<?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 Design Nomad on Medium]]></title>
        <description><![CDATA[Stories by Design Nomad on Medium]]></description>
        <link>https://medium.com/@ganesh_govalkar?source=rss-1a15a3cfe255------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*jFeAkqyjqWAAA62CJ7XReQ.jpeg</url>
            <title>Stories by Design Nomad on Medium</title>
            <link>https://medium.com/@ganesh_govalkar?source=rss-1a15a3cfe255------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 11:19:05 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@ganesh_govalkar/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[When the Canvas Becomes the Chat]]></title>
            <link>https://medium.com/@ganesh_govalkar/when-the-canvas-becomes-the-chat-99e02278b4ad?source=rss-1a15a3cfe255------2</link>
            <guid isPermaLink="false">https://medium.com/p/99e02278b4ad</guid>
            <category><![CDATA[ux-design]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[figma]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[product-design]]></category>
            <dc:creator><![CDATA[Design Nomad]]></dc:creator>
            <pubDate>Thu, 19 Nov 2026 06:21:01 GMT</pubDate>
            <atom:updated>2026-11-19T06:21:01.597Z</atom:updated>
            <content:encoded><![CDATA[<p><em>What if AI stopped living in a side panel and joined you on the board instead?</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*l3apCGlCH6QZKCbAchDV4A.png" /></figure><p>This morning, I opened a design tool and saw the same thing I see everywhere.</p><p>A small chat box on the side. Waiting for me to type.</p><p>Then I read a piece by Monika Pawlak, a designer working on AI at Miro. She asked a simple question. What if AI was not a chat at all?</p><h3>The chat box problem</h3><p>Most AI products look the same today. A prompt box, a side panel, and a conversation that scrolls up.</p><p>Miro is a big shared whiteboard. People go there to draw ideas and work together. Not to chat.</p><p>So the team tested a chat panel that several people could share. Users were not excited. One said, “I don’t come to Miro to chat.”</p><p>That told them a lot. A side panel sits next to the work, not on it. And your team can’t see it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*GZFMtSSf3PwLtupOcjkPbA.png" /></figure><h3>Bringing AI to where people already talk</h3><p>Next, they asked a better question. Where do people already talk in Miro?</p><p>In comments. Comments are already shared, and they are pinned to the thing you are discussing. So the AI gets the context for free.</p><p>They also tried “cursor chat”. You type a short line next to your mouse pointer, and it fades away. A small, fast model decides if you are talking to the AI.</p><p>I liked one detail. On a low score, the agent stays quiet. Above about a third, it offers to help. It only acts when the model is very sure, around 0.86.</p><p>Knowing when to stay quiet is good design.</p><h3>A board where agents work beside you</h3><p>The boldest test was called Cursor Town. There is no chat at all.</p><p>You write a goal on the board. A team of agents breaks it down and works on it. They show their thinking as sticky notes and drawings, not walls of text. When they need you, they ask on the board too.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*L_TFUOaJXZTdBrzk_1BbaQ.png" /></figure><p>Monika is honest that these last two ideas have not been tested with users yet. They might turn out to be wrong. That is fine. The point is to learn quickly.</p><h3>What I am taking away</h3><p>The big idea is simple. Don’t copy the chat box just because everyone else does. Make AI fit the way your product already works.</p><p>For a shared canvas, the canvas itself can be the prompt and the answer.</p><p>I keep wondering what this means for the tools I use every day.</p><p>Where do your users already talk? Could your AI live there?</p><p><strong>Source:</strong> <a href="https://monika.fyi/writing/agents-as-collaborators-not-chatbots">Agents as collaborators, not chatbots (Monika Pawlak, 16 Nov 2026)</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=99e02278b4ad" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Button That Turns AI Off]]></title>
            <link>https://medium.com/@ganesh_govalkar/the-button-that-turns-ai-off-8f93d32a1fef?source=rss-1a15a3cfe255------2</link>
            <guid isPermaLink="false">https://medium.com/p/8f93d32a1fef</guid>
            <category><![CDATA[ui]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[design]]></category>
            <category><![CDATA[ux-design]]></category>
            <dc:creator><![CDATA[Design Nomad]]></dc:creator>
            <pubDate>Wed, 18 Nov 2026 08:48:21 GMT</pubDate>
            <atom:updated>2026-11-18T08:48:21.222Z</atom:updated>
            <content:encoded><![CDATA[<p>An email tool just shipped a single switch that removes AI from your account. I think more products should.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*H25mdYeg_XyLeVukYKu8_w.png" /></figure><p>Yesterday, I read a product launch that made me smile. Not because of a new AI feature. Because of a way to turn AI off.</p><h3>What happened</h3><p>Flodesk is an email marketing tool used by small businesses and creators.</p><p>In June, it launched an AI design app called Studio. It kept Studio separate on purpose, so it could prove itself first.</p><p>Now Studio lives inside the main product. You can start an email from a sentence, a template, or a blank page.</p><p>Flodesk says Studio cut the average time to make an email from two hours to 18 minutes. It also says emails needed a median of two prompts before they were ready.</p><h3>Chat or hands, your pick</h3><p>The part I like most is how the two ways of working sit together.</p><p>You can describe an email in chat. Then you can switch to the manual builder to finish it.</p><p>You can drag and drop blocks. Or you can type a change, like a new layout or colour, and the builder makes the edit.</p><p>AI gets you most of the way. Your hands do the last part.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NqNQkWgf8nTZmGnOyg4QdQ.png" /></figure><h3>One switch, all of it</h3><p>Then there is the kill switch.</p><p>It is one toggle in settings. Flodesk says it turns off AI across the whole account. That includes writing emails for you, subject line ideas and text refining.</p><p>What is left is the full builder, in your hands.</p><p>Flodesk says most companies have refused to build this, or their off switch does not fully turn AI off. I can’t check that claim. But I have rarely seen an “off” that means everything.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yvX22zoDwl5-nJkLMOiXcg.png" /></figure><p>Why this matters for designers</p><p>Many of us add AI features because we can. Fewer of us design the way out.</p><p>Some people don’t want AI writing in their voice. Some have work rules. Some just like doing it themselves.</p><p>A clear off switch respects all of them. It also makes the product honest. If the AI is good, people will turn it back on.</p><p>It also asks a hard question of our designs. Does the product still work well without the AI? If not, the AI was hiding a weak core.</p><p>So here are my questions for you. Could your product work with every AI feature switched off? And where would that switch live?</p><p><strong>Source:</strong> <a href="https://www.prnewswire.com/news-releases/flodesk-brings-studio-inside-its-platform-with-a-whole-new-way-to-create-emails-302941983.html">Flodesk Brings Studio Inside Its Platform With a Whole New Way to Create Emails (PR Newswire, 17 November 2026)</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8f93d32a1fef" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[When Your AI Agent Wants to Pay]]></title>
            <link>https://medium.com/@ganesh_govalkar/when-your-ai-agent-wants-to-pay-17e526a581bc?source=rss-1a15a3cfe255------2</link>
            <guid isPermaLink="false">https://medium.com/p/17e526a581bc</guid>
            <category><![CDATA[fintech]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[product-design]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[ux-design]]></category>
            <dc:creator><![CDATA[Design Nomad]]></dc:creator>
            <pubDate>Tue, 17 Nov 2026 05:16:52 GMT</pubDate>
            <atom:updated>2026-11-17T05:16:52.234Z</atom:updated>
            <content:encoded><![CDATA[<p>ChatGPT Wallet lets an AI agent buy things for you, but it still has to ask you first, every single time.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BOABl0Hii6LJrxOHtx1QLQ.png" /></figure><p>I often ask an AI assistant to compare things before I buy them.</p><p>It does the research well. Then, at the payment page, it stops and hands the work back to me.</p><p>I always feel a little relieved. And a little curious. What would it take to let it press “Pay”?</p><p>On 13 November, OpenAI gave one answer. It is called ChatGPT Wallet.</p><h3>What it is</h3><p>ChatGPT Wallet lets an AI agent buy things on your behalf. In ChatGPT, these agents are called “dots”.</p><p>You add a card once. After that, the agent can finish a purchase without you typing your card details again.</p><p>For now, it is only for ChatGPT Pro users in the US, aged 18 or over, with eligible US cards. OpenAI built it with American Express, Mastercard and Visa.</p><h3>The agent never sees your card</h3><p>This is the clever part.</p><p>When you approve a purchase, the payment network creates a token. A token is a stand-in for your real card. It has its own card number, security code and expiry date.</p><p>The shop gets paid, and your card is charged. But the shop and the agent never see your actual card details.</p><p>You also keep your card’s usual rewards, and its protection for eligible disputes, for example when an item arrives faulty.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xGDBgB4pPiGL9AGRYQK4iQ.png" /></figure><h3>You still say yes, every time</h3><p>Before each purchase, you see a review screen. It shows the items, the total and the card you chose.</p><p>Then you approve it with a security check, such as Face ID or a fingerprint.</p><p>Each approval leaves a clear record of what you allowed the agent to buy.</p><p>OpenAI says that, over time, people may be able to set rules instead, like a spending limit, a type of purchase or a time period. That part is not here yet.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SXzaMPrQDs2fpmQEKPIttg.png" /></figure><h3>Why this matters to designers</h3><p>Most of us have designed a checkout. Very few of us have designed one where the shopper is not a person.</p><p>The most important screen is now the review step. It has one job: show exactly what is about to happen, in a way you can read in a few seconds.</p><p>Shops have work to do too. OpenAI says some websites may block these agents by mistake, so their systems need to learn the difference.</p><p>I am not sure yet if I would trust an agent with my card. But I like that the first version keeps a person in the loop.</p><p>Would you let an agent pay for you? And what would the review screen need to show before you said yes?</p><p><strong>Source:</strong> <a href="https://openai.com/index/introducing-chatgpt-wallet/">Introducing ChatGPT Wallet (OpenAI, 13 November 2026)</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=17e526a581bc" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The AI Assistant That Knows When to Stop]]></title>
            <link>https://medium.com/@ganesh_govalkar/the-ai-assistant-that-knows-when-to-stop-ca482a8132f6?source=rss-1a15a3cfe255------2</link>
            <guid isPermaLink="false">https://medium.com/p/ca482a8132f6</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[ux-research]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[product-design]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <dc:creator><![CDATA[Design Nomad]]></dc:creator>
            <pubDate>Sat, 14 Nov 2026 05:46:42 GMT</pubDate>
            <atom:updated>2026-11-14T05:46:42.608Z</atom:updated>
            <content:encoded><![CDATA[<p>A small UK government team designed the hard answers first, and it made their assistant easier to trust.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rL_vrFBuvg4RzH7-YM2gow.png" /></figure><p>This week, I read a short blog post from a UK government team.</p><p>It was about a chat assistant. No big launch. No bold claims.</p><p>And it changed how I think about designing AI.</p><h3>A helper for one clear job</h3><p>Defra is the UK department for the environment, food and farming. Its team built a tool called “Ask the Toolkit”.</p><p>Staff use it to ask questions about using AI at work. It answers only from the toolkit’s own guidance pages.</p><p>Every answer ends with “Guidance used” links. So you can open the page and check it yourself.</p><p>It also reminds you that AI can make mistakes. And it offers a way to report a problem or talk to a person.</p><h3>Checking before showing</h3><p>The team said the worst failure is a wrong answer about a rule.</p><p>AI can sound very sure while being wrong. So they check each answer before anyone sees it.</p><p>They remove any link that does not point to one of their own pages.</p><p>They also check quotes. A quote only appears if its exact words are on the source page. If not, the quote is removed. A plain explanation and a link are shown instead.</p><p>That sometimes makes the answer less impressive. They chose that on purpose.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SiVaiXxC_vB2tmxlOI8Aww.png" /></figure><p>Designing the hard moments first</p><p>This is the part I loved.</p><p>They designed the difficult cases first, because that is where trust is won or lost.</p><p>The assistant has six outcomes, each with its own heading and message: answered, needs more detail, cannot answer, talk to a person, blocked, and error.</p><p>Words mattered too. For blocked questions, they avoid words like “flagged” or “filtered”. To someone new to AI, those can sound like an accusation.</p><p>When an answer fails, the question stays in the box. The failed answer is not added to the chat.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*n0tsT7PaP7IzwdSq-eRVnw.png" /></figure><p>Small details, big trust</p><p>Some of the best changes were tiny.</p><p>The browser tab says “Answer 2 of 3”, so you know where you are. Error titles start with “Error:”, so screen reader users hear the problem first.</p><p>The follow-up box grew from one line to three. A single line quietly says, “keep it short”.</p><p>They even built it with prepared answers before the AI model was ready. That let them test the wording and every outcome early.</p><p>The author calls AI “an assistant, not an autopilot”. I agree.</p><p>As designers, we often polish the happy answer. This team polished the moment the AI has to stop.</p><p>How does your AI feature behave when it does not know? Have you designed that screen yet?</p><p><strong>Source:</strong> <a href="https://defradigital.blog.gov.uk/2026/11/10/designing-an-ai-assistant-people-can-trust/">Designing an AI assistant people can trust, Defra digital, data, technology and security blog (Farhana Abdul Alim, 10 November 2026)</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ca482a8132f6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Screen You Forgot to Design]]></title>
            <link>https://medium.com/@ganesh_govalkar/the-screen-you-forgot-to-design-a13dd7e794c4?source=rss-1a15a3cfe255------2</link>
            <guid isPermaLink="false">https://medium.com/p/a13dd7e794c4</guid>
            <category><![CDATA[ux-research]]></category>
            <category><![CDATA[design-thinking]]></category>
            <category><![CDATA[product-design]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[ux-design]]></category>
            <dc:creator><![CDATA[Design Nomad]]></dc:creator>
            <pubDate>Wed, 11 Nov 2026 03:46:14 GMT</pubDate>
            <atom:updated>2026-11-11T03:46:14.268Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Zeplin’s new AI Flows builds a user journey from your screens, and points out the paths you missed.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*x54UUM6IxlJzeFTLlYtFNg.png" /></figure><p>Last week, a developer asked me a simple question. “What happens if the email link expires?”</p><p>I had no answer. There was no screen for it.</p><p>Every designer knows this moment. The happy path looks perfect. The gaps show up later.</p><p>On 9 November 2026, Zeplin announced a beta feature that tries to catch those gaps earlier. It is called AI Flows.</p><h3>What a flow is</h3><p>Zeplin is a tool designers use to hand finished screens to developers.</p><p>A flow is a map of a user journey. Screens sit on a canvas. Arrows, called connectors, link one screen to the next. Short labels explain what moves the user forward.</p><p>Building one by hand takes a lot of clicking. You drag screens, draw arrows, write labels, then move everything to make room for the path you forgot.</p><h3>You describe, it maps</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jtZWR8VBBQN-eyMDlrBRIQ.png" /></figure><p>With AI Flows, you add your screens to the canvas. Then you write what the journey should be, in plain words.</p><p>Zeplin’s own example is an onboarding flow. Sign up, verify email, create a workspace, invite teammates.</p><p>The AI puts the screens in order, connects them and labels the arrows. It reads what is already in Zeplin, like screen names, layers and components. So it is not guessing from pictures alone.</p><p>Zeplin says a simple flow usually takes 20 to 30 minutes by hand. With AI, you get a first draft in under a minute. You can then edit it as much as you like.</p><h3>The part that surprised me</h3><p>While building the flow, the AI also flags cases you may have missed.</p><p>An error path with no screen. A dead end with no way back. An edge case you mentioned but never designed.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vLyBaTUYBgsZMbZ15nSjew.png" /></figure><p>You decide what to do with each one. Design it now, add a placeholder, or leave it out of this release.</p><p>It also works on old flows. You can ask it to tidy a messy map, or fit a new screen, like “Forgot password”, into the right spot.</p><h3>Why this matters</h3><p>For me, the time saved is nice. The real value is the question it asks.</p><p>It will not replace thinking about users. It can be wrong, and it is still in beta. But a second pair of eyes on the unhappy paths, before handoff, feels genuinely useful.</p><p>Do you check your flows for dead ends before handoff? What is the screen you most often forget?</p><p><strong>Source:</strong> <a href="https://blog.zeplin.io/product-news/ai-flows/">Flows that build themselves (well, almost), Zeplin Gazette, 9 November 2026</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a13dd7e794c4" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Space of Your Own, Built in Chat]]></title>
            <link>https://medium.com/@ganesh_govalkar/a-space-of-your-own-built-in-chat-ec71749d0acf?source=rss-1a15a3cfe255------2</link>
            <guid isPermaLink="false">https://medium.com/p/ec71749d0acf</guid>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[ai-automation]]></category>
            <category><![CDATA[muse]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[meta]]></category>
            <dc:creator><![CDATA[Design Nomad]]></dc:creator>
            <pubDate>Tue, 10 Nov 2026 06:42:45 GMT</pubDate>
            <atom:updated>2026-11-10T06:42:45.887Z</atom:updated>
            <content:encoded><![CDATA[<p>Meta’s Muse app can now turn a short description into a living screen inside the app.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_FTTQGQzxUReQrD7AxHtLg.png" /></figure><p>I keep half-finished lists in my notes app. A gift list. A trip checklist. A page of client deadlines.</p><p>None of them quite deserve their own app. But they also get lost in a long file of notes.</p><p>On 6 November 2026, Meta showed a different path in its Muse app. It is called Spaces.</p><p>You describe an idea. Muse builds a screen for it, inside the app you already use.</p><h3>You start with a sentence</h3><p>A space is a custom experience you create in Muse. The examples in Meta’s post are simple. A holiday gift tracker. A daily word game. A planning board for a DIY project.</p><p>Your first description is only a starting point. You can keep talking to Muse as you use the space. You can add what is missing, try a new look, or change how it works.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qHgVM3CxqQNMuz9Bx94FEQ.png" /></figure><p>You also choose where the space lives. You can add it to the sidebar, move it into the main tabs, or hide it when you do not need it. You can even schedule when it appears.</p><p>Meta gives a clear example. Keep your weekly planner up front during the workweek. Then swap it for your weekend plans on Saturday.</p><h3>It stays with you</h3><p>The space becomes part of Muse. You can come back to it at any time.</p><p>If you connect the apps you already use, the space can update itself in the background. New concert dates, price changes or savings progress can show up without you asking.</p><p>You can check things off, add notes and make edits right in the space. Those changes stay there when you return.</p><p>Meta lists other uses too. A shopping hub. A school dashboard. A music tracker. A wedding planner. A trip planner. Even a client hub for freelance work.</p><p>Sharing with other people is not ready yet. Meta says it is still exploring that.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*r3vhJy0oLoiqhJYCDa0EIw.png" /></figure><h3>What this means for designers</h3><p>For years, a personal tool meant a new app, or a messy notes file.</p><p>Spaces sit in the middle. The person describes a job. A screen is built for that job. Then it can be pinned, hidden or timed, like any other part of the product.</p><p>That is a new design problem. We are no longer only designing one layout for everyone. We are designing a system that can make many small layouts, and a home for them inside the app.</p><p>I would still want to know how clear the first version is, and how easy it is to undo a change. But the idea itself is calm and useful. The screen should fit the person’s day, not the other way around.</p><p>Have you ever wished a product could grow a screen just for your routine? What would you ask it to build first?</p><p><strong>Source:</strong> <a href="https://about.fb.com/news/2026/11/introducing-spaces-customize-muse-app/">Design the Muse App, Just for You (Meta, 6 Nov 2026)</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ec71749d0acf" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[From Paper Sketch to Live App]]></title>
            <link>https://medium.com/@ganesh_govalkar/from-paper-sketch-to-live-app-e75b9d463674?source=rss-1a15a3cfe255------2</link>
            <guid isPermaLink="false">https://medium.com/p/e75b9d463674</guid>
            <category><![CDATA[ux-design]]></category>
            <category><![CDATA[ux-research]]></category>
            <category><![CDATA[product-design]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[design]]></category>
            <dc:creator><![CDATA[Design Nomad]]></dc:creator>
            <pubDate>Mon, 09 Nov 2026 06:13:45 GMT</pubDate>
            <atom:updated>2026-11-09T06:13:45.795Z</atom:updated>
            <content:encoded><![CDATA[<p>Gemini’s Canvas can turn a photo of your drawing into something you can click.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tc-rr0pqh7AOG_lqzsnFFQ.png" /></figure><h3>The three steps</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gPz1rN7PQ62wDEOEBDtaxg.png" /></figure><p>First, you take a photo of your sketch. Google’s post mentions examples like an app mock-up, a habit tracker, or a game board.</p><p>Next, you upload that photo to Gemini and select the Canvas tool. You ask something plain: “Can you code this app?”</p><p>Gemini reads the visual and writes the code. You then test the tool in a preview window. It can adjust colours, fonts, or features while you watch.</p><p>That is the whole flow. No long brief. No separate design file first.</p><h3>Why this matters for designers</h3><p>Many of us still think best with a pen. Whiteboards, sticky notes, a page torn from a notebook.</p><p>Until recently, that sketch was only a handoff note. Someone else had to rebuild it in Figma or in code.</p><p>This path keeps the sketch as the source. You draw first. You prompt second. You refine in the preview.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*h9h4Ryf-AU8fI1REXut2-g.png" /></figure><p>It will not replace careful product design. Layouts still need hierarchy, accessibility, and real user checks. But for a quick idea test, the gap between paper and clickable is much smaller.</p><h3>What I would try next</h3><p>I would start with one clear screen, not a full product. Label the main actions on the paper so the model has less to guess.</p><p>Then I would change one thing at a time in the preview colour, type, or a single feature and see what holds.</p><p>If the result feels close, I would still rebuild the keepers in Figma with proper components. The sketch path is for speed. The design system is for lasting quality.</p><p>Have you tried turning a desk sketch into a live preview yet? What broke first the layout, the copy, or the interactions?</p><p><strong>Source:</strong> <a href="https://blog.google/innovation-and-ai/products/gemini-app/turn-drawings-into-apps-gemini/">Turn sketches into apps with Gemini (Google Keyword, 5 Nov 2026)</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e75b9d463674" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Best Bug Report Is a Screen Recording]]></title>
            <link>https://medium.com/@ganesh_govalkar/the-best-bug-report-is-a-screen-recording-6832feb80135?source=rss-1a15a3cfe255------2</link>
            <guid isPermaLink="false">https://medium.com/p/6832feb80135</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[claude-code]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[ux-design]]></category>
            <dc:creator><![CDATA[Design Nomad]]></dc:creator>
            <pubDate>Wed, 04 Nov 2026 11:05:40 GMT</pubDate>
            <atom:updated>2026-11-04T11:05:40.082Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ezwCFGIAMPoq4Rag7Pz3kA.png" /></figure><p>How one small team fixed a jumpy AI chat screen in two days, starting with a phone video.</p><p>I read a short engineering post that changed how I think about bug reports.</p><p>It came from Ditto, a team that makes an AI chat app. On 1 November, their Android chat screen was hard to use.</p><p>Text jumped while a reply streamed in. The text box jumped when the keyboard opened. An empty gap appeared under the last message. Sometimes, words you had just sent came back into the box.</p><p>Two days later, they shipped a new chat screen.</p><p>What caught my eye was not the speed. It was how they started.</p><h3>They pressed record</h3><p>Instead of writing a bug report, a teammate recorded four minutes of normal use on a real phone.</p><p>Typing. Sending. Scrolling. Opening a menu. Closing the keyboard.</p><p>A written report loses the order of events. A recording keeps it. It showed, for example, that the gap appeared after the keyboard closed, not before.</p><h3>An AI described what it saw</h3><p>Next, they gave the video to an AI model that had never seen their code. On purpose.</p><p>They asked it to describe everything on screen. It came back with a timestamped play-by-play.</p><p>The description was precise. But its guesses about why things broke were mostly wrong. It assumed a native Android app. Ditto’s app is actually a web app.</p><p>It also picked up things the app never drew, like the screen recorder’s buttons and the keyboard’s toolbar. A model watching pixels can’t tell which program drew them.</p><p>So the team used the AI for the “what”, and the code for the “why”.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BikPV2zLBMYEJeItvPiaNg.png" /></figure><h3>One owner for each job</h3><p>The code showed the real cause. Sixteen different code paths were moving the chat’s scroll position, each on its own timer.</p><p>The new screen follows one rule: each piece of state has one owner. Only one part of the code may move the scroll. Everything else has to ask it.</p><p>They measured before and after. On their test phone, the old screen left a gap after 7 of 12 keyboard movements. The new one left none.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pDPDKlZbXvYVj80DQji6aA.png" /></figure><h3>What I’m taking from this</h3><p>I’m not an engineer, and I can’t judge their code.</p><p>But two ideas feel useful for designers.</p><p>First, a screen recording beats a paragraph. It keeps the timing, and timing is often where “it feels janky” lives.</p><p>Second, AI is good at describing what’s on screen, and less reliable at explaining why. I want to use it for the first job and stay careful with the second.</p><p>Next time a prototype feels off, I’ll record it before I write a word.</p><p>Do you record your bugs, or still write them up? Have you tried asking AI to describe a screen recording?</p><p><strong>Source:</strong> <a href="https://heyditto.ai/blog/from-a-screen-recording-to-a-chat-rewrite">From a screen recording to a chat rewrite (Ditto, 3 Nov 2026)</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6832feb80135" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[When AI Works for You, It Should Show Its Work]]></title>
            <link>https://medium.com/@ganesh_govalkar/when-ai-works-for-you-it-should-show-its-work-c05cca257785?source=rss-1a15a3cfe255------2</link>
            <guid isPermaLink="false">https://medium.com/p/c05cca257785</guid>
            <category><![CDATA[design-thinking]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai-agents-in-action]]></category>
            <category><![CDATA[product-design]]></category>
            <category><![CDATA[ux-design]]></category>
            <dc:creator><![CDATA[Design Nomad]]></dc:creator>
            <pubDate>Tue, 03 Nov 2026 03:44:54 GMT</pubDate>
            <atom:updated>2026-11-03T03:44:54.649Z</atom:updated>
            <content:encoded><![CDATA[<p>A simple guide to how AI agents should tell you what they did, what they need, and what went wrong.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1fRLkKr6eQzeIdpXgvC5GQ.png" /></figure><p>Yesterday, I asked an AI assistant to tidy up a folder of files.</p><p>A minute later, it said “Done”. Just that.</p><p>I had no idea what it had moved, what it had skipped, or whether anything had broken. So I checked everything myself, which took longer than doing the job.</p><p>Then I read a short guide from GradientDeck about how AI agents should show their activity. An agent is an AI that does tasks for you, step by step, not just answers questions. The guide put words to what I felt.</p><h3>Say what happened, in plain words</h3><p>An agent does a lot. It opens files, searches, changes its plan and runs checks.</p><p>Showing every step with the same weight makes things longer, not clearer.</p><p>The guide suggests a simple top layer: a clear verb, a thing you recognise, and its state. “Read src/auth/session.ts” is easier to follow than “Tool execution complete”. A short result like “12 tests passed” beats a wall of output.</p><p>The full detail can sit behind a tap, for when you need it.</p><h3>A plan, a yes and a result are different things</h3><p>This was the part that stayed with me.</p><p>A plan is what the agent intends to do. An approval is your decision. A result is what actually happened.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*k7kKu6ZaCFdfeuCzCoMVxg.png" /></figure><p>If all three turn into one green tick, the screen looks calm at the exact moment it should be precise. An approved action can still fail. A finished plan can still leave a check undone.</p><h3>Give every wait a name</h3><p>Queued, paused, waiting for approval and waiting for you can all look like nothing is happening.</p><p>The guide says to give each of these a short label. And when something fails, say what you can do next: retry, change the request, carry on without that source, or stop.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-mZJm7PoNZQ-EgAX-_vPAQ.png" /></figure><p>It also asks for a lasting record. When you come back later, you should see what changed, what was checked and what was skipped.</p><h3>Why this matters to all of us</h3><p>More apps now have AI that acts for us. That only works if we can trust it.</p><p>Trust does not come from a nice animation. It comes from a clear, honest record you can check.</p><p>As a designer, I find that both humbling and useful. The hard part is not the AI. It is the words and states we put around it.</p><p>What does your AI tool say when it finishes a task? Does it tell you enough?</p><p><strong>Source:</strong> <a href="https://gradientdeck.com/blog/designing-readable-ai-agent-activity">Designing agent activity people can actually follow (GradientDeck, 2 Nov 2026)</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c05cca257785" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Figma Is Teaching Its AI to Make Small Edits Well]]></title>
            <link>https://medium.com/@ganesh_govalkar/figma-is-teaching-its-ai-to-make-small-edits-well-d030b19440cf?source=rss-1a15a3cfe255------2</link>
            <guid isPermaLink="false">https://medium.com/p/d030b19440cf</guid>
            <category><![CDATA[figma]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai-design-tool]]></category>
            <category><![CDATA[ux-design]]></category>
            <category><![CDATA[product-design]]></category>
            <dc:creator><![CDATA[Design Nomad]]></dc:creator>
            <pubDate>Sat, 31 Oct 2026 16:47:35 GMT</pubDate>
            <atom:updated>2026-10-31T16:47:35.157Z</atom:updated>
            <content:encoded><![CDATA[<p>Why a design tool is training its own model for the everyday changes designers make all day.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YeLXtpj__UUlZUv6c3dHLw.png" /></figure><p>Yesterday, I read a post from Figma’s engineering team.</p><p>I expected a big launch. Instead, it was about something much smaller.</p><p>Tiny edits. The kind I make fifty times a day.</p><h3>The problem with big AI models</h3><p>Most AI tools in design use large, general models. They are clever, but they were not made for design files.</p><p>Figma says that when it started, those models could not understand how a Figma file is put together. Making them work needed long, expensive instructions every time.</p><p>So Figma tried a different path. It decided to train its own model for one job: editing work that is already on the canvas.</p><h3>How they taught it</h3><p>First, they used something called fine-tuning. In simple words, that means learning from examples.</p><p>Each example had three parts: a starting design, an edit request, and a good version of the finished change.</p><p>One of their early models learnt from more than 250,000 of these examples. Figma says it became competitive with the top models of that time on editing tasks.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*odO1t60sQQ8WrUcz6wz0Mw.png" /></figure><p>But learning only by copying has limits. The model gets good at things it has seen, and weaker at things it has not.</p><p>So now they are also using reinforcement learning. That means the model practises, gets scored, and improves. The scores are shaped by criteria written by experts, based on what designers actually care about.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*je0Hb7ACzAv2oPVZnucEHA.png" /></figure><h3>One agent, more than one model</h3><p>Here is the part I found most useful.</p><p>The Figma agent does not use just one model. For precise edits, it uses Figma’s own. For things like image generation or brand-new explorations, it uses the best outside model available.</p><p>Figma says typical tasks became about 20% faster between the beta and the full release.</p><p>On the same day, Canva announced Brush, its own design-editing model. So this is not only a Figma idea. Design tools are starting to build AI that knows how design files work.</p><h3>What this means for us</h3><p>Smaller, focused models may matter more than bigger ones for daily work.</p><p>A good edit is not only fast. It changes what you asked for and leaves everything else alone.</p><p>That is a design standard, not a tech one. And it is nice to see it taken seriously.</p><p>Do you trust AI with small edits yet? Or do you still make them by hand?</p><p><strong>Source:</strong> <a href="https://www.figma.com/blog/training-a-figma-fluent-editing-model/">How we’re training a Figma-fluent editing model (Figma Blog)</a>. Also mentioned: <a href="https://www.canva.com/newsroom/news/canva-brush/">Introducing Canva Brush (Canva Newsroom)</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d030b19440cf" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>