing on the silver bridge over the Colorado river</figcaption></figure><h3>Transactive dynamics at 4:00 AM</h3><p>The value of these goals lies as much in the failures as the successes. When I put “catch a big fish” on my list, my coworker and friend Michael immediately volunteered to help.</p><p>In psychology, <a href="https://doi.org/10.1037/a0039654">Transactive Goal Dynamics (TGD) Theory</a> explains this transition from individual goals to shared effort. TGD theory suggests that in close relationships, individuals do not regulate their goals in isolation. Instead, they operate as a single, self-regulating system. We share cognitive load, coordination, and physical effort.</p><p>When Michael jumped in, my goal became our system’s goal. He brought the boat, I brought the list, and we coordinated our schedules.</p><p>Our first trip was a complete failure. We spent hours on the water and caught nothing. But we shared the disappointment. We sat on the boat and talked. In a transactive system, failure is not an individual burden; it is a shared experience. That shared letdown brought us closer than a quick, easy victory would have.</p><p>The second trip was different. Michael wanted a guaranteed success, so he took me out of Half Moon Bay to a spot where he regularly caught Lingcod. He was a patient teacher, and I caught multiple fish that morning. The boat ride back was spectacular: we saw sea lions swimming, big pelicans flying, and a whale was hanging around the harbor with a crowd watching from shore. The triumph was sweet, but the memory of the early morning boat ride and the shared effort was the real prize. I would have slept through that particular sunrise if I had kept my list to myself.</p><figure><img alt="A smiling woman on an aluminum fishing boat holds up a large, mottled gray-and-brown Lingcod caught during an ocean fishing trip off Half Moon Bay." src="https://cdn-images-1.medium.com/max/1024/1*st8xFu4OoG669QkRSNvAXw.jpeg" /><figcaption>Me on Michael’s boat, with the biggest fish I’ve ever caught.</figcaption></figure><h3>The mathematics of accountability</h3><p>A midlife crisis list is contagious. When you share your goals, you give others permission to voice their own.</p><p>Empirical research supports this public sharing. A <a href="https://scholar.dominican.edu/cgi/viewcontent.cgi?article=1002&amp;context=psychology-faculty-conference-presentations">well-known study by Dr. Gail Matthews</a> at the Dominican University of California found that writing down goals, making action commitments, and sharing progress reports with a friend increases goal success to over 70 percent. In contrast, those who keep goals private succeed only 35 to 43 percent of the time.</p><p>Matthews’ study shows that sharing the list is not just a social catalyst. It is an execution framework. When we make our goals public, we leverage social commitment. The social physics of the list are also the mathematics of follow-through.</p><p>This year, my physical goals are demanding. I am training for my first pull-up and working toward a 200lb squat. I am also taking my daughter camping, getting more tattoos, and I have practiced Spanish for over 100 consecutive days.</p><p>When people see these pursuits, they do not just offer polite support. They start talking about their own hidden dreams. The list shifts from a personal project to a collective rebellion against the waiting trap.</p><h3>Finding your people through deliberate crises</h3><p>After talking with people for a while, I often find myself asking, “Have I told you about my very deliberate midlife crisis?” Then I tell them about the lists.</p><p>The response is consistent. I have never received anything less than enthusiastic warmth and support. It turns out that the people who lean in are the exact type of people I want in my life: individuals who love the pursuit of hard goals and good fun. Those are my people.</p><p>If you feel disconnected or stuck in the waiting trap, do not just schedule more coffee chats. Stop waiting for the perfect moment.</p><p>Write a list of hard, weird, and exciting things you want to do. Share it with your friends, your coworkers, and your family. Let them see what you care about. Some of them will walk away, but the right ones will tell you their favorite whiskey, hike the Grand Canyon with you, or show up at 4:00 AM with a boat to help you catch a fish.</p><figure><img alt="A group of eight women runners smiling together in running gear and finisher medals beneath the Tampa Convention Center Gasparilla Distance Classic marquee." src="https://cdn-images-1.medium.com/max/1024/1*uMsK662UMnSlAgIlIope_A.jpeg" /><figcaption>A collection of my running club friends after finishing the Gasparila Half last year</figcaption></figure><h3>Recommended reading</h3><ul><li><a href="https://medium.com/@megak/my-mid-life-crisis-list-aa568c581285?sharedUserId=megak">My mid-life crisis list</a> (article) — The original 2024 essay that initiated this three-year experiment in stopping waiting.</li><li><a href="https://doi.org/10.1037/a0039654">Transactive Goal Dynamics</a> (academic journal) — The foundational paper by Fitzsimons, Finkel, and vanDellen introducing how close relationships self-regulate as a single system.</li><li><a href="https://doi.org/10.1037/0022-3514.77.2.293">Close partner as sculptor of the ideal self: The Michelangelo phenomenon</a> (academic journal) — Drigotas et al.’s research demonstrating how relationships sculpt partners toward their ideal selves.</li><li><a href="https://scholar.dominican.edu/cgi/viewcontent.cgi?article=1002&amp;context=psychology-faculty-conference-presentations">The Impact of Commitment, Accountability, and Written Goals on Goal Achievement</a> (academic paper) — Dr. Gail Matthews’ landmark empirical study showing that sharing goals and weekly progress reports with a friend increases achievement to over 70 percent.</li><li><a href="https://en.wikipedia.org/wiki/Bowling_Alone">Bowling Alone</a> (book) — Robert D. Putnam’s seminal study on the decline of social capital and the importance of active community building.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e6371331a7b7" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The psychological cost of moving too fast]]></title>
            <link>https://uxdesign.cc/the-psychological-cost-of-moving-too-fast-867fb3830722?source=rss-5e4d9b1fe93------2</link>
            <guid isPermaLink="false">https://medium.com/p/867fb3830722</guid>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[design]]></category>
            <category><![CDATA[ux]]></category>
            <category><![CDATA[product-development]]></category>
            <category><![CDATA[product-design]]></category>
            <dc:creator><![CDATA[Meg Kurdziolek]]></dc:creator>
            <pubDate>Sat, 06 Jun 2026 12:15:41 GMT</pubDate>
            <atom:updated>2026-06-06T12:15:41.990Z</atom:updated>
            <content:encoded><![CDATA[<h4>Why discovery still matters in the age of AI</h4><figure><img alt="A minimalist, charming children’s picture book style watercolor illustration of the Tortoise and the Hare metaphor applied to software development. The hare sprints frantically on the main path toward a crumbling dead end cliff, while the tortoise navigates safely onto a beautiful ‘happy path’ that branches off." src="https://cdn-images-1.medium.com/max/1024/1*ydyT3iGEx2xP_WOys-yNqQ.png" /></figure><p>Historically, I used to pitch investment in early UX work as a cost-saving effort. Engineering time and resources were the most expensive assets a company possessed. You couldn’t afford to build the wrong thing, so you invested in discovery to ensure the target was correct before firing the arrow.</p><p>Today, the math has changed. With the rise of autonomous AI coding agents like Antigravity, Cursor, and Claude Code, the financial cost of engineering is plummeting. You can spin up a working prototype in an afternoon. The friction of going from idea to functional software has practically vanished.</p><p>But while the financial cost of building the wrong thing has dropped, the psychological cost has not. There is a hidden, cognitive toll to skipping the ideation and discovery phases. When we move too fast, we fall victim to design fixation. We harden our ideas too early and lose the ability to pivot.</p><h3>The trap of Design Fixation</h3><p>The psychological cost of speed is the cost of hardening your ideas early. When an idea remains on a whiteboard or in a Figma file, it is malleable. We are open to criticism. We are agile. But the moment an idea becomes <em>solid</em>, our brains shift.</p><p>Researchers examining the <a href="https://www.researchgate.net/publication/260288902_Role_of_Sunk_Cost_in_Engineering_Idea_Generation_An_Experimental_Investigation">Role of Sunk Cost in Engineering Idea Generation</a> use a brilliant term for this: <em>Design Fixation</em>. In an experimental study by Viswanathan et al., researchers tested how the time and effort spent building physical prototypes affected innovation. They divided participants into groups: one group only sketched ideas, another built models with easy-to-use metal materials (low time cost), and a third group built models with plastic materials that required high time and effort.</p><p>The results were striking. The participants who spent the most effort building the plastic models generated ideas with significantly less novelty and variety. The study concluded that prototypes themselves don’t cause fixation. It is the <em>sunk cost</em> of the effort invested that limits a designer’s willingness to explore alternatives.</p><p>Even when the financial investment of AI coding is low, the emotional and cognitive investment of seeing a functional prototype triggers this same sunk cost fallacy. We become subject to confirmation bias. We start looking for data that proves our creation is right, rather than data that proves our initial idea was wrong. At the exact moment a team needs to be accepting of new information, they dig in their heels.</p><h3>Design Science: Prototyping as inquiry</h3><p>We can look to the academic framework of <a href="https://www.researchgate.net/publication/287561334_An_Introduction_to_Design_Science">Design Science</a> to structurally enforce discovery when going straight to code feels frictionless.</p><p>At its core, as outlined by researchers like Paul Johannesson and Erik Perjons, Design Science is a process of “nested problem solving.” It distinguishes between solving <em>practical problems </em>(improving a situation for a stakeholder) and solving <em>knowledge problems</em> (learning something new about the world).</p><p>The framework offers clear advice for teams: treat practical problems as engineering tasks that require changing the world (building an artifact), but treat knowledge problems as research tasks that require observing the world (empirical testing). You cannot solve a knowledge problem just by building more code; you have to stop and measure.</p><p>In practice, Design Science relies on an iterative design cycle. You design an artifact to solve a problem, but you treat that artifact as a method of inquiry. The prototype is not the answer; it is the question. By adopting a design science mindset, teams use rapid prototyping to promote empirical, validated learning. You build quickly, but you do so explicitly to test and break your assumptions.</p><h3>Speed to MVP = speed to confirmation bias</h3><p>If we view it through this lens, the original Lean Startup philosophy was simply an extension of design science thinking. The concept of a Minimum Viable Product (MVP) was introduced as a means of maximizing validated learning about a business idea with the least amount of effort and time invested. The MVP was never the end destination in and of itself, rather a quick stepping stone to learning.</p><p>But teams trying to follow the MVP philosophy often fall victim to design fixation. Studies like <a href="https://essay.utwente.nl/fileshare/file/74732/Schwery_BA_BMS.pdf">Lean Startup Orientation: Empirical Evidence on Venture Success</a>, show how rapid prototyping reduces uncertainty, but also warn that confirmation bias leads startups to misinterpret test results. When you view the MVP as the product, you twist user data to validate the code you already wrote.</p><p>Today, because AI can build an MVP so quickly, the problem is exacerbated. Teams are moving even faster towards something they feel is a solid product when in fact they are working with an untested, unvalidated, concept.</p><h3>Repositioning the value of early UX</h3><p>This increased speed of development also fundamentally changes the power dynamics of the modern product team. Historically, UX could gatekeep bad ideas because engineering was an expensive, scarce resource. Now, founders and product managers can bypass engineering entirely and build the prototype themselves in a weekend. Without the friction of engineering costs, UX must shift their value proposition. You can no longer pitch discovery as a way to save engineering time. You must pitch it as a way to <em>save market and reputational risk</em>. UX must advocate for discovery by showing that while building is cheap, launching something nobody wants still destroys brand equity and user trust.</p><p>This rapid onset of design fixation is exactly why the concept of the <a href="https://medium.com/@megak/the-mve-minimum-viable-experiment-b2498fda9f93">Minimum Viable Experiment (MVE)</a> is so critical. The MVE helps reframe the purpose of the MVP. An MVE is not a product; it is a probe. It should be considered a strictly “codeless” way of testing an idea, an experiment designed to test the single, riskiest assumption with minimal setup. For example, at Intrinsic over the past year, our UX team has been helping product teams build out <a href="https://medium.com/@megak/ux-mini-lesson-storyboards-e6c9935a33c8">storyboards </a>as a practice of mocking up ideas for product concepts. This allows us to get early feedback before any real prototype has been developed. By separating the experiment from the product, teams are forced to stay in the discovery mindset long enough to actually learn something. And when a teammate argues that building an MVP with AI is so fast that launching it should just be the experiment, we must remind them of the psychological toll: the MVE isn’t there to save us time anymore. It is there to protect us from falling in love with our own code.</p><h3>Strategies for Design Science in the age of AI: Scenario-Focused Engineering</h3><p>The very AI tools that push us into creating functional prototypes too soon also offer an unprecedented opportunity to practice good design science, if we use them right. Because iteration is now so cheap, we can run empirical tests and pivot faster than any previous generation of engineers.</p><p>To do this, we need frameworks that anchor us to the user rather than the code. Popularized by Microsoft, <a href="https://www.oreilly.com/library/view/scenario-focused-engineering-a/9780133967241/">Scenario-Focused Engineering (SFE)</a> is a customer-centric methodology that shifts teams away from building technical feature lists and toward designing end-to-end user experiences. SFE relies on fast feedback loops and scenarios. It is the perfect framework for modern AI development practices because when an AI agent can build a feature instantaneously, you need a system that forces you to define the <em>human scenario</em> before you unleash the AI on the <em>technical execution</em>.</p><p>Here is what a modern, AI-augmented SFE practice could look like:</p><ol><li><strong>The Interrogation Prompt: </strong>Before generating any code, use AI agents to aggressively challenge your core premise. Ask the agent to play the role of a skeptical user or a competing product manager. Force the team to generate three alternative scenarios before writing a single line of code.</li><li><strong>The “Disposable” Milestone:</strong> Treat the first AI-generated prototype as explicitly disposable. Keep it outside the production environment entirely. Set a rigid team rule: “we always throw away the first one.” To enforce this, design the prototype with deliberate visual friction. Tell the AI agent to explicitly watermark the interface with “TO BE THROWN AWAY,” design it entirely in black-and-white, or use a generalized wireframe design system. By making it visually obvious that the prototype is not final, you add friction to any attempt to push it to production without proper review, directly combating design fixation.</li><li><strong>Bias-Check Reviews:</strong> Integrate methods of investigation at specific points in the product development lifecycle. We can systematically check for bias using the very tools that helped us build the prototype. After a user testing session, feed the raw transcripts to an LLM alongside your product team’s summary of the findings. Prompt the AI to act as an objective third-party auditor: “<em>Identify any instances where the team’s summary dismisses negative feedback as user error, or highlight data points in the transcript that contradict the team’s conclusions.”</em></li></ol><p>However, we must anticipate the messy human reactions to some of these strategies. When a team is dug in, having a machine tell them they are biased can spark hostility. Facilitating this requires positioning the AI not as an accuser, but as an impartial mirror. The AI can be impartial precisely because it is unaware of company politics and does not have any real desires in and of itself. To make this work, UX teams must establish this framing at the earliest possible time in development. Talk to the team about addressing confirmation bias and sunk cost fallacy as a core part of the way you work together. Pre-frame the review session by saying, “We know we are too close to the code, so we are having the AI stress-test our conclusions.” By shifting the critique to the AI, we can safely facilitate a discussion about uncomfortable data without triggering a defensive argument between teammates.</p><h3>The enduring value of discovery</h3><p>I have worked as a UX Researcher for a long time. In all those years, across countless projects and product teams, one truth has remained absolute: I have never conducted a study that didn’t get a reaction. However, depending on how built-out the product was, the team’s reaction to that research changed drastically. Teams who received disappointing results on a loose wireframe were grateful for the data and immediately got to work on the next idea. Teams who received disappointing results on something that was nearly fully built would react in ways typical of confirmation bias: denial, discomfort, sudden disinterest, and rationalizing away the negative feedback as “user error.”</p><p>My strategy for getting in front of team confirmation bias has always been to work with them earlier in the design process. Now AI has radically compressed that timeline, shrinking the window from weeks after initial product effort is launched down to just a couple of days. This is why we must continually reframe necessary product discovery work. It is no longer about saving engineering time. It is the only way to mitigate the massive market and reputational risk of building the wrong thing at warp speed.</p><p>The barrier to writing functional code has never been lower. But speed of execution is useless if we are running in the wrong direction. We must embrace AI for its ability to build quickly, but we must fiercely defend the discovery phase. If we don’t, we will just find ourselves building the wrong thing faster than ever before.</p><h3>Recommended reading</h3><ul><li><a href="https://essay.utwente.nl/fileshare/file/74732/Schwery_BA_BMS.pdf">Lean Startup Orientation: Empirical Evidence on Venture Success</a> (thesis) — Empirical evidence on how rapid prototyping reduces uncertainty, and the dangers of misinterpreting test results in early startups.</li><li><a href="https://www.researchgate.net/publication/287561334_An_Introduction_to_Design_Science">An Introduction to Design Science</a> (book)— A foundational overview of using artifacts and prototypes as methods of inquiry.</li><li><a href="https://medium.com/@dmdinithipurna/design-science-research-43c55d36e583">Design Science Research: A Simple Introduction</a> (article)— A great, mile-high overview and introduction to Design Science.</li><li><a href="https://www.researchgate.net/publication/260288902_Role_of_Sunk_Cost_in_Engineering_Idea_Generation_An_Experimental_Investigation">Role of Sunk Cost in Engineering Idea Generation</a> (article)— An experimental investigation into design fixation and the sunk cost fallacy in prototyping.</li><li><a href="https://www.oreilly.com/library/view/scenario-focused-engineering-a/9780133967241/">Scenario-Focused Engineering</a> (book)— Practical guidance on using early prototypes to facilitate course corrections and maintain customer-centricity.</li><li><a href="https://medium.com/@megak/the-mve-minimum-viable-experiment-b2498fda9f93">The MVE: Minimum Viable Experiment</a> (article)— A breakdown of why shifting from MVP to MVE helps entrepreneurs test ideas without the pressure of building.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=867fb3830722" width="1" height="1" alt=""><hr><p><a href="https://uxdesign.cc/the-psychological-cost-of-moving-too-fast-867fb3830722">The psychological cost of moving too fast</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[Practical agentic workflows for UXR]]></title>
            <link>https://medium.com/design-bootcamp/practical-agentic-workflows-for-uxr-56b94fbd1ecb?source=rss-5e4d9b1fe93------2</link>
            <guid isPermaLink="false">https://medium.com/p/56b94fbd1ecb</guid>
            <category><![CDATA[ep]]></category>
            <category><![CDATA[user-experience]]></category>
            <category><![CDATA[user-research]]></category>
            <category><![CDATA[agentic-workflow]]></category>
            <category><![CDATA[ux-research]]></category>
            <dc:creator><![CDATA[Meg Kurdziolek]]></dc:creator>
            <pubDate>Sun, 17 May 2026 16:20:23 GMT</pubDate>
            <atom:updated>2026-05-17T16:20:23.754Z</atom:updated>
            <content:encoded><![CDATA[<h4>Moving beyond zero-shot prompts to scale your research</h4><figure><img alt="A human researcher guiding a structured, glowing AI data pipeline" src="https://cdn-images-1.medium.com/max/1024/1*UW_yMtvvyxC3wnVbo9bbIg.png" /></figure><p>The noise surrounding AI in qualitative research is deafening. Every week brings a new promise that researchers will soon be replaced by autonomous systems capable of conducting interviews, analyzing sentiment, and synthesizing reports without human oversight. Yet, when practitioners actually sit down to use these tools, the reality is often underwhelming. Dumping a raw, two-hour transcript into a ChatGPT window and asking for “the top five themes” yields generic, surface-level summaries. It strips away the nuance and context that makes qualitative research valuable in the first place.</p><p>UX research inherently requires a human, empathetic lens to transform raw data into actual meaning. We cannot treat complex analytical processes as single-step prompts. Recent work on agentic workflows, however, offers a promising way to augment this interpretive process without replacing it.</p><p>Agentic workflows allow researchers to scale qualitative efforts. Imagine analyzing 1,000 interviews instead of 10, without sacrificing the hermeneutic rigor of manual analysis. It shifts the AI from an untrustworthy, autonomous oracle into a verifiable mechanical extractor. The human researcher remains firmly in the seat of interpretation.</p><h3>The Spectrum of Agentic Assistance</h3><p>AI adds the most immediate value when it tackles the rote, high-friction aspects of the research lifecycle.</p><ul><li><strong>Participant Recruitment:</strong> Automating screener logic to verify free-text responses against complex persona profiles.</li><li><strong>Transcription Formatting: </strong>Cleaning up raw audio text and perfectly attributing quotes to the correct speakers.</li><li><strong>Longitudinal Tracking:</strong> Parsing weeks of daily diary entries to track shifting emotional trajectories over time.</li><li><strong>App Review Scraping:</strong> Structuring thousands of unstructured app store reviews into categorized feature requests.</li><li><strong>Survey Data Standardization: </strong>Cleaning massive datasets of free-text responses (e.g., standardizing 50 spellings of “Software Engineer”).</li><li><strong>Heuristic Parsing: </strong>Sorting raw observation notes from a usability test into recognized usability heuristics.</li><li><strong>Data Analysis:</strong> The most fraught area. Models can tag data and extract quotes, but the researcher must dictate the codebook to scale interpretation rather than replace it.</li></ul><h3>What Others Are Trying: A Literature Review of Current Practices</h3><p>The research community is shifting away from simple text analysis toward structured, hybrid pipelines. The literature heavily favors human-in-the-loop workflows over autonomous sensemaking. Researchers recognize that rigorous qualitative data analysis requires a transparent chain of reasoning.</p><p>The authors of “<a href="https://link.springer.com/article/10.1140/epjds/s13688-025-00548-8">Scaling hermeneutics: a guide to qualitative coding with LLMs for reflexive content analysis</a>” offer a practical strategy for maintaining rigor with AI tooling. The researchers detailed how they maintain qualitative validity by retaining human-led codebook development. They adapt traditional definitions specifically for machine comprehension. They use explicit rules and constraints to guide the LLM. The models surface patterns across massive datasets, but the human retains the final say on what those patterns actually mean.</p><p>A similar strategy is demonstrated by the creators of <a href="https://arxiv.org/html/2501.00775v3">MindCoder</a>. They emphasize “efficiency with rigor” through transparent reasoning chains. Their system separates the mechanical extraction of data from the human act of validation. The LLM handles the rote sorting and tagging, while human researchers actively curate and verify the output at every stage.</p><p>The core argument for this rigor relies on establishing an audit trail. These authors prove that by forcing LLMs to provide a rationale before applying a code (Chain-of-Thought), researchers create a verifiable trail. This protects the qualitative integrity of the data. If an AI misinterprets a quote, the researcher can read the explicit rationale and correct the codebook definition.</p><p>We can recreate these workflows locally using tools like Antigravity. Practitioners can build deterministic pipelines that force the LLM to output a rationale step. By separating the coding task from the synthesis task, we ensure the system behaves predictably.</p><h3>Case Study: From Transcripts to Reports</h3><p>The difference between a zero-shot prompt and an agentic workflow becomes clear when you try to build one. I recently designed a set of custom skills for my own qualitative data analysis. The pipeline relies on three specialized agents.</p><p>First, the `qual_coder` skill goes transcript by transcript to code the interviews. It has one job. It reads the raw transcript chunk by chunk and applies specific tags. As it works, it builds up a master codebook artifact. At every point, the skill pauses to ask for my feedback. I adjust what it produces before it moves forward.</p><p>Second, that master codebook artifact is passed to a `thematic_mapper` skill. This agent looks only at the structured codes, not the messy raw transcript. It groups those codes into high-level themes based on parameters I define (e.g., passing in the original study plan and field notes to ensure the themes actually answer the intended research questions). It also requests my input to verify its thematic groupings.</p><p>Finally, the approved themes go to a `report_writer` skill. This agent outlines the themes, pulling in the supporting details and direct evidence. I then use this structured outline to write my final report.</p><p>This deterministic, step-by-step approach is infinitely safer than a generic LLM window. It forces the AI to “show its work.” If the report surfaces a strange insight, I can trace it directly back to the thematic mapper, down to the qual coder, and finally back to the raw transcript. The human remains in the loop, driving the interpretation while benefiting from the automation.</p><p>We can visualize this specific pipeline, and its critical human feedback loops, below:</p><figure><img alt="A flowchart showing a three-stage agentic workflow for UX research. Raw transcripts are processed by a qual_coder agent to extract codes, which undergo human review to become a master codebook. This codebook is then processed by a thematic_mapper agent to group themes, which undergo a second human verification step. Finally, the approved themes are sent to a report_writer agent to generate a final structured outline." src="https://cdn-images-1.medium.com/max/436/1*XFrL6M2-JYo2DeeepUHejg.png" /></figure><p>To make this workflow even more robust, we can integrate strategies from recent literature. The creators of MindCoder recommend introducing explicit “Sensemaking Nudges.” For example, the `qual_coder` could be prompted to identify the codes it is <em>least</em> confident about. This forces the human researcher to focus their attention on the most ambiguous data.</p><p>Additionally, we can enforce a “Memo Writing” step. By having the workflow pause after the `qual_coder` finishes to ask the researcher for a written reflection, we preserve a “Codebook Development Trajectory.” This creates an audit trail that proves exactly how the AI was constrained and how the human led the interpretation.</p><h3>Augmentation, Not Replacement</h3><p>Human intuition cannot be automated. The empathy required to read between the lines of an interview (to notice the hesitation before an answer, or connect an irrelevant anecdote to a broader product strategy) is uniquely human.</p><p>Agentic workflows do not threaten this intuition. They protect it. By offloading the mechanical burden of coding, formatting, and synthesizing, these tools free up mental bandwidth. They give us the space required for deep, meaningful analysis.</p><p>Start by automating your highest-friction step. Build a single, deterministic skill to format transcripts or parse screener surveys. By offloading the mechanical extraction, you preserve the mental bandwidth required for actual hermeneutic analysis.</p><h3>Recommended Reading</h3><ul><li><a href="https://arxiv.org/html/2501.00775v3">Efficiency with Rigor! A Trustworthy LLM-powered Workflow</a> : Provides practical guidelines for separating mechanical data extraction from human validation.</li><li><a href="https://link.springer.com/article/10.1140/epjds/s13688-025-00548-8">Scaling hermeneutics: a guide to qualitative coding with LLMs</a>: <br> Outlines strategies for adapting human codebooks into constraints that maximize LLM performance.</li><li><a href="https://link.springer.com/chapter/10.1007/978-3-032-08872-7_11">Perspectives on LLMs in Management and Qualitative Research</a>: Discusses the transition from basic text analysis tools to sophisticated analytical assistants.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=56b94fbd1ecb" width="1" height="1" alt=""><hr><p><a href="https://medium.com/design-bootcamp/practical-agentic-workflows-for-uxr-56b94fbd1ecb">Practical agentic workflows for UXR</a> was originally published in <a href="https://medium.com/design-bootcamp">Bootcamp</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What feels right, and what feels wrong, to entrust to agentic AI in user research]]></title>
            <link>https://medium.com/design-bootcamp/what-feels-right-and-what-feels-wrong-to-entrust-to-agentic-ai-in-user-research-7e8268a1dda7?source=rss-5e4d9b1fe93------2</link>
            <guid isPermaLink="false">https://medium.com/p/7e8268a1dda7</guid>
            <category><![CDATA[agentic-uxr]]></category>
            <category><![CDATA[agentic-workflow]]></category>
            <category><![CDATA[ux-strategy]]></category>
            <category><![CDATA[ux-research]]></category>
            <category><![CDATA[future-of-work]]></category>
            <dc:creator><![CDATA[Meg Kurdziolek]]></dc:creator>
            <pubDate>Sun, 03 May 2026 00:43:32 GMT</pubDate>
            <atom:updated>2026-05-03T14:53:59.462Z</atom:updated>
            <content:encoded><![CDATA[<h4>Avoid insight washout by drawing the boundaries of delegation</h4><figure><img alt="A frazzled human researcher next to a perfectly organized AI robot" src="https://cdn-images-1.medium.com/max/1024/1*6WwjF_iIj_uZmdeiTCkYxw.png" /></figure><p>During graduate school, I spent months manually coding 100 semi-structured qualitative interviews. I highlighted transcripts until my eyes blurred, creating a dizzying taxonomy of margin notes, codes, themes, and scratch docs synthesizing my thoughts and feelings about the data. It was grueling. It was inefficient. But by the end of that project, I knew those 100 interviews backwards and forwards. I understood the pregnant pauses in their audio recordings. I recognized the unspoken anxieties hiding behind their polite answers.</p><p>Today, if faced with those same 100 interviews, I could hand the entire dataset to an agentic AI. Within seconds, it would spit out a well-formatted thematic analysis, complete with representative quotes and synthesized takeaways. The immediate relief of that automation is undeniable. The creeping sense of loss, however, is much harder to articulate. It’s the tension of wanting to use the powerful new tools at our disposal, while also recognizing that the very process that gives us deep expertise and intimate understanding of our users might be exactly what we’re automating away.</p><p>Agentic AI is fundamentally altering the UX research lifecycle. We are shifting from operating static research tools to delegating complex tasks to autonomous systems. To navigate this transition, we must consciously choose what we delegate. If we fail to draw clear boundaries, we risk automating away the empathy and nuance that makes our craft valuable in the first place.</p><h3>The Era of the UXR Delegator</h3><p>The role of the UX Researcher is undergoing a radical shift. We are no longer just operators. We are delegators managing <a href="https://www.forbes.com/councils/forbestechcouncil/2025/07/10/from-ux-to-ax-the-future-of-agentic-experiences-in-the-age-of-ai/">agentic experiences</a>.</p><p>Automating drudgery provides the immediate upside. Recruiting participants, cleaning transcripts, and pulling baseline synthesis are historically the most time-consuming parts of the job. Furthermore, a massive portion of day-to-day product research isn’t deep, foundational discovery; it is providing quick, tactical answers to unblock product teams during a sprint. Agents crush this fast-paced workload, elevating researchers from data-processors to strategic decision-makers.</p><p>Their sheer efficiency, however, is exactly what can create a false sense of security. Because these tools are so frictionless, the temptation is to use them for <em>everything</em>, and that is where the danger lies.</p><h3>The Danger of the “Insight Washout”</h3><p>The anxiety surrounding AI in research isn’t just about job security; it is about the degradation of data quality. Large Language Models are fundamentally designed to predict the most statistically probable outcome. They find the average.</p><p>The problem is that the most profound UX insights rarely live in the average. They live in the extremes, the weird edge cases, and the emotional friction points. If we overly rely on an AI for the <em>entire</em> qualitative synthesis process, blindly accepting the output rather than inserting human judgment at crucial junctions (as modeled in Brad Orego’s excellent <a href="https://greatquestion.co/ux-research/ai-analysis-synthesis">6-step automated pipeline</a>) we risk washing out the nuance. An AI will give you the most common theme. It will likely miss the quiet, singular user whose bizarre workaround just revealed a million-dollar product pivot.</p><h3>Drawing the Delegation Boundary: Right vs. Wrong</h3><p>To harness the speed of AI without sacrificing the intimacy of human research, we have to draw hard lines around our workflows.</p><h4>What feels RIGHT to entrust:</h4><ul><li><strong>Drafting initial interview scripts or survey structures.</strong> Use AI to overcome the blank-page problem.</li><li><strong>Transcription and raw data cleaning.</strong> Let the machine handle the transcribing, formatting, and cleaning of the data.</li><li><strong>Pattern recognition at scale.</strong> When dealing with massive datasets, agents excel at finding the “what” (e.g., “70% of users clicked the back button here”). When phenomena are easy to identify, agents provide a new capability for identifying those phenomena across larger datasets than humans could feasibly review.</li><li><strong>Creating presentations or reports from data. </strong>Once you’ve done the interpretive heavy lifting, agents can help you visualize and package your findings. For example, you can automatically aggregate representative quotes for a given theme or finding directly into your final report.</li></ul><h4>What feels WRONG to entrust:</h4><ul><li><strong>Initial “margin notes” and coding of qualitative data.</strong> This is where the human researcher exercises the art of careful curation. If you don’t provide that initial layer of human sense-making, you lose your connection to the data.</li><li><strong>Final interpretive synthesis.</strong> Agents can find the “what,” but human researchers must tell the story of the “why.”</li><li><strong>Navigating deeply emotional or sensitive user feedback.</strong> Empathy cannot be simulated by a statistical model.</li><li><strong>Identifying the human intent behind edge cases.</strong> An outlier data point might be a simple error, or it might be the most important discovery of the quarter. Only a human researcher has the context and intuition to investigate the “why” behind the weirdness.</li><li><strong>Guiding product teams through critical strategic conversations.</strong> The true craft of UXR often happens *after* the research is done. Presenting data, reading the room, and managing stakeholder politics to drive product decisions is a deeply human discipline that no AI can replicate.</li></ul><h3>Preserving Human Intuition</h3><p>The future of UXR isn’t fighting the agents. It is learning how to co-pilot with them while fiercely guarding the intimacy of our data. We must learn to leverage automation to buy back our time, and spend that time diving deeper into the human experience.</p><p>Agents can process data at the speed of light, but only humans can truly understand what it means to be human.</p><h3>Recommended Reading</h3><p>If you’re grappling with how to integrate agents into your workflows without losing the human touch, I highly recommend checking out these pieces:</p><ul><li><a href="https://medium.com/uxr-microsoft/why-ai-agents-should-change-ux-research-837bc6a2a488">Why AI Agents (Should) Change UX Research</a> by the Microsoft UXR Team. This is a fantastic foundational argument for how agents will alter the research lifecycle and why we should embrace the shift from operators to delegators.</li><li><a href="https://greatquestion.co/ux-research/ai-analysis-synthesis">AI Analysis &amp; Synthesis for UX Research: A 6-Step Pipeline</a> by Brad Orego. A practical look at how teams are successfully automating the coding and theming of transcripts, while clearly identifying the crucial points where human oversight is still absolutely necessary.</li><li><a href="https://www.philipburgess.net/post/agentic-ai-the-future-of-ux-research-workflows">Agentic AI: The Future of UX Research Workflows</a> by Philip Burgess. A deep dive into how smarter automation and autonomy can provide decision support, elevating researchers to focus on high-impact insights rather than repetitive tasks.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7e8268a1dda7" width="1" height="1" alt=""><hr><p><a href="https://medium.com/design-bootcamp/what-feels-right-and-what-feels-wrong-to-entrust-to-agentic-ai-in-user-research-7e8268a1dda7">What feels right, and what feels wrong, to entrust to agentic AI in user research</a> was originally published in <a href="https://medium.com/design-bootcamp">Bootcamp</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[My mid-life crisis list]]></title>
            <link>https://medium.com/@megak/my-mid-life-crisis-list-aa568c581285?source=rss-5e4d9b1fe93------2</link>
            <guid isPermaLink="false">https://medium.com/p/aa568c581285</guid>
            <category><![CDATA[adventure]]></category>
            <category><![CDATA[goals-in-life]]></category>
            <category><![CDATA[midlife]]></category>
            <dc:creator><![CDATA[Meg Kurdziolek]]></dc:creator>
            <pubDate>Fri, 09 Feb 2024 15:50:25 GMT</pubDate>
            <atom:updated>2024-02-09T21:33:35.967Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HsyTGkEBHoIKwKChen1jgg.jpeg" /><figcaption>The midlife crisis list</figcaption></figure><p>2023 did not start well. Just before the new year I had lost my grandmother, and the grief was still fresh. Then a couple weeks into January I was laid off from the job I loved. And a few days after that the cat I had for 17 years passed away. By the time I started February I was a mess. Somehow I managed to interview and land a new job by March. Then April was spent trying to get my feet underneath me again. But then in May I had my 40th birthday. It took me by surprise.</p><p>As the rest of 2023 progressed I spent more and more time thinking about what it meant to be 40. I spent a lot of time thinking about the life I led to this point and what I would do differently if I could go back.</p><p>The one realization I came to was that I had spent a good bit of my 30’s waiting. I was waiting for my kids to get older. I was waiting for my body to go back to some pre-baby state. I was waiting for the pandemic to be over. I was waiting for my husband’s health to be better. I was waiting for work to be easier. I was waiting for the moment to be just right before I did the things I really wanted to do. I decided that in my 40’s I would stop waiting. My 40’s are going to be all about the doing.</p><p>So for 2024 I made a list. Its not a bucket list of things I want to do before I die, its a list of things I want to try to do <em>this year</em>. I do not believe I’ll be able to accomplish all the things on the list. It is aspirational. But the point is to stretch myself every moment I can spare to do things I’ve always wanted to. I’ve been jokingly calling it my mid-life crisis list. Here it is:</p><ol><li>Get a tattoo</li><li>Write a book</li><li>Become fluent in Spanish</li><li>Go backpacking</li><li>See the northern lights</li><li>Go stargazing</li><li>Go on a night hike</li><li>Go snowshoeing</li><li>Travel to Hawaii</li><li>Adopt 2 kittens</li><li>Go on a friends camping trip</li><li>Go on a family camping trip</li><li>Take Anna to a fancy tea party</li><li>Take Tommy to an arcade</li><li>Make a perfect margarita</li><li>Host a party for some of my favorite people</li><li>Celebrate something with a champagne shower</li><li>Plant an epic summer garden</li><li>Improve my 200m swim time</li><li>Swim over 100 miles throughout the year</li><li>Go kayaking</li><li>Catch a big fish</li><li>Go on a few bike rides</li><li>Go on monthly dates</li><li>Go to a hockey game</li><li>Go to a baseball game</li><li>Take the family to Niagara falls</li><li>Travel to Puerto Rico</li><li>Complete a triathlon</li><li>Podium at a race</li><li>Go to breakfast with a friend once a month</li><li>Smoke a high quality cigar</li><li>Drink some high quality bourbon</li><li>Drink a Buena Vista Irish coffee</li><li><em>….I’m still adding to the list whenever a new thought hits me.</em></li></ol><p>What’s been funny to me is that the more people I tell about the list the more people want to join in. I’ve already done a few of the things on this list and so far it has always been with a friend. ❤ If you’ve got more ideas for my list, I want to hear them. And if you’ve also found yourself in the waiting trap, I encourage you to make your own!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=aa568c581285" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Human + Machine = Something new]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/the-ux-book-club/human-machine-something-new-f81f04a80ed0?source=rss-5e4d9b1fe93------2"><img src="https://cdn-images-1.medium.com/max/750/1*vVqMqYXEvr566l_5qsg0OA.jpeg" width="750"></a></p><p class="medium-feed-snippet">Review of &#x201C;Human-Machine Reconfigurations&#x201D; by Lucy Suchman</p><p class="medium-feed-link"><a href="https://medium.com/the-ux-book-club/human-machine-something-new-f81f04a80ed0?source=rss-5e4d9b1fe93------2">Continue reading on The UX Book Club »</a></p></div>]]></description>
            <link>https://medium.com/the-ux-book-club/human-machine-something-new-f81f04a80ed0?source=rss-5e4d9b1fe93------2</link>
            <guid isPermaLink="false">https://medium.com/p/f81f04a80ed0</guid>
            <category><![CDATA[thumbs-up]]></category>
            <category><![CDATA[ux]]></category>
            <category><![CDATA[machine]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[hci]]></category>
            <dc:creator><![CDATA[Meg Kurdziolek]]></dc:creator>
            <pubDate>Mon, 17 Jun 2019 16:36:00 GMT</pubDate>
            <atom:updated>2019-06-17T16:36:00.860Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[The UX Research FAQ]]></title>
            <link>https://medium.com/@megak/the-ux-research-faq-7ba655d9f6eb?source=rss-5e4d9b1fe93------2</link>
            <guid isPermaLink="false">https://medium.com/p/7ba655d9f6eb</guid>
            <category><![CDATA[research]]></category>
            <category><![CDATA[product-design]]></category>
            <category><![CDATA[ux]]></category>
            <category><![CDATA[ux-research]]></category>
            <category><![CDATA[design]]></category>
            <dc:creator><![CDATA[Meg Kurdziolek]]></dc:creator>
            <pubDate>Fri, 02 Nov 2018 17:03:41 GMT</pubDate>
            <atom:updated>2019-02-06T15:18:17.214Z</atom:updated>
            <content:encoded><![CDATA[<p>Sometime people ask me questions about UX Research, and sometimes I’m able to give them good answers on-the-fly. For all other times, I’ve created this FAQ.</p><h4>Q: Gathering feedback seems time consuming and a waste. People don’t know what they want anyways — didn’t Steve Jobs say that?</h4><p>There is a big difference between running a useless survey and actually putting in the work of gathering feedback on a design or new tool.</p><p>At the absolute minimum, gathering feedback can help you from making mistakes like this…</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/308/1*XZceBUGNL8DU10oc8AOwCQ.jpeg" /><figcaption>That looks like it hurts</figcaption></figure><p>Or this…</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/406/1*FxX9FchuLz1KAwV9xiTSpg.jpeg" /><figcaption>Fashion Fart</figcaption></figure><p>And if you have a UX Researcher helping you out, they will be thinking about the longevity of the product and prevent you from making mistakes like this…</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/680/1*JEnFyO92o50SkQuRZ_Jkxg.jpeg" /></figure><p>Lastly, if you invoke Steve Jobs to justify your reluctance to do research then you are just trying to get out of doing work in the lamest way possible. (And you should read this: <a href="https://medium.com/@mktgwithmeaning/that-steve-jobs-research-quote-should-rip-e8f3335ec66">That Steve Jobs Research Quote Should RIP</a>)</p><h4>Q: Doesn’t research cost lots of money?</h4><p>No. UX research actually<em> saves</em> your company tons of money.</p><p>It is way more costly to spend engineering time and resources on an idea that is flawed than to spend time in research getting the details right first.</p><h4>Q: What is the best organization structure for UX Research on product teams? Who should they report to?</h4><p>I’ve seen the org-chart for UX Research done one of two ways:</p><ol><li>Internal research consultancy — a small team of researchers report to a “lead researcher”. They receive research requests from all different parts of the company. They prioritize those requests and assign researchers as jobs start and end. Researchers get a wide-variety of experiences working on a wide-variety of products.</li><li>Embedded on product teams — each researcher is assigned to a single product team. They become experts on their product and target customers. They report to the team product manager. (But may also have a “dotted line” research manager.)</li></ol><p>Typically I see “structure one” at small companies that have few resources and/or not enough work on one product to occupy a researcher full-time. I currently work at a company that operates like this. There are only a handful of researchers and we take on “jobs” from all different parts of the company. Our “customers” include sales, operations, engineering, and product.</p><p>“Structure two” is what you see at bigger companies, like Google and Amazon. Researchers are deeply embedded in a particular product domain. They may collaborate and share ideas with other researchers, but typically their day-to-day interactions are with the product managers and engineers on their team.</p><p>(Have you seen any different kind of structures for UX Research? I’d love to hear about it!)</p><h4>Q: Don’t you need a control group and a treatment group and then run a statistical test to really get a valid answer?</h4><p><strong>The shortest answer:</strong> No. You don’t.</p><p><strong>The slightly longer answer:</strong> Quantitative research methods are appropriate and valid in some situations. Qualitative research methods are valid in others. Good UX researchers have studied and trained in a variety of different methodologies (quantitative and qualitative) and are skilled at selecting the right method for the particular research question.</p><p><strong>The longest answer:</strong> Look, I know that you took science classes in high-school that have biased you into thinking that science is <em>really</em> only done when you have a hypothesis and an A/B test. I also know that they have biased you into thinking that looking at “anecdotal evidence” is a cardinal sin. And to you, qualitative research looks <em>a lot</em> like anecdotes and this is giving you the heebie-jeebies. Let me put you at ease.</p><p>First, a point: the foundation of <em>all</em> science is observation and documentation. The Theory of Evolution is based on observation and documentation. The Law of Gravity grew from initial observations, which were documented. Everything we know about animal behavior, from their migration patterns to mating behaviors, is based on observation and documentation. Science = observation and documentation. So whenever I do anything in my job, I always observe and document. It is the absolute minimum requirement to doing science.</p><p>Even in “quantitative research” you observe the experiment and document everything you see. This is because the world is inherently uncontrollable (especially when you deal with complex animals like humans). When you observe and document in a quantitative study, you are looking for a couple things, like threats to the validity of the study, and further explanation as to why the treatment worked (or didn’t work as the case may be).</p><p>When you observe and document in a qualitative studies, on the other hand, you are uncovering “themes.” Themes are, in a nut-shell, the most common behaviors you see when you observe people interacting in a specific scenario. For example, when you conduct a usability test you are trying to document the most-common behaviors, problems, and interaction flows participants have with a specific interface. There is a lot of history, science, and tradition within the fields of anthropology and sociology that UX Researchers rely on to conduct these kind of tests.</p><h3>Add more to my FAQ!</h3><p>If you have questions about UX Research — ask them and I’ll try to answer.</p><p>If you are a UX Researcher, what questions do you routinely get asked?</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7ba655d9f6eb" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[3 ways to validate your startup idea without building a damn thing]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@megak/3-ways-to-validate-your-startup-idea-without-building-a-damn-thing-ea30b8635efc?source=rss-5e4d9b1fe93------2"><img src="https://cdn-images-1.medium.com/max/1800/1*DjJb1__PA6hbCaisz-YZYQ.png" width="1800"></a></p><p class="medium-feed-snippet">An article from the User-Centered Startup blog</p><p class="medium-feed-link"><a href="https://medium.com/@megak/3-ways-to-validate-your-startup-idea-without-building-a-damn-thing-ea30b8635efc?source=rss-5e4d9b1fe93------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@megak/3-ways-to-validate-your-startup-idea-without-building-a-damn-thing-ea30b8635efc?source=rss-5e4d9b1fe93------2</link>
            <guid isPermaLink="false">https://medium.com/p/ea30b8635efc</guid>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[design]]></category>
            <category><![CDATA[mvp]]></category>
            <category><![CDATA[ux]]></category>
            <category><![CDATA[startup-life]]></category>
            <dc:creator><![CDATA[Meg Kurdziolek]]></dc:creator>
            <pubDate>Thu, 21 Sep 2017 13:44:13 GMT</pubDate>
            <atom:updated>2019-06-11T12:58:56.062Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[The MVE: Minimum Viable Experiment]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@megak/the-mve-minimum-viable-experiment-b2498fda9f93?source=rss-5e4d9b1fe93------2"><img src="https://cdn-images-1.medium.com/max/1024/1*WTwYKWN24GgrAif22iuXRw.png" width="1024"></a></p><p class="medium-feed-snippet">Entrepreneurs would find a lot more freedom to explore and grow their ideas if they focussed less on MVP, and more on MVE.</p><p class="medium-feed-link"><a href="https://medium.com/@megak/the-mve-minimum-viable-experiment-b2498fda9f93?source=rss-5e4d9b1fe93------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@megak/the-mve-minimum-viable-experiment-b2498fda9f93?source=rss-5e4d9b1fe93------2</link>
            <guid isPermaLink="false">https://medium.com/p/b2498fda9f93</guid>
            <category><![CDATA[mvp]]></category>
            <category><![CDATA[idea-validation]]></category>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[minimum-viable-product]]></category>
            <category><![CDATA[lean-startup]]></category>
            <dc:creator><![CDATA[Meg Kurdziolek]]></dc:creator>
            <pubDate>Wed, 31 May 2017 15:41:00 GMT</pubDate>
            <atom:updated>2019-06-11T14:44:24.213Z</atom:updated>
        </item>
    </channel>
</rss>