... 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 ...

... 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 ... ... useful for everyone. The real value comes from knowing how to use it with your own expertise. Designers + AI can be just as powerful as developers + AI. Comment from Irsa Doham(@irsa_doham): I would out source taste

... 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

... every imaginable problem. If anyone can turn an idea into software before lunch, building is no longer the barrier. Soon, why maintain old software when an agent can simply rebuild it? So what remains valuable: distribution, trust, proprietary data, taste, community or nothing? Are coders becoming superpowered product creators, or are we watching the economic value of code collapse in real time? I m fascinated. And slightly terrified. Which side ...

... 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

... live yet, or it would've been a big issue. Comment from Stacy Wycoff(@stacywycof83995): I hit this same tension but across businesses rather than terminals. I run more than one venture at once and the babysit versus trust problem shows up constantly, except instead of watching an agent I am watching Slack, email, and four different dashboards trying to catch the one thing that actually needs a decision from me. I ended up building FounderFlow partly to solve this ... ... like a similar itch to what you are describing with agents, just one layer up at the business level. Comment from Stacy Wycoff(@stacywycof83995): @aadilghani  I don't run coding agents but I have the 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

... entire roadmap. Comment from Nolan Vu(@nolan_vu): @thenameisarian  Interesting, alternative to LinkedIn you said "Transparency, not hiding" as the positioning is sharp. The whole industry is moving toward pretending AI involvement isn't happening, so calling it out directly and making it a feature instead of a flaw flips the script in a really smart way. Recruiters needing to know what was AI vs

... breaking, and so on. - Vaizle - I work with Vaizle, but it has very easily emerged as my fav platform. Not even lying. I can keep track of ad performance, check what competitors are posting on social. And best part is that I can feed it my own website + competitors

... 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 Nika(@busmark_w_nika): @1mirul  that's true. More people are likely to put something into GPT and have an output. But not so many people ... ... 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 ... ... write emails, code, summarize docs, and generate strategies. But when your bathroom is leaking, your tiles are crooked, or someone needs a great haircut before an important event, you still need a real human with real hands, 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

... generates personalized weekly meal plans (breakfast, lunch, dinner, snacks) in about 30 seconds, tailored to user goals, dietary preferences, allergies, cooking time, and even fridge/pantry contents via photo scan.7ad66c Key Features Instant AI Plans: Built around calorie/protein targets, tastes, family size (up to 6), and restrictions. Regenerates weekly as needs change. Smart Grocery Lists: Auto-generated, aisle-grouped, shareable. AI Coach: Chat for adjustments like "make it keto," "cheaper," "higher protein," or "kid-friendly." Fridge/Pantry Scan (Premium): Snap a photo ...

... usually includes constraints, tradeoffs, and small details that are hard to fake. In B2B, for example, a generic AI post might say, build trust with suppliers. A more human answer would mention the actual messy parts: unclear MOQ, sample vs production specs, certification documents that do not cover the exact SKU, packaging files that are not ready, or buyers comparing quotes that are not based on the same assumptions. That kind of detail does not have to sound perfect ... ... 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

... folks it's been exactly 21 days since i launched PopTask here and y'all dropped helpful comments .. i read every single one v2.0 is basically that list, but built. small taste of what's coming: - real 2-way calendar sync .. google, apple, reminders - voice input .. hold a key, dictate, done - rest i'm keeping hidden till launch day -.- check out what's coming, visit PopTask Website a few things still cooking but launch is close .. like, this week close ...