I recently built a benchmark for an AI retrieval system. The most revealing cases weren t the ones where the right answer ranked second instead of first. They were the ones where no relevant context existed at all.
A system that confidently retrieves something vaguely related can look functional while being completely wrong. So I added absence cases where the correct result is nothing.
It changed how I judge retrieval quality. Finding context matters, but knowing when there isn t any might matter more.
If you re testing an AI feature, do your evals include cases where the correct answer is I don t know ?
Seeing more AI products move beyond chat into execution: browser agents completing transactions, cloud agents opening pull requests, and analysts generating ready-to-share reports.
The demos are increasingly convincing. The harder question is whether these agents remain dependable when the workflow becomes ambiguous, permissions change, or a step fails midway.
When evaluating an agent, what matters most to you: successful completion, visibility into each action, or the ability to review and recover from mistakes?
Two months and 100 users can mean very different things. I doubt YC cares much about the number alone. The stronger signals are how quickly you built, whether users return without being chased, how deeply they use the product, and what you have learned that others have missed.
If you were evaluating such a startup, what would matter most: retention, weekly growth, founder insight, speed of execution, or evidence that users genuinely need the product? Where would the bar be highest?
I m at the stage where building the product feels easier than getting the first real users.
Not traffic. Not friends signing up. I mean people who genuinely try the product, come back, and tell you what s useful or broken.
I keep hearing the usual answers - Product Hunt, Reddit, X, communities, cold outreach - but I m more interested in what actually worked when you had no audience yet.
Did your first users come from one channel, or was it mostly manual conversations one by one?
I'm building the core feature of my product right now, and I keep noticing the same pattern.
I start on the one thing the product is supposed to do, and while building it, small necessary-looking pieces attach themselves. An empty state needs handling. Errors need messages. Settings need a place to live. None of these feel optional in the moment, but a week later I look up and I've spent most of my time on the branches instead of the trunk.
So my question is about where you draw the line while you're still in the middle of building.
I've spent the last year building writing tools, and the thing that surprised me most is where the difficulty actually sits.
Long-form is easy for a model. You give it a topic and it fills a page. Nobody's day is blocked waiting on a page.
The two-line reply is the hard one. To get it right you need to know what you're replying to, what was said three messages ago, what you already promised this person, and how blunt you're allowed to be with them. That context is scattered across a thread, a file, an old email, and your memory. So the ninety-second reply takes six minutes, and you do that forty times a day.
Early on, marketing is just as important as building the product, especially if you want a successful launch.
Some say the fit in Product-Market Fit is really about marketing. I tend to agree. For me, when I founded my first company, I was more comfortable writing code than copywriting. When I started trying to sell the first product I developed, I realised just how much I didn t know about Go-To-Market (GTM) and marketing. I ve tried everything from cold outreach to reading all the books I could find. After my second company failed, I actually went back to being an operator and only took "business" roles to try and learn as much as possible from doing.
So, if you don t have an MBA or business background, how are you tackling this side of things?
Are you talking to mentors, reading books, taking online courses, using LLMs, or something else entirely?
Whats up with this huge upvote panel taking up half the screen lmao? (The navbar/header is also a bit big but whatev). Also your Launch Guide is messed up some of the markdown isn't rendering and the section titles show the hashtag markup ###Section title, and there isn't any line break between the title head and the content. You're ProductHunt WTH?
Two of my blog posts flipped from "Submitted and indexed" to "Crawled, currently not indexed" this week. No edits, no redirects, nothing changed on my end.
I went through everything GSC actually exposes: robots meta says index,follow, the canonical points to itself, HTTP status is 200. Technically clean by every check available. Google indexed them once, then quietly pulled them back out, with no error and no explanation anywhere in Search Console.
