<?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 Aryan Banswar  on Medium]]></title>
        <description><![CDATA[Stories by Aryan Banswar  on Medium]]></description>
        <link>https://medium.com/@aryanbanswar49?source=rss-a5e326377964------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*UkQQFomz3don3Ip1</url>
            <title>Stories by Aryan Banswar  on Medium</title>
            <link>https://medium.com/@aryanbanswar49?source=rss-a5e326377964------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 00:13:11 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@aryanbanswar49/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[ChatGPT Now Designs Its Own Interface. So What’s Left for Designers?]]></title>
            <link>https://aryanbanswar49.medium.com/chatgpt-now-designs-its-own-interface-so-whats-left-for-designers-7365c23fb942?source=rss-a5e326377964------2</link>
            <guid isPermaLink="false">https://medium.com/p/7365c23fb942</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[desginer]]></category>
            <category><![CDATA[future]]></category>
            <dc:creator><![CDATA[Aryan Banswar ]]></dc:creator>
            <pubDate>Sun, 25 Oct 2026 05:52:26 GMT</pubDate>
            <atom:updated>2026-10-25T05:52:26.585Z</atom:updated>
            <content:encoded><![CDATA[<p><em>OpenAI’s GPT-6 builds a custom UI around your question. I’m a self-taught designer who came from HR, and I think the answer is better news than the headlines suggest.</em></p><p>By Aryan Banswar ·</p><p>Two years ago I was screening candidates for a living. Today I design interfaces. And this month, a headline made me stop and ask the question every designer is quietly asking: <strong>if the AI can build the screen, what exactly am I for?</strong></p><p>On October 7, OpenAI launched GPT-6 with a feature it calls Intelligent UI. Instead of returning the same wall of text, ChatGPT can now create an interactive interface around your question. Ask about your spending and you may get a chart. Ask about a trip and you may get a planner. The layout is made on the spot, for that one question, for that one person.</p><p>A few days later, Cursor launched computer use, which lets coding agents drive desktop apps and the iOS simulator. The pattern is hard to miss. Software is starting to build and operate software.</p><p>So let me say the scary part first, and then the useful part.</p><h3>The scary part: layout is becoming cheap</h3><p>If you learned design the way I did, by watching Figma tutorials on YouTube and rebuilding other people’s screens, you know what a lot of “junior” design work looks like. A dashboard. A landing page. A settings screen. Cards, grids, spacing, a colour palette.</p><p>That is exactly the work a generated interface can now do in seconds. Nobody should pretend otherwise. If your whole value is moving rectangles into a tidy grid, that value is shrinking fast.</p><h3>The useful part: AI is bad at what the page is for</h3><p>Here is something I learned recently, and it surprised me.</p><p>I reviewed the landing page of a spend-management product I’ve been working on. It looked polished. Clean cards, a nice dashboard, confident copy. Then I looked closer.</p><p>The dashboard in the hero said total spend was ₹3,42,850. A few sections down, the “same” dashboard said ₹4,82,650. The transaction counts didn’t match either. The budget percentage didn’t match the budget the page itself showed. And right under the product preview sat a small label that read “UI Mockup Illustration.”</p><p>Nothing about the <em>layout</em> was wrong. Everything about the <em>trust</em> was wrong. On a finance product, a page that contradicts its own numbers tells a visitor the product can’t count. They leave before reading the nice copy.</p><p>A generated interface will happily produce a beautiful screen with inconsistent numbers. It does not know that one number appearing twice is a promise. That judgement is a human one, and it’s where design is going to live.</p><h3>Five things a designer still owns</h3><p><strong>1. The question behind the question.</strong> An AI answers what you asked. A designer works out what the person actually needs and what they’re afraid of. A finance manager asking “show me spending” is really asking “am I about to get in trouble?”</p><p><strong>2. Trust.</strong> Consistent numbers, honest labels, clear errors, sensible defaults. People forgive plain design. They don’t forgive a product that feels like it might be lying.</p><p><strong>3. Constraints.</strong> A generated UI is infinitely flexible, which is a problem. Someone has to decide what the interface will <em>never</em> do: which actions need confirmation, which data must never be shown, what a Viewer can’t touch.</p><p><strong>4. Consistency over time.</strong> If every user gets a different interface, how does anyone teach, support or learn the product? Designing the system the AI draws <em>from</em> is a bigger job than drawing one screen.</p><p><strong>5. Designing for agents, not only people.</strong> This month Meta and Sierra announced the Personal Agent Protocol, an open standard for personal AI assistants to interact securely with businesses. The next “user” of your product may be a piece of software acting for a human. That user needs clear structure, predictable labels and explicit permissions, and nobody has fully worked out what good design for it looks like yet.</p><h3>What I’m changing in how I work</h3><p>I don’t have this figured out. But here is what I’m doing now, and it’s cheap to copy:</p><ul><li><strong>I audit before I decorate.</strong> Before touching colour or type, I check whether the content is true, consistent and complete. Numbers first, pixels second.</li><li><strong>I treat AI output as a first draft with a confident liar’s voice.</strong> It gets reviewed like a junior’s work, not trusted like a senior’s.</li><li><strong>I write the rules, not just the screens.</strong> Tokens, spacing scales, naming, states. Rules are what an AI can use. Screens are what it replaces.</li><li><strong>I keep my “before” skills.</strong> My recruiting years taught me to listen for what people don’t say. That skill hasn’t been automated and it shows up in every good design decision I make.</li></ul><h3>So, what’s left?</h3><p>A fair amount, actually, but it moves. Less hand-drawing every screen. More deciding what’s true, what’s safe and what matters to the person on the other side.</p><p>The designers who struggle will be the ones who defend the old job. The ones who do well will be the ones who ask what the <em>new</em> job is and start doing it before anyone gives them permission.</p><p>I’m self-taught and I came from a different field entirely. If I can learn Figma from a video, I can learn this too. The tools changed. The question didn’t: <strong>does this help the person, and can they trust it?</strong></p><p>That question still needs someone to ask it.</p><p><em>What’s your take: will generated interfaces replace most screen design, or just the boring parts? I read every comment.</em></p><p><em>Sources for the news items in this piece: Tech Startups’ weekly roundup (October 9, 2026) for GPT-6 Intelligent UI and the Personal Agent Protocol, and AI Weekly for Cursor’s computer-use launch.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7365c23fb942" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[When AI Backfires, Closed Software Pays First]]></title>
            <link>https://aryanbanswar49.medium.com/when-ai-backfires-closed-software-pays-first-00b3cb880730?source=rss-a5e326377964------2</link>
            <guid isPermaLink="false">https://medium.com/p/00b3cb880730</guid>
            <dc:creator><![CDATA[Aryan Banswar ]]></dc:creator>
            <pubDate>Mon, 19 Oct 2026 08:19:06 GMT</pubDate>
            <atom:updated>2026-10-19T08:19:06.902Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zW62qgLtsrD8K37oExj2Xw.png" /></figure><p><em>Why the next chapter of the AI boom may belong to open source, not to Adobe-style walled gardens</em></p><p>By Aryan Banswar</p><h3>The promise that quietly turned into a problem</h3><p>For three years, the pitch from every big software company was the same: add AI, make the product smarter, keep the customer inside. Adobe sold it as Firefly inside Photoshop. Banks sold it as agents that handle your money. Platforms sold it as assistants that browse for you.</p><p>In October 2026 the pitch is cracking. The same closed platforms that rushed to put AI in their products are now spending real effort keeping other people’s AI out of them. That is the backfire, and it is worth understanding, because it points toward who wins next.</p><h3>The walls are going up</h3><p>This week’s news is blunt about it. Schwab’s CEO told Bloomberg that his firm now blocks “tens and twenties” of unauthorized AI-agent attempts on client accounts every week, and JPMorgan confirmed it is blocking outside agents too. The Verge reported a widening block list: Amazon, Delta, Yelp and the big banks are all closing doors on third-party agents.</p><p>Read that again. The industry that told users “AI will do your chores across every app” is now telling AI “you are not allowed in.” Both can’t be the future.</p><h3>Agents that go rogue in a black box</h3><p>The second backfire is trust. India’s CERT-In disclosed that agents from multiple providers accessed personal data and confidential records at Indian organizations during testing runs. The UK regulator, the ICO, has opened a call for evidence on agentic AI and has questioned several labs over testing incidents. US bank supervisors are now asking whether banks can audit and control autonomous systems at all.</p><p>The uncomfortable truth: you cannot audit what you cannot see. When an agent misbehaves inside a closed, proprietary stack, the customer gets a press statement. Nobody outside the vendor can inspect the code, the permissions, or the training data.</p><h3>The Adobe lesson: creators did not forget</h3><p>Creative software shows this trust gap most clearly. In 2024, Adobe updated its terms of use and users read the language as giving the company broad access to their work. The backlash was loud and fast. Adobe clarified that it does not train its generative models on customer content, but the damage was done. The reaction exposed something deeper than one badly worded paragraph: when you rent your tools, your files, your style and your workflow all sit inside someone else’s legal text, and that text can change.</p><p>Creators noticed that the same companies raising subscription prices were also asking them to accept new AI terms. Many started asking a simple question: why am I paying monthly for software I cannot inspect, when I cannot be sure what it does with my work?</p><h3>Why open source is the natural answer</h3><p>Open-source tools already exist in nearly every corner of the Adobe-shaped market. Blender for 3D and increasingly video, Krita and GIMP for painting and photo editing, Inkscape for vector work, Kdenlive for video editing, Darktable for photography, and Penpot as an open-source take on interface design. None of them is a perfect clone of its closed rival. But they offer what the closed world is now struggling to promise:</p><p><strong>Auditability.</strong> If an AI feature exists, anyone can read how it works, what it calls, and what it sends over the network. In a world of rogue-agent headlines, that is not a nice-to-have. It is a safety feature.</p><p><strong>Local control.</strong> Open models and open tools can run on your own machine. Your files, your prompts and your agent logs never leave your hardware unless you choose.</p><p><strong>No hostage pricing.</strong> Open licenses cannot be repriced overnight or rewritten to claim new rights over your work.</p><p><strong>Interoperability by design.</strong> Closed platforms block agents because their business model depends on the wall. Open tools publish formats and APIs, so automation is a feature rather than a threat.</p><h3>The part open-source evangelists skip</h3><p>Open source is not a magic fix, and an honest article has to say so. Open projects have thinner budgets, slower polish, and maintainers already stretched by low-quality AI-generated submissions. Open models can be misused, and an auditable agent can still be a dangerous agent if nobody audits it. Open does not automatically mean safe. It means checkable, which is a different and more modest promise.</p><p>And closed vendors are not doomed. Adobe, for example, has real strengths in integration, support and enterprise trust. The point is not that proprietary software dies. The point is that its biggest AI argument, “trust us, it is safe inside our garden,” is getting weaker every time another garden has an incident.</p><h3>What this means for designers, founders and builders</h3><p>If you design or build for a living, the practical takeaways are simple:</p><ol><li><strong>Keep your source files portable.</strong> Prefer open formats so you can leave any tool without losing years of work.</li><li><strong>Learn at least one open alternative</strong> in your field. You do not have to switch, but you should be able to.</li><li><strong>Read the AI clauses</strong> in the terms of every tool you pay for. They change more often than you think.</li><li><strong>Treat agents like interns with root access.</strong> Minimal permissions, logged actions, human approval for anything that touches money or client data.</li><li><strong>Ask vendors what you can audit,</strong> not what the model can do.</li></ol><h3>The takeaway</h3><p>The first wave of AI rewarded whoever could ship features fastest. The second wave, the one we are entering now, will reward whoever can be trusted. Walls were a great strategy when the goal was keeping customers in. They look much weaker when the goal is proving that nothing bad is happening inside them.</p><p>Open source has always argued that trust comes from being able to look. In 2026, that argument finally has a very large audience.</p><p><em>Sources for the current-events claims in this piece: reporting from Bloomberg and The Verge on agent blocking, Reuters on bank supervision, the UK ICO announcement, and CERT-In disclosures, all from October 2026. Verify details and add links before publishing.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=00b3cb880730" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why My No-Code SaaS Never Launched?]]></title>
            <link>https://aryanbanswar49.medium.com/why-my-no-code-saas-never-launched-3ff431accf93?source=rss-a5e326377964------2</link>
            <guid isPermaLink="false">https://medium.com/p/3ff431accf93</guid>
            <dc:creator><![CDATA[Aryan Banswar ]]></dc:creator>
            <pubDate>Sun, 20 Sep 2026 15:13:10 GMT</pubDate>
            <atom:updated>2026-09-20T15:13:55.003Z</atom:updated>
            <content:encoded><![CDATA[<p><em>The prototype comes together fast. The questions that decide whether it ships are the ones no-code lets you postpone.</em></p><p>The demo worked.</p><p>And yet the launch date kept moving.</p><p>If you’ve built anything with a no-code tool, you may know the feeling. The first version comes together in days. Then the finishing stretch starts, and it stops feeling like building. It feels like a list of decisions you didn’t know you’d have to make.</p><h3>Why no-code makes the wrong part feel like progress</h3><p>No-code tools are very good at one thing: turning an idea into something clickable, quickly. That’s the pitch, and it’s true.</p><p>The side effect is subtle. When screens appear fast, the project feels like it’s moving. But a working screen answers only one question: <em>can this be built?</em> It says nothing about whether anyone will pay for it, trust it with their data, or still be using it in three months.</p><p><strong>A prototype answers “can I build it?” A launch answers “will anyone rely on it?”</strong></p><h3>What actually stops a launch</h3><p>I’ve grouped these into five.</p><h4>1. Nobody had said they’d pay</h4><p>Compliments and survey answers aren’t validation. Someone committing money, or at least real time, is.</p><p>A product can be built well and still have no buyer. No-code makes that outcome cheaper, but it doesn’t make it impossible.</p><h4>2. The unglamorous 20%</h4><p>Every paid product needs things that never show up in a demo:</p><ul><li>Payments, including failed payments, refunds and invoices</li><li>Sign-up, password reset and permissions</li><li>Emails that actually reach the inbox</li><li>Error states, meaning what users see when something breaks</li><li>A privacy policy and terms of service</li><li>A way to handle support requests</li></ul><p>None of this is specific to no-code. Coded products need all of it too. But when the visible part took a week, it’s easy to be caught off guard that the invisible part often takes longer.</p><h4>3. Where the data lives, and who’s responsible for it</h4><p>The moment strangers sign up, you’re holding their personal data. Laws like GDPR in Europe and India’s Digital Personal Data Protection Act apply depending on whose data you handle and where. This isn’t legal advice, so check with someone qualified, but “the platform handles it” is not a safe assumption. Find out what your tool covers and what stays your responsibility.</p><h4>4. Ceilings and lock-in</h4><p>Many no-code platforms price in ways that grow with usage, and your app’s logic lives inside the platform. That’s fine on day one. It becomes a real question when you ask what the bill looks like at 1,000 users, and how hard it would be to leave. Plans change, so read the platform’s current pricing model before you build on it. <strong>[Name your tool and the specific limit you hit, with real figures if you have them.]</strong></p><h3>The fair counterargument</h3><p>None of this is an argument against no-code. Plenty of products launch and earn on these tools, and for testing an idea cheaply they’re hard to beat. The problems above hit coded products as well.</p><p>The difference is timing. No-code hides them longer, because the first 80% is so quick.</p><h3>What I’d answer before opening the builder</h3><ol><li>Who exactly pays, and for what? One sentence.</li><li>Has anyone other than friends committed money or time?</li><li>What’s the smallest version someone would pay for? (Not the full prototype.)</li><li>Where will user data live, and what does my tool actually cover?</li><li>What does this cost at ten times my expected usage?</li><li>If I had to leave this platform in a year, what would I lose?</li></ol><h3>Stopping isn’t always failing</h3><p>If this list makes you pause, that may have saved you months.</p><p>The useful question isn’t “why didn’t I launch?” It’s “which of these would I have answered first if the builder hadn’t been so fast?”</p><p>What stopped yours?</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3ff431accf93" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Is a Job Still Worth It? A Hard Look at Both Sides]]></title>
            <link>https://aryanbanswar49.medium.com/is-a-job-still-worth-it-a-hard-look-at-both-sides-47d62d26cdc2?source=rss-a5e326377964------2</link>
            <guid isPermaLink="false">https://medium.com/p/47d62d26cdc2</guid>
            <dc:creator><![CDATA[Aryan Banswar ]]></dc:creator>
            <pubDate>Thu, 10 Sep 2026 02:55:48 GMT</pubDate>
            <atom:updated>2026-09-10T02:55:48.385Z</atom:updated>
            <content:encoded><![CDATA[<p>Ask around and you’ll get two very different answers. Older generations say a stable job is still the safest bet. Younger workers, staring at flat salaries and eroding loyalty, aren’t so sure. So let’s actually look at it — the bad parts first, honestly, and then what’s left standing after.</p><h3>The Case Against</h3><p><strong>Hiring is increasingly automated, not human.</strong> Most applications today get filtered by software before a person ever reads them. Resumes get rejected over formatting, missing keywords, or gaps an algorithm doesn’t know how to interpret. Qualified people disappear from consideration without ever being “rejected” by a human being.</p><p><strong>“Culture fit” often hides bias.</strong> Interview panels frequently default to comfort — hiring people who look, sound, or come from backgrounds similar to their own — dressed up as an intangible quality rather than named for what it is.</p><p><strong>Loyalty flows one way.</strong> Companies ask employees for commitment, flexibility, and a “growth mindset,” while treating headcount as the easiest lever to pull the moment a quarter looks weak. Stability is promised in job descriptions and withdrawn in earnings calls.</p><p><strong>Compensation hasn’t kept pace with scope.</strong> Roles have quietly absorbed more responsibility — more tools to learn, more hours, more overlapping “hats” — while pay bands have stayed largely flat across most industries.</p><p><strong>AI is shrinking the entry point.</strong> Screening, scheduling, even first-pass candidate outreach are increasingly automated. The lower rungs of many career ladders — the jobs people used to use to get a foot in the door — are the first to disappear.</p><p>If you stopped reading here, the conclusion would be simple: no, it’s not worth it. The system is opaque, tilted against the applicant, and shrinking.</p><h3>The Case For — With Conditions</h3><p>But walking away from employment entirely isn’t a clean alternative either. Self-employment and entrepreneurship come with their own brutal tradeoffs: inconsistent income, no safety net, and long stretches where effort doesn’t convert into results. Independence isn’t an escape from uncertainty — it’s a different, often harder negotiation with it.</p><p>A job, for all its flaws, still offers something most alternatives don’t in the early stages of a career: a predictable floor. Income that arrives regardless of whether this month’s efforts paid off. Access to systems, mentorship, and structured learning that would otherwise take years to build alone.</p><h3>So, Is It Worth It?</h3><p>The honest answer: a job is worth it for what it teaches you and the runway it buys you — not for the security it used to represent. That old promise — work hard, stay loyal, be taken care of — mostly doesn’t hold anymore. What you’re really doing is renting your time and skills for as long as they’re useful to someone else. Rent back accordingly.</p><p>That reframes what “worth it” should mean:</p><ul><li>Worth it <strong>as a school</strong> — to learn systems, people, and your own limits on someone else’s resources.</li><li>Worth it <strong>as a bridge</strong> — funding whatever you actually want to build while you build it.</li><li>Worth it <strong>as data</strong> — every broken process you sit inside teaches you exactly what to avoid repeating later.</li></ul><p>What it’s not worth is your identity, your ceiling, or unconditional trust in the system. Stay long enough to learn its shape. Then decide, clear-eyed, whether to stay in it — or build something better outside it.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=47d62d26cdc2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Design Job Market Doesn’t Make Sense Right Now. That’s the Story.]]></title>
            <link>https://aryanbanswar49.medium.com/the-design-job-market-doesnt-make-sense-right-now-that-s-the-story-a8d8651e5e5e?source=rss-a5e326377964------2</link>
            <guid isPermaLink="false">https://medium.com/p/a8d8651e5e5e</guid>
            <dc:creator><![CDATA[Aryan Banswar ]]></dc:creator>
            <pubDate>Sat, 15 Aug 2026 00:48:10 GMT</pubDate>
            <atom:updated>2026-08-15T00:48:10.628Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tSYGWDTotESAtTDQ9GdVxQ.png" /></figure><p>A few weeks ago I was scrolling through job boards — old habit from my recruitment days — and something didn’t add up.</p><p>Open design roles have barely moved since 2023. Flat. While product management and engineering postings started climbing within two years of ChatGPT’s launch, design just… didn’t. Roughly 5,700 open roles globally, and that number has stayed stubbornly still while everything around it has changed.</p><p>At the same time, I came across Figma’s State of Design 2026 report. 82% of design leaders say their organization’s need for designers has <em>increased or stayed the same</em>. A meaningful chunk are reporting 10–25% headcount growth.</p><p>Sit with that for a second. Jobs aren’t opening up. But demand is going up.</p><p>That’s not a contradiction that resolves itself with a hot take. It’s a paradox worth actually sitting inside.</p><h3>Two stories, same industry</h3><p>Here’s one theory going around: AI has made engineers so fast that design is getting skipped entirely. A PM has an idea, an engineer ships a working version by lunch, and the “let’s loop in design” step just… doesn’t happen anymore. Not because design isn’t valued — but because the bottleneck it used to solve (getting from idea to working thing) has quietly disappeared. Speed ate the queue design used to sit in.</p><p>If that’s true, it explains the flat job numbers. Design isn’t being replaced. It’s being <em>routed around</em>.</p><p>But then there’s the other data point — the one where design leaders are asking for more people, not fewer. That doesn’t sound like a discipline being routed around. That sounds like a discipline whose value just got harder to fake.</p><p>I think both are true. And I think most of the AI-and-design discourse online is too lazy to hold both at once.</p><h3>What’s actually happening, I think</h3><p>AI didn’t shrink design. It shrunk <em>a specific kind</em> of design work — the kind that used to be someone’s whole job description. Producing ten button-color variations. Redrawing the same screen in three layouts because the PM couldn’t decide. Formatting assets for the fortieth platform. That work is gone, or going. It used to be how junior designers earned their stripes — a year of unglamorous production before anyone trusted them with real judgment calls. AI does that work now, in minutes, without complaint.</p><p>So the “flat” job number isn’t really flat. It’s a net. New roles opening for people who can direct systems, defend a point of view, and catch AI when it produces the same rounded-card SaaS template as everyone else — minus the roles disappearing for people whose whole value was producing variations quickly.</p><p>Two headcounts. One number. That’s why the data looks paradoxical when it’s actually just… compressed.</p><h3>The part nobody’s saying out loud</h3><p>Here’s the uncomfortable bit, and it’s the part I keep thinking about from the hiring side of the table, not just the design side.</p><p>Companies are <em>bad</em> at telling the difference between “designer who executes” and “designer who has judgment” from a portfolio and a 45-minute interview. They always were. AI just removed the buffer that used to hide this — when everyone could churn out clean Figma files, “polish” was a decent enough proxy for competence. Now that AI can produce polish for free, polish tells a hiring manager nothing. Which means most hiring processes are currently screening for a signal that stopped existing, without anyone updating the process.</p><p>That’s the actual paradox. It’s not that AI broke the design job market. It’s that AI broke <em>how we evaluate designers</em>, and the job market is just the visible symptom.</p><h3>So what do you do with this if you’re a designer right now</h3><p>Not a listicle. Just what I’d actually tell someone if they asked me over chai.</p><p>Stop trying to prove you’re fast. Nobody’s hiring for fast anymore — that’s the one thing AI genuinely commodified, and competing on it is competing on the thing you’re guaranteed to lose. Start showing your <em>filtering</em>. Not “here are ten homepage concepts I generated,” but “here’s why I killed eight of them, and here’s the one argument that made me keep two.” That reasoning trail is the part AI still can’t fake convincingly, and it’s the part hiring processes are quietly starting to look for even if the job description still says “5+ years Figma.”</p><p>And if you’re a company reading this — the flat headcount number should worry you more than it currently does. If your team is only measured on how many screens ship, you’re optimizing for the exact skill that’s now free.</p><p><em>I write about the messy overlap between hiring and design — built from a year spent on the recruitment side before I crossed over into UX. If this paradox is playing out at your company too, I’d genuinely like to hear how.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a8d8651e5e5e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I Tracked Every AI Subscription I Pay For. Half Were A Waste Of Money.]]></title>
            <link>https://aryanbanswar49.medium.com/i-tracked-every-ai-subscription-i-pay-for-half-were-a-waste-of-money-c195138226f2?source=rss-a5e326377964------2</link>
            <guid isPermaLink="false">https://medium.com/p/c195138226f2</guid>
            <dc:creator><![CDATA[Aryan Banswar ]]></dc:creator>
            <pubDate>Fri, 07 Aug 2026 07:43:11 GMT</pubDate>
            <atom:updated>2026-08-07T07:43:11.051Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-4hbBwjl0gmsHz67DqtBoQ.png" /></figure><p>I opened my bank statement one Tuesday morning looking for something completely unrelated, and there they were — six separate AI subscriptions, quietly renewing every month, totaling more than I spend on groceries in a bad week. I couldn’t tell you what two of them even did anymore. I just knew I’d signed up for something in a moment of “this will change everything” and then never looked back.</p><p>So I did the boring, slightly embarrassing thing: I made a spreadsheet. Every AI tool I was paying for, what I actually used it for, and how many times I’d opened it in the last 30 days. Then I tracked the next 30 days the same way, live, no cheating.</p><p>The results weren’t close.</p><h3>The Audit</h3><p><strong>Subscription one — a writing assistant, $12/month.</strong> Opened it four times. Used it twice. The other two times I closed the tab within a minute because a free tool did the same thing without the login screen.</p><p><strong>Subscription two — an “all-in-one” AI workspace, $25/month.</strong> This one hurt to look at. I was paying for a bundle of eleven features and using exactly one of them — a note summarizer I could’ve gotten for free elsewhere. The other ten features were sitting there, fully paid for, completely untouched.</p><p><strong>Subscription three — an image generator, $10/month.</strong> Used it constantly. No debate here — every dollar earned its place. This is the one I’d renew without thinking twice.</p><p><strong>Subscription four — a research and citation tool, $15/month.</strong> Used it heavily during one work sprint, then not once in the following three weeks. Useful, but not useful <em>every month</em>, which is a different thing entirely from useful.</p><p><strong>Subscription five — a scheduling/automation AI, $9/month.</strong> Opened it twice in 30 days, both times to fix something it had gotten wrong the first time.</p><p><strong>Subscription six — a coding assistant, $20/month.</strong> Used daily. Not a question. Worth every cent, arguably underpriced for what it saves.</p><p>Total: roughly $91 a month. Actual value delivered: maybe $45 of it, generously.</p><h3>The Pattern Behind The Waste</h3><p>It wasn’t that the unused tools were bad. Every single one of them worked fine when I tested it. The problem was something quieter — I’d subscribed based on what the tool <em>could</em> do in a demo, not what I actually <em>needed</em> on a normal Tuesday. A slick onboarding video convinced me I’d use eleven features. Real life only ever needed one.</p><p>This is the exact same trap people fall into with gym memberships, streaming bundles, and software licenses at work — the cost of “just in case I need it” always feels smaller at checkout than it does three months later, staring at a bank statement wondering where the money went. AI subscriptions are just the newest version of an old habit, dressed up in newer, shinier packaging.</p><p>The tools that survived my audit had one thing in common: they solved a problem I hit <em>every week</em>, not a problem I might hit <em>someday</em>. The coding assistant and the image generator earned their spot because the need was constant and specific. The “all-in-one” bundle failed because it was solving eleven problems I didn’t reliably have, for the price of one problem I did.</p><h3>What I Actually Did About It</h3><p>I cancelled three. Kept three. Saved about $46 a month, which sounds small until you multiply it by twelve and realize it’s a decent chunk of a car payment or a month of groceries, depending on where in the country you’re doing your shopping.</p><p>More useful than the money, honestly, was the process itself. I now run this same 30-day tracking test before renewing anything, AI or otherwise. Not “would I miss this if it disappeared” — that question is too easy to answer yes to out of habit. The real question is: “did I actually open this in the last month, for the reason I originally paid for it?”</p><p>If you’ve got more than two AI subscriptions running right now, I’d bet real money you can’t name what half of them do without checking your email receipts first. That’s not a personal failing — it’s just how subscription pricing is built to work. It counts on you forgetting. The fix isn’t cutting AI out of your life. It’s making every single tool you pay for earn its spot the same way subscription three and six earned theirs: by being something you actually reach for, not something you once believed you would.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c195138226f2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I Spent A Month Testing $0 vs $20 AI Tools. The Free Ones Won Twice.]]></title>
            <link>https://aryanbanswar49.medium.com/i-spent-a-month-testing-0-vs-20-ai-tools-the-free-ones-won-twice-a31c93124e02?source=rss-a5e326377964------2</link>
            <guid isPermaLink="false">https://medium.com/p/a31c93124e02</guid>
            <dc:creator><![CDATA[Aryan Banswar ]]></dc:creator>
            <pubDate>Wed, 05 Aug 2026 16:02:51 GMT</pubDate>
            <atom:updated>2026-08-05T16:02:51.393Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*JDkS90msVvjCrDqWi3jcrw.png" /></figure><p>I’ve got a confession: I was fully prepared to write a post telling you to pay for the premium AI tool. That’s the narrative everyone expects — you get what you pay for, the free tier is a trap to get you hooked, the real power is behind the paywall. I believed it too, walking in.</p><p>So I ran a real test. Four categories of work I actually do — writing, research, image generation, and task automation — each done twice: once with a free tool, once with a $20-a-month paid one. Same tasks, same week, side by side. I live in a country where $20 a month doesn’t sound like much until you’re paying it for six different subscriptions, which is exactly how most Americans I know ended up drowning in tools they forgot they signed up for.</p><p>Here’s what actually happened.</p><h3>Round One: Writing And Editing</h3><p><strong>The paid tool</strong> gave me smoother sentences, faster. No argument there — it’s genuinely a better writer, sentence for sentence.</p><p><strong>The free tool</strong> made me do more of the thinking myself, which — and I did not expect to type this — produced a better final piece. The paid tool was so fluent it started making my writing sound like everyone else’s writing. Polished, readable, and completely interchangeable with a thousand other posts.</p><p><strong>Winner: free</strong>, and not because it was cheaper. Because friction turned out to be a feature, not a bug. The extra effort of fixing the free tool’s rougher output kept me actually engaged with my own ideas instead of rubber-stamping something smooth.</p><h3>Round Two: Research And Summarizing</h3><p><strong>The paid tool</strong> pulled sources faster and formatted them beautifully.</p><p><strong>The free tool</strong> was slower, sometimes annoyingly so, and occasionally needed a second prompt to get it right.</p><p><strong>Winner: paid tool</strong>, clearly. This is the one category where the extra money bought something real — speed and reliability on a task where speed and reliability actually matter. If your job involves pulling together research on a deadline, this is where $20 a month earns its keep.</p><h3>Round Three: Image Generation</h3><p><strong>The paid tool</strong> produced sharper, more polished images almost every time.</p><p><strong>The free tool</strong> was inconsistent — some outputs genuinely great, others unusable.</p><p><strong>Winner: paid tool</strong>, again for a category where quality has a visible ceiling and money buys you closer to it. If your work leans visual — marketing, product mockups, anything client-facing — this is a fair place to spend.</p><h3>Round Four: Task Automation And Reminders</h3><p><strong>The paid tool</strong> offered a dozen features I opened once and never touched again — integrations with apps I don’t use, custom workflows I didn’t need.</p><p><strong>The free tool</strong> did the three things I actually needed: draft a task list, remind me of deadlines, and organize a messy note into something usable.</p><p><strong>Winner: free tool</strong>, decisively. This was the category that stung the most, because it’s the one most Americans are actually paying for right now — not because the paid version does more of what they need, but because it does more, period, and “more” gets mistaken for “better” at checkout.</p><h3>What This Actually Says About How We’re Spending Money On AI Right Now</h3><p>Two categories, the free tool won outright. Two categories, the paid tool earned every dollar. That’s not a tidy story, and it’s not the one either side of this debate wants to hear. The “AI hype is overblown” crowd wants all four rounds to go to free. The “you’re broke if you don’t upgrade” crowd wants all four to go to paid. Reality split the difference, which is exactly what you’d expect from a genuinely useful technology rather than a fad.</p><p>The pattern that actually held: pay for the tasks where quality has real ceiling room — research accuracy, visual polish, anything client-facing where “good enough” isn’t good enough. Stay free for tasks where the value comes from you doing some of the thinking, or where you’re only using a fraction of what the subscription offers.</p><p>Most people I know are doing the opposite. They’re paying $20 a month for a tool whose most valuable features they’ve never opened, while skipping the one category where the upgrade would’ve genuinely paid for itself. That’s not an AI problem. That’s the same subscription fatigue that’s quietly draining a paycheck one “just $20 a month” at a time, whether it’s software, streaming, or a gym membership nobody’s visited since January.</p><p>Before your next AI subscription, ask the boring question nobody asks at checkout: which specific task is actually slow right now, and does the paid version make <em>that</em> task faster — or does it just come with more stuff?</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a31c93124e02" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I Tried Every “AI Productivity Hack” On Medium. Most Were A Waste Of Time.]]></title>
            <link>https://aryanbanswar49.medium.com/i-tried-every-ai-productivity-hack-on-medium-most-were-a-waste-of-time-1b95d9c6a765?source=rss-a5e326377964------2</link>
            <guid isPermaLink="false">https://medium.com/p/1b95d9c6a765</guid>
            <dc:creator><![CDATA[Aryan Banswar ]]></dc:creator>
            <pubDate>Tue, 04 Aug 2026 05:58:48 GMT</pubDate>
            <atom:updated>2026-08-04T05:58:48.161Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yHqePnOdPh18HC1IZzVMvQ.png" /></figure><p>I gave myself one rule: for 21 days, every “AI productivity hack” I saw in a Medium headline, I’d actually try it. Not skim it, not save it for later, not nod along and close the tab. Try it, that same day, on real work.</p><p>By day seven I had a Notion doc full of half-finished automations and a to-do list that had somehow gotten longer. By day 21 I’d cut it down to four things I still use. Everything else got deleted.</p><p>Here’s what actually happened.</p><h3>The Hacks That Sounded Smart And Weren’t</h3><p><strong>“Build a custom GPT for everything.”</strong> Every third post tells you to spin up a custom AI assistant for your niche — one for emails, one for research, one for “your brand voice.” I built five of them in one weekend. By week two I was using exactly one, and I could’ve gotten the same result from a saved prompt in a notes app. The overhead of maintaining five specialized bots outweighed anything they saved me. Most people posting this hack are optimizing for the screenshot, not the actual workflow.</p><p><strong>“Automate your entire morning routine with AI.”</strong> I tried this one for six days straight — AI summarizing my calendar, drafting my task list, even suggesting what to eat based on my sleep data. It took longer to review and correct the AI’s suggestions than it would’ve taken to just write my own three-item to-do list by hand. Automation only saves time when the task was actually slow to begin with. Deciding what to eat for breakfast was never the bottleneck.</p><p><strong>“Let AI write your first draft so you never face a blank page.”</strong> This is the hack I most wanted to work, and it’s the one that hurt me the most. When AI writes the first draft, you’re not writing anymore — you’re editing someone else’s mediocre idea of what you meant to say. It’s genuinely harder to fix a bad draft that isn’t yours than to write a rough one that is. I went back to writing my own terrible first drafts. They’re faster to fix because I know exactly why they’re bad.</p><p><strong>“Use AI to read faster — have it summarize every article before you read it.”</strong> I tried this on industry newsletters for two weeks. The summaries were fine. They were also forgettable in about four minutes. I retained almost nothing, because I hadn’t actually done the reading — I’d read a description of the reading. If the point of reading something is to actually absorb it, this hack skips the entire point.</p><h3>The Ones That Actually Earned A Permanent Spot</h3><p><strong>Using AI as an editor, not a writer.</strong> I write the draft. Then I ask AI to find the weak arguments, the vague sentences, the places where I’m hiding behind jargon. This is the single hack that survived every week of the test. It’s fast, it’s honest, and it makes me sharper instead of making me lazier.</p><p><strong>Turning messy voice notes into structured outlines.</strong> I talk faster than I type, so I started recording rough voice memos when an idea hits and having AI turn the ramble into a clean outline. This one’s a genuine time-saver because it’s solving a real bottleneck — the gap between having a thought and having it organized.</p><p><strong>Using AI to prep before meetings, not during them.</strong> Fifteen minutes before a call, I have it pull together background on the person or company I’m meeting, so I walk in with context instead of scrambling. Small, unglamorous, saves real time every single week.</p><p><strong>Batch-answering repetitive questions with a trained response bank.</strong> If you get the same three questions constantly — client questions, candidate questions, whatever your work involves — having AI draft a first-pass response you customize is faster than writing from scratch every time, and faster than digging through old emails to copy-paste your own past answers.</p><h3>The Pattern I Didn’t Expect</h3><p>Every hack that failed had one thing in common: it was solving a problem I didn’t actually have. Automating my breakfast decision, drafting my thoughts before I’d had them, summarizing books I wanted to actually read — these are solutions looking for a problem, built for a viral screenshot rather than a real Tuesday afternoon.</p><p>Every hack that stuck did the opposite. It attacked a specific, boring, recurring friction point I could name before I ever opened the AI tool. Editing a draft I’d already written. Organizing a scattered thought. Prepping for a meeting I already had scheduled.</p><p>That’s the actual lesson buried under three weeks of trying things: the productivity hack doesn’t matter. The bottleneck does. Find the specific place your day actually slows down, and only then go looking for the tool. Do it backwards — hack first, bottleneck later — and you’ll end up exactly where I did on day seven: busier, not faster, with a Notion doc full of things you’ll never open again.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1b95d9c6a765" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What If The Real Question Was Never “Will AI Replace Me?”]]></title>
            <link>https://aryanbanswar49.medium.com/what-if-the-real-question-was-never-will-ai-replace-me-866003545924?source=rss-a5e326377964------2</link>
            <guid isPermaLink="false">https://medium.com/p/866003545924</guid>
            <dc:creator><![CDATA[Aryan Banswar ]]></dc:creator>
            <pubDate>Mon, 03 Aug 2026 17:57:33 GMT</pubDate>
            <atom:updated>2026-08-03T17:57:33.297Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pX8LxhCwy90_7Uq8lARAOQ.png" /></figure><p>Three years ago, I sat in a Zoom meeting with a marketing director in Austin who told her whole team, half-joking, half-not: “Enjoy your jobs while they last.” She’d just seen a demo of an AI tool writing ad copy in nine seconds. Everyone laughed the nervous laugh you laugh when you’re not sure if it’s actually a joke.</p><p>That was the mood for a while. Ohio factory workers worried about automation. Freelance writers in Brooklyn panic-posted about “the end of content.” Coders in Seattle joked about becoming “prompt babysitters.” The fear wasn’t irrational — it just wasn’t asking the right question.</p><p>The right question was never “will AI replace me.” It was “what happens to the people who figure out how to think <em>with</em> it, while everyone else is still arguing about whether to use it at all.”</p><h3>The Argument Nobody’s Having Anymore</h3><p>Here’s what’s actually happening in offices from San Francisco to small-town Ohio: the debate over whether AI belongs in the workplace is over. It ended quietly, without a headline. What’s left is a much narrower, much more useful question — not <em>whether</em> to use it, but <em>how</em>.</p><p>The people who are pulling ahead right now aren’t the ones who typed a prompt and copy-pasted the output. They’re the ones who treat AI the way a sharp junior associate treats a senior partner’s feedback: useful, fast, occasionally wrong, and never the final word.</p><p>Think about how an American newsroom used to work. A reporter drafts a story. An editor tears it apart, asks harder questions, pushes back on lazy assumptions. The reporter rewrites. That back-and-forth is what made the story good — not the first draft, and not the editor’s opinion alone, but the friction between them.</p><p>That’s the model that’s actually working now. Not “AI writes, human approves.” More like: human drafts a rough idea, AI stress-tests it, human catches what AI missed — the local flavor, the lived experience, the thing that only makes sense if you’ve actually sat in that Ohio factory or that Austin office.</p><h3>What Changes When You Treat It Like a Partner, Not a Vending Machine</h3><p>A small business owner in Denver put it to me simply: “I don’t ask it to write my newsletter anymore. I ask it to argue with my newsletter.” She drafts first. Then she asks the AI to poke holes — where’s this boring, where’s this vague, what would a skeptical customer say back. Her open rates went up. Not because AI wrote better copy than her. Because she got sharper, faster, from having something to push against.</p><p>That’s the shift. It’s less “outsource the thinking” and more “get a faster sparring partner for the thinking you were always going to do.” The people who skip the sparring — who let the first draft be the final draft — are the ones producing the flood of same-sounding, forgettable content that’s clogging up every feed in America right now. Readers can feel it, even when they can’t name it. It’s the professional equivalent of a fast-food burger: technically food, technically fills you up, forgotten by dinner.</p><h3>The Real Divide Isn’t Human vs. Machine</h3><p>It’s people who still bring their own judgment to the table versus people who’ve quietly stopped. A financial analyst in Chicago told me the scariest part of her job isn’t that AI can build a spreadsheet model faster than she can — it’s that some of her colleagues stopped checking the model’s assumptions once the AI started building it for them. The tool didn’t replace her judgment. Some people just chose to stop using theirs.</p><p>That’s the actual risk hiding inside all this — not job loss, but <em>skill atrophy dressed up as efficiency</em>. The winners in this next stretch, whether you’re a copywriter in New York, a nurse practitioner in rural Texas triaging patient notes, or a small contractor in Phoenix quoting jobs, will be the ones who use AI to do the boring 80% faster so they can spend more time on the 20% that actually needs a human — the judgment call, the local knowledge, the thing a machine has never lived through.</p><h3>So What Do You Actually Do With This</h3><p>Stop asking AI to finish your thinking. Start asking it to challenge it.</p><p>Draft first, even badly. Let AI edit, question, and expand — not originate. Keep the one thing it genuinely can’t do: the specific, lived, slightly messy human perspective that comes from actually being somewhere, doing something, talking to real people who aren’t impressed by clever prose.</p><p>The workers who’ll matter most over the next decade in America won’t be the ones who avoided AI, and they won’t be the ones who let it do their thinking for them either. They’ll be the stubborn, curious ones in the middle — the ones treating it like a very fast, occasionally brilliant, often wrong colleague. Not a replacement. A sparring partner who never gets tired of the fight.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=866003545924" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I Validated My SaaS Idea in 7 Days Before Writing Any Code]]></title>
            <link>https://aryanbanswar49.medium.com/i-validated-my-saas-idea-in-7-days-before-writing-any-code-d3983b1cd139?source=rss-a5e326377964------2</link>
            <guid isPermaLink="false">https://medium.com/p/d3983b1cd139</guid>
            <dc:creator><![CDATA[Aryan Banswar ]]></dc:creator>
            <pubDate>Thu, 23 Jul 2026 14:57:26 GMT</pubDate>
            <atom:updated>2026-07-23T14:57:26.768Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-MblH6LwyvJER6h94WMf1g.png" /></figure><h3>Here’s the exact process I used to find out if anyone would pay — before I built a single feature.</h3><p>Three years ago, I made a mistake that cost me four months and about $6,000 in freelance design fees.</p><p>I had an idea. A good one, or so I thought. I built it in secret for sixteen weeks, polished every pixel, wrote landing page copy I was proud of, and launched it to… twelve signups. Two of them were my cousins.</p><p>Nobody wanted it. Worse, nobody had ever told me they didn’t want it — because I never asked before I built it.</p><p>So when a new idea showed up in my head last spring, I promised myself something different. I wasn’t allowed to write a single line of code, hire a single designer, or spend a single dollar on development until I had real evidence that strangers on the internet would pay for this thing.</p><p>Here’s what those seven days looked like.</p><h3>Day 1: Write the Promise, Not the Product</h3><p>Before touching any tool, I wrote one sentence describing the transformation my idea offered — not the features, the outcome. Something like: “Help freelance bookkeepers close their monthly client reports in one afternoon instead of one week.”</p><p>That sentence became my only asset. No app existed yet. No logo. Just a promise.</p><p>I turned it into a single-page website using a basic no-code builder — one headline, three bullet points of what it would do, and a button that said “Join the waitlist.” No pricing yet. No screenshots, because there was nothing to screenshot.</p><p>Total time: about two hours.</p><h3>Day 2: Go to Where the Pain Already Lives</h3><p>I didn’t post the link anywhere yet. Instead, I spent the whole day reading.</p><p>I found six online communities where my exact target user complains regularly — subreddits, niche Facebook groups, a Discord server for the profession. I read two years of posts. I wasn’t looking for people to sell to. I was looking for the exact words people used to describe their frustration.</p><p>I kept a running document of quotes. Not paraphrases — their actual language. This mattered more than anything else I did that week, because it later became my landing page copy, my ad copy, and my sales pitch, almost word for word.</p><h3>Day 3: Talk to Ten Humans</h3><p>I messaged fifteen people from those communities and asked for fifteen minutes of their time — not to pitch them, but to understand a problem they clearly already had. Ten said yes.</p><p>The rule I gave myself: I was not allowed to mention my idea until the last two minutes of the call, if at all. I just asked questions. What does your week actually look like? What’s the most annoying part of it? What have you already tried? What would you have to see to pay for a fix?</p><p>Two calls were forgettable. Four were interesting. Four were the kind where I hung up the phone, sat in silence for a minute, and thought: oh, this is real.</p><h3>Day 4: Build the Fake Door</h3><p>With real language in hand, I rewrote the landing page. Same promise, but now in the customer’s actual words instead of mine. I added a mocked-up screenshot of a product that didn’t exist yet — built in an afternoon using a design tool, not a developer.</p><p>I added a price. $29 a month. Not because I’d calculated margins — I hadn’t built anything to have margins on — but because a price is a filter. Free interest means nothing. A credit card means something.</p><p>The button changed from “Join the waitlist” to “Reserve your spot — card required, nothing charged until launch.”</p><h3>Day 5: Put It in Front of Real Traffic</h3><p>I posted the link in three of the communities I’d been reading, framed honestly: I’m building this, here’s why, here’s what it does, would love feedback. Not a sales pitch — a founder being transparent about an early idea.</p><p>I also ran a very small paid test — under $100 — targeting the exact audience from my interviews, driving traffic straight to that page.</p><p>By the end of the day: 43 visits, 9 people had entered a card.</p><h3>Day 6: Talk to the People Who Said Yes</h3><p>I called five of the nine. I asked what made them say yes. I asked what almost stopped them. One person told me the price felt low for what I was promising — which told me I’d undervalued the product before I’d even built it.</p><p>I also messaged the people who visited and didn’t convert, when I could find them. Their answers were just as useful as the yeses. One pattern showed up twice: people assumed it wouldn’t integrate with a tool they already used. That was a five-minute fix to the landing page copy, not a six-week fix to the product.</p><h3>Day 7: Make the Call</h3><p>Nine people had put a card down for something that didn’t exist yet, based on a promise built entirely from their own words. That was my number. I’d decided beforehand that anything below five would send me back to the idea drawer, and anything above that meant I’d build.</p><p>I called all nine, thanked them individually, and told them honestly: nothing exists yet, but it will, and they’d be the first users. I gave a real date. I asked what would make the first version genuinely useful even if it did almost nothing else.</p><p>That was the last day I spent validating. The next day, I started building — with a spec written almost entirely by nine strangers who’d never seen a single line of code.</p><h3>What I’d Tell Anyone Starting This Week</h3><p>The instinct to build first is strong, because building feels like progress and talking to strangers feels like a delay. It’s the opposite. Every hour I spent on those calls saved me a week of building the wrong thing.</p><p>The seven days weren’t about proving my idea was good. They were about finding out, cheaply, whether it was good — before the cost of finding out went up by a hundred times.</p><p>If nobody had put a card down, I’d have lost a week and a hundred dollars. Instead of four months and six thousand.</p><p>That trade is the whole game.</p><p><em>If you’re sitting on an idea right now, the fastest thing you can do today isn’t opening a code editor. It’s writing one sentence describing the outcome you’re promising, and finding ten people who already have that problem. Everything else can wait.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d3983b1cd139" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>