<?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 Anka Tomaka on Medium]]></title>
        <description><![CDATA[Stories by Anka Tomaka on Medium]]></description>
        <link>https://medium.com/@ankatomaka?source=rss-cc6b0024f17c------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*MuX9h6CPl92VYaEWmi0EkQ.jpeg</url>
            <title>Stories by Anka Tomaka on Medium</title>
            <link>https://medium.com/@ankatomaka?source=rss-cc6b0024f17c------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 20:29:18 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@ankatomaka/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[AI didn’t speed up my discovery]]></title>
            <link>https://medium.com/design-bootcamp/ai-didnt-speed-up-my-discovery-7a15e58c58f4?source=rss-cc6b0024f17c------2</link>
            <guid isPermaLink="false">https://medium.com/p/7a15e58c58f4</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[design-ethics]]></category>
            <category><![CDATA[product-discovery]]></category>
            <category><![CDATA[slow-design]]></category>
            <category><![CDATA[ux-desig]]></category>
            <dc:creator><![CDATA[Anka Tomaka]]></dc:creator>
            <pubDate>Mon, 18 May 2026 12:42:31 GMT</pubDate>
            <atom:updated>2026-05-18T12:42:31.147Z</atom:updated>
            <content:encoded><![CDATA[<h4>It bought me time to slow down</h4><p><em>On AI product discovery for complex visual tools. And what slow design means now.</em></p><p>Slow design, in the AI era, is what you do with the time AI gives back. AI product discovery doesn’t speed up the hard parts of design. It changes which parts are hard.</p><figure><img alt="InsureFlow reviewer queue mockup generated by Google AI Studio. A black AI EVALUATION panel displays TRUST SCORE 98% in cyan. A Verification Matrix below marks Identity, Policy Coverage, and Compliance as OK. A large APPROVE TASK button dominates the screen. A SELECT toggle at the top of the queue enables bulk approval." src="https://cdn-images-1.medium.com/max/1024/1*j-Yx4LEKURS9XGKtx6wieg.png" /><figcaption><em>InsureFlow mockup — what AI Studio proposed when I asked for a reviewer queue.</em></figcaption></figure><p>I asked Google AI Studio to draft a queue for claim and policy reviewers on an insurance core platform. It returned a screen built around a single number. TRUST SCORE 98%, cyan, in a black panel labelled AI EVALUATION, with a verification matrix below it marking Identity, Policy Coverage, and Compliance as OK. A large APPROVE TASK as the primary action. A SELECT toggle at the top of the queue for bulk approval. The screen had other things on it, too. Tabs, side panels, secondary actions. None of them mattered. The visual hierarchy answered the question before the reviewer asked it. The pattern was fluent. It was also a rubber-stamping nudge built on the average case, served on a screen whose entire job is to review the cases that are not average.</p><p>Discovery is my favourite phase of any project. I expected AI to make it faster. It did something else. <a href="https://findingourway.design/2025/08/08/61-the-paradoxes-of-product-discovery-ft-teresa-torres">AI absorbs the laborious parts of discovery</a> and hands me back the time, and the harder work lives in the time AI gave me back.</p><h4><strong>What AI Product Discovery Actually Gives Me</strong></h4><p>For over four years, I have been running discovery workshops at Synergy Codes, designing diagramming and workflow tools for regulated industries like insurance, energy, and industrial automation. The clients have not changed much. The toolkit has. Over the past year, AI stepped into the discovery phase, and two things have changed about how I do it.</p><p><strong>It shortens the distance to the domain and takes over the boring parts.</strong></p><p>Reading my way into insurance regulations, industrial control system documentation, or AV signal flow specifications used to take weeks. Now it takes hours. The information-gathering, synthesis, and first-pass analysis that used to consume a third of my time and produce nothing memorable are now automated. AI is good at this work.</p><p><strong>It makes the visual layer of discovery cheap.</strong></p><p>There is a thing that happens in domain conversations. We talk about something, agree we understand it, and proceed. Then I show a sketch and the client says, “No, that is not what I meant.” The talking did not catch the gap. The visual did.</p><p>AI lets me put a draft visual in front of a power user in the time it takes to have a conversation about it. Half-formed. Sometimes wrong on purpose. The draft becomes the thing we discuss. The client, who is the actual expert, validates or breaks it. The expensive part of the design moves earlier, into the cheap part.</p><p>On the same project, I showed an AI-generated flow to the Product Owner. He flagged a bulk-edit pattern in seconds. The rule had never been articulated because in the domain, it was settled, not an open question. Then his team walked through the same flow end-to-end and named gaps in their own thinking that the project had been carrying silently. The talking version of discovery would not have surfaced any of it. The visual one did.</p><h4><strong>The Slow Design Paradox</strong></h4><p>This is what I expected, walking into AI as a working designer.</p><p>Discovery would get faster. The six weeks I usually spend would be compressed to two or three. My UX research process would shrink. I would save time. I would ship faster.</p><p>This is what actually happened.</p><p>Discovery did not get faster. And I do not want it to. The boring parts of discovery got automated, and the time that opened up went somewhere I did not expect. It went into decisions.</p><p>This is the paradox I have been trying to articulate to people who ask me about AI for UX designers.</p><p>Information gathering, synthesis, analysis. These are not the hard parts of design. They are the laborious parts. AI takes them. Good. The hard parts are decisions. Which pattern fits this domain, and where does it stop fitting? Which assumption can ride and which one needs to be killed before it gets into the build? What we ship and what we deliberately do not ship. These decisions take time. They take attention. They take a stop.</p><blockquote>AI does not decide. AI proposes. The decision is mine.</blockquote><p>That same project came with a phrase that read like a brief: “AI-driven automation with human-in-the-loop.” Nobody had stress-tested what “in the loop” meant once written into a UI. At least four interpretations were live: a final approval gate, an inline suggestion to accept or edit, a batch audit on a sample, or a manual trigger that called AI on demand. Each implied different screens, different permissions, and different audit trails. The decisions were mine because no one else’s job was to make them. We landed on an inline suggestion the reviewer could accept or edit, with a separate audit trail for the cases where AI was overridden. Two weeks to get there. None of those weeks looked like discovery.</p><p>The time AI returns to me is exactly the time I now have to spend deciding. With more information, more hypotheses to validate, more directions to reject, the cognitive load of decision making goes up, not down. AI does not decide. AI proposes. The decision is mine, and it should be mine, because I am the one accountable for what gets built.</p><p>So we work faster and slower at the same time. Faster in handling information. Slower, deliberately, in deciding.</p><p>This is what <a href="https://static1.squarespace.com/static/52dfe66be4b0cad36168429a/t/53499733e4b09ac51176571d/1397331763401/CtC_SlowDesignPrinciples.pdf">slow design</a> means in the AI era. Not nostalgia for older tools. Not a craft revival. A conscious choice about where to spend the time you just got back.</p><p>If you do not make that choice, AI will fill the time for you. It will generate more options. It will draft more variants. It will produce more documentation. The boring will expand to fill the space the boring used to occupy. Nothing will have changed.</p><h4><strong>The Invisible Women Problem</strong></h4><p>There is one more thing AI changes about discovery, and it cuts in the opposite direction.</p><p>Caroline Criado Perez’s <em>Invisible Women</em> <a href="https://uxdesign.cc/invisible-women-when-you-only-design-for-50-of-the-population-6738f5b60b88">documents how products designed for the “universal” user end up designed for an average man</a>, and how the rest of the population pays the cost. Crash test dummies. Voice recognition. Drug dosages. Public transit schedules. The average user is a fiction, and the fiction is expensive.</p><p>AI proposes patterns. The patterns are averaged. Trained on what was already published, already shipped, already considered normal. When AI suggests a workflow pattern, what it is really suggesting is the pattern most documented in its training data. That pattern may be dominant for reasons that have nothing to do with whether it is right for the underwriter in the room with you.</p><p>Go back to the screen I opened with. The 98% trust score in cyan, the verification matrix marking everything OK, the bulk SELECT toggle, the APPROVE TASK that was wanted to be tapped. That is exactly the average answer.</p><p>Without a deliberate practice of validating against the real user in the real domain, AI becomes an accelerator for design bias. It makes the average answer arrive faster. It makes the average answer feel more authoritative because it is polished. It hides the question that should have been asked.</p><p>The question to ask, every time AI hands you a fluent pattern: who is sitting in front of this queue? If you can’t see that person, the averaged pattern wins by forfeit. Pattern recognition is half the work. The other half is knowing which patterns are local and which ones are universal. AI is fluent in patterns. It is not fluent in your client’s underwriter.</p><p>Deyan Sudjic’s <em>The Language of Things</em> argues that designed objects speak. Whose voice does an average AI-generated screen speak in? The voice of the most common process in its training data. Not the voice of the underwriter in your client’s office. Without a designer choosing, AI picks the voice by default.</p><p>Look at the trust score again. 98%, cyan, in a panel labelled AI EVALUATION. That number speaks the voice of a credit score, of a heart rate readout, of a search engine confidence rating. A measurement voice. It tells the underwriter: this is not your judgment, this is a fact. The underwriter’s own voice would sound different. This case has these flags. This paperwork is missing. This looks like the one we saw in November. Two languages, two decisions. The screen picks one.</p><p>On the project that opened this piece, we replaced the trust score with an AI insights panel that recommended where to look without concluding. Editable fields. Three action paths instead of one primary CTA. Same data, different voice.</p><p>In practice, slow design means two things. Ask “what am I rejecting?” on every AI output before asking “what am I changing?”. And go to the client, the team, and the technological constraints for the decisions AI cannot make.</p><p>AI does not fix any of this. What AI gives the designer is time. If you don’t choose how to spend it, AI just speeds up the gap-filling we were already doing badly.</p><p><em>Disclosure: I used AI to help structure earlier drafts of this piece. The final text, examples, and arguments are mine.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7a15e58c58f4" width="1" height="1" alt=""><hr><p><a href="https://medium.com/design-bootcamp/ai-didnt-speed-up-my-discovery-7a15e58c58f4">AI didn’t speed up my discovery</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>
    </channel>
</rss>