... realize how much of my job was taste until I started generating code fast. When you can produce ten versions of a feature in an afternoon, the bottleneck moves. It is no longer how fast you can build. It is how well you can tell which version is good. And that skill lives completely outside the tool. I have generated plenty of things that technically worked and were the wrong choice. Too much surface. Wrong emphasis. The kind of thing ... ... care about judgment. What to build, what to cut, what to leave ugly on purpose. That part is still entirely mine and I am not sure it can be outsourced. Do you feel the same? Has speed made the taste

... everyone Thanks for all the support on Taste Lab. A lot of you used it to pull the visual DNA out of websites you love. But I kept running into the next problem. Even with the right visual style, an AI-built page can still feel off, because the agent doesn't know what belongs on the page, in what order, or why. So I built Proof , and it launches in PH tomorrow. Taste

... also to solo-build or optimise the personal life. We are passed the point of discussion AI IS GOOD / AI IS BAD. It just is a new technology. Now it's up to individuals to use it in good taste, bad taste

... side. For the past several months I ve been building Cupmark, an Android app for people who want to make better coffee at home. I started it because of a problem I kept having. I d brew a cup, taste it, and think this is a bit sour or this is too bitter. Then I had no clear idea what to change. Most apps give you a recipe and leave it there. Here is what Cupmark does: Guided brewing. Pick ... ... method (V60, Chemex, AeroPress, French press, moka pot, espresso, or cold brew), follow a recipe, and brew with step-by-step timing. One suggested change. After you taste

... same as the same. The part I don't have is detection. You'd need a fixed set of prompts running on a schedule with a person looking at the outputs, which is an eval suite for taste, and I've never seen one that works. So in practice we find out when someone writes in to say it used to look right. If you're building on models you don't own, do you check for drift ... ... have seen work is not an eval suite but a folder of about twenty real user files run on a schedule, where a person only looks when the output size or duration jumps. That catches breakage and misses taste

... Product Hunt I'm Ethan Walkrman, a chef turned vibe builder at Matestable.com. Years on the pass taught me that taste doesn t fit inside a star rating it lives in what you order again, who you trust, and the table you keep talking about. Now I m building an agent discovery community that learns those signals and turns them into better restaurant discoveries. Early days. Shipping in public. Good to meet ...

... there and say nothing Comment from Toto De Brant(@toto_db): Depends, you can generate a lot of ideas with Stitch from Google, or with Claude, so if you're at MVP stage and have some level of taste, do it yourself, if you are like a hardcore backend developer, yeah perhaps get someone in for at least a few screens. But a design system is even more important, so all screens have a coherence between them, otherwise your ... ... stitch for ui mostly...its been amazing and the credits almost seem to never get over Comment from Shayan Basu Roy(@shayan_basu_roy): I feel you don't need a designer but you have to have good taste

... cactus and dried in the sun. Its own words: Seventy thousand of them for a pound of colour that once dressed cardinals and kings. Getting the cochineal right is a fact-check. Making the page breathe between chapters is taste. Fable 5.1 did both without being asked for either. Most cinematic websites are a hero image and a parallax trick. This one is edited like a film. World Three The snow globe that knew about the spacebar Winterholm a Three.js ... ... scientific computing) 52.6% Fable 5: 24.7% Opus 5: 29.0% Context window 1,000,000 tokens hold a whole codebase at once Max output 128,000 tokens a full build in one pass Prompt-cache reads up to 75% cheaper vs

... automation. Doesn't scale, but might be correct at my size. Where I'm stuck is that I can't measure the thing I'd need to decide. I don't know what share of retries are genuine misses versustaste. I could ask on the retry "what was wrong with it?" but I suspect that mostly annoys the people I most want to keep. For anyone running something generative and metered: how do you handle a failed generation ...

... number captures it. The uncomfortable takeaway: in the era where anyone can build the feature, the eval is the moat. It's the one artifact an agent can't hand you, because it's the encoding of your taste and your knowledge of your users. The makers who win the next stretch won't be the ones who ship fastest. They'll be the ones who can tell, reliably and before their users do, whether what they shipped is good ... ... teammate-with-zero-context trick is a good one, though i'd guess it has a shelf life too - the third time you pull the same person in for a tie break they've absorbed enough of your taste

... implementation tickets and works them one at a time. This is  its  altitude. I never open one. I never name a function. I never see a branch. That line is the whole design. The moves that need taste stay with me; everything mechanical including the agent's own mistakes stays below the line. The blueprint is the only thing I read end to end Before any code exists, the loop hands me one artifact: the  blueprint ... ... keyboard, or even at home. The agent owns everything mechanical, including its own mistakes. I own the two things that were ever mine: are these the right stories, and does each one actually work. Keep the taste

... build a fragrance app, so I spend a lot of time in how people describe what they own. One pattern keeps surprising me. The bottle someone bought and never reaches for predicts their taste better than the ones they love. A favorite tells you what you aspire to. A regret tells you what you actually are, and that turns out to be the more useful signal. So a question for the room, in two parts. If you have built recommendation ... ... other categories, do you find negative signal beats positive signal for figuring out a person? And a question for the fragrance people here. What bottle do you regret buying the most, and what did it teach you about your taste

... landing page, product pages, design system, even our promo video, all of it came from AI tools plus taste. Five years ago this would have needed at least one designer on payroll and tons of Figma design effort. Here's the uncomfortable part. The output is genuinely fine. Not award winning, but clean, consistent, and shipped in weeks instead of months. Nobody who visits our site asks if we have a designer. But I keep noticing what's missing. Every ... ... would never suggest because it breaks a pattern instead of following one. So my honest read is the junior designer role is in real trouble, because "make it clean and consistent" is exactly what AI does well now. But taste

... tasks the agent completed autonomously without me needing to undo the work. after ten successes in a category, that category becomes auto-approve. one failure resets it to watched. slow to build but you know exactly what to trust vs what to babysit. Comment from Gal Dayan(@galdayan): @aadilghani  same read-only-vs

... across every surface. Photograph a plate and a vision model recognizes the dish, then the same deterministic engine attaches the macros (I call it MitraVision). Ask for a diet swap and a fusion of culinary knowledge and your learned taste profile keeps the dish authentic: a Thai curry should still taste

... actually deserves real care. I caught myself lovingly engineering parts no user will ever notice, while rushing the one flow that actually mattered. So I've started thinking about vibecoding less as a speed tool and more as a taste filter. The AI handles the typing. The judgment about what to keep, what to throw away, and what to quietly rebuild later is still very much on me. I'd genuinely love to hear how others approach this: When ... ... vibecode something, what's the one part you always slow down and do carefully by hand? Have you ever shipped something fast and been surprised by what users actually cared about (vs

... live on PH Aug 22, so this thread's been good practice at saying it in one paragraph.) Comment from Demetrio Koro(@aktivist): siOsi upload one selfie and it reads your 12-season color type, undertone, gold vs silver, and which hair colors suit you, then grades the actual makeup in the photo (flash cast, oxidation, coverage), which most tools skip. Free, no signup. Built it because every beauty sub has a queue of "what season am I" posts waiting ... ... verified AI disclosure. others will not. over time the receipts with agent transparency outperform on credibility because the alternative

... both content and   product work.   2. Google Search Console I basically live in it. For anything SEO, it's the only honest source of truth   about what's actually happening vs. what I assume is happening.   3. Ahrefs keyword and competitor

... part people argue with: it never calls an LLM. Pure Python, offline, deterministic. Same brief in, same system out. A model improvising is what makes the slop, so I took the model out of the decision and wrote the taste down as rules you can read and switch off. There's a real before/after in the repo: a skip-hire site, generic template versus

... Describe the why, not just the what. Don't tell the agent what to build. Tell it how it should feel. The intention. The user context. I've found that when agents understand the why, they go further . Taste is now the expensive thing. When code is cheap, taste

... Oasis still needs a clear line between  guest  and  signed-in  so people know what to expect. Insight for other users: Guests  can try  limited assistant flows  (good for a quick taste before you commit). Signed-in  users get the full experience:  cloud routing ,  voice transcription ,  chat history in IndexedDB  keyed to your user, and the  plan-aware daily token bar . Typical sign-in moments:  Voice ... ... /docs/welcome-back-after-sign-out  |  https://kahana.io/docs/chat-hist... Questions we would love your take on: Try before sign-up How much should guests get without an account before it feels like a crippled demo? Sign-in prompts When should Oasis block vs

... system. Manus analyzed the database and figured out they were abusing Indonesian free subdomains like: . my.id . web.id It then blocked those TLD patterns automatically. That kind of investigative debugging honestly surprised me. The Honest Downsides You Still Need Product Taste Manus builds what you ask for. If your product decisions are bad, you'll simply get a very well-built bad product. The what to build part is still completely your responsibility. Debugging Requires Good Communication Sometimes a feature doesn ... ... belink.to if you want to check it out. Happy to answer questions about the workflow. Comment from Juan Guillermo Rozo(@juan_rozo17): Great write-up, Trey One thing that really resonated with me was: "You still need product taste

... already be partially automated with AI. Blue-collar work is physically harder to replace at scale because robotics is expensive, slower to deploy, and depends heavily on real-world environments. But I don t think it becomes humans vs AI. More likely: small teams + AI replacing large teams without AI. Comment from Haritha Vijayakumar(@harithavijayakumar): @busmark_w_nika  Yes, that s another huge shift. AI increases productivity, but at the same time it also changes perceived value and pricing ... ... expectations. Suddenly people expect: faster work cheaper work more output Which probably means the long-term advantage becomes less about access to AI and more about: taste

... Yesterday s discussion about who is more likely to be replaced (white-collar vs. blue-collar workers) raised another question for me. So I m asking: What do we need to do to keep our jobs in the age of AI and robotisation? What skills do we need to master and become self-sufficient in and better at than any machine? Just for fun, I thought about how students at the University of Arizona booed Eric Schmidt after he told ... ... because robots may replace physical work soon. Comment from Tony Spiro(@tonyspiro): The people I see thriving aren't the ones who know the most tools. They're the ones who've developed strong taste

... Claude by Anthropic launched their frontier coding models last month, and @sama posted: we still get looksmaxxed on frontend a little but we IQmog hard now In other words: GPT-5.5 has the brains; Claude still has the taste. To verify this statement, the @Kilo Code team gave both models the same 5 UI design prompts - a SaaS landing page, an analytics dashboard, a settings page, a sign-up screen, and a pricing page. TL,DR: Yes. Opus 4.7 produced ... ... details, refine spacing, and polish components. Claude Opus 4.7 shows better understanding of typography, color, spacing, and UI details. The per-token pricing suggests GPT-5.5 is the most expensive model ($30 per 1M output tokens vs