... 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 ...

... design problem explains zero interest in the thing itself, and solo makers usually spend the design money before they know which one they have. adana, of the products you've seen struggle, how many were actually a polish problem versus nobody wanting it? Comment from Sansa Grey(@sansa_grey): One thing i have learned that users dont care that the UI took you two days to perfect. they care that they found the button they needed in five seconds. Comment ... ... running a few SaaS products solo with Codex that handles the UI/UX and everything for me. Comment from Adana Marukhyan(@adana): @jordan_bulk   @evgeny_yakushev  What an interesting product! Does it have competitors

... 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

... humans, but it will make average work much harder to sell. Average copy. Average designs. Average code. Average research. Average support. Average strategy decks. Average I ll do what the prompt can already do services. But people who understand taste, context, trust, distribution, customer psychology, product judgment, and real-world constraints may become more valuable, not less. So maybe the real question is not: Will AI replace humans ? Maybe it is: What kind of human work will still be worth ...

... 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

... Ghani(@aadilghani): @virko_kask   Agent-state instead of task-type as the urgency signal is better than what I had, and I think you just moved my whole priority model. Category decides if it may reach me, idle-vs-busy decides how loud, that stacks cleanly and it's more honest than any static rule because it measures actual wasted time, not assumed importance. The critical-path piece you can't automate is a genuine third axis ... ... hard for a precise reason: idle-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

... Cursor: coding Supabase: backend and auth Vercel: hosting GitHub: version control Google Search Console: SEO Campix: my own project, camp software, yes I'm biased Comment from Hans Desjarlais(@ismaelyws): Gmail Calendly Claude Code Google Docs & Sheets VS Code Comment from Bill Davis(@billspaced): @nolan_vu  Slack, not because I love it but because everything dies without it. YESSSSSSSSSSSSS Comment from Bill Davis(@billspaced): @riya_pariyar   @benfleming   @asapx  same question... never heard ... ... solo founder, these are the five tools I use almost every day: ChatGPT VS

... worth ad dollars at a tiny budget, and which are a money pit? I'm doing daily build-in-public content on TikTok, YouTube Shorts, and X, but I honestly can't tell yet what's driving downloads versus just views. If you've cracked early distribution for a consumer app, I'm all ears. Product Hunt Launch: https://www.producthunt.com/prod... Comment from Sajid Javid(@sajid_javid): @nolan_vu  Thanks, Nolan. That's a really helpful perspective. I originally ... ... verified AI disclosure. others will not. over time the receipts with agent transparency outperform on credibility because the alternative

... 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

... similar way. It will remove or compress many repetitive office tasks, raise the bar for junior roles, and make one person capable of doing more. But it does not automatically replace judgment, accountability, client trust, domain expertise, taste, or decision-making. Blue-collar work feels even safer over the next 10 20 years, especially skilled trades. A robot might work well in a controlled warehouse, but repairing a roof, renovating a bathroom, fixing plumbing, dealing with an old building ... ... similar is happening again now so time will tell I guess. Comment from Alper Tayfur(@alpertayfurr): I think white-collar work gets affected faster because most of it already lives inside software. But the real split is not blue vs

... 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 ... ... impressive my CV could be, the current job market is not reliable. 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

... instead of building another shopping app we started building a personal commerce agent Velvet learns your style from what you actually wear, and turns that into an agent that can act on your behalf. something that understands your taste, styles you, and can act on your behalf. Virtual try-on was just step one. Now it can: Learn your style from your closet Style outfits for you Let you explore looks through a social layer with your friends and taste

... them, photo profiles were special in a similar way. When I double checked the profiles, the links were not working. And I hate finding myself interacting with bots. What is the limit that bothers you in interacting with AI vs what feels fine? A/ Interacting with a bot B/ Interacting with content being 100% generated C/ The person giving the idea, and asking AI to write an answer. D/ Asking AI to rewrite E/ Correcting with AI It was funny ... ... safe and always AI-compatible. The thing that can get you fired if you apply it to the wrong deal that's the signal. On images and video: agree with Saad that effort is the better frame than AI vs

... iterations. In my experience, they re great for exploration and saving time, but they also tend to push outputs toward similar patterns and solutions. It feels like the more we rely on them, the more important our own taste and judgment become. Personally, I don t see AI replacing designers just like having a camera doesn t make someone a photographer. The tool helps, but the outcome still depends on how you use it. What s been interesting ... ... early exploration, copy variations, quick UI directions, and summarizing research. That part is genuinely useful. But I ve noticed the same thing you mentioned: the outputs get solid fast, and also a bit same-y fast. That s why taste