... Ben. I ve spent 15+ years working with low-code, collaboration and content management systems in government. A couple of years ago, I started experimenting with AI to see how far I could take an idea of my own. A few restarts and a lot of scope expansion later, I launched Lunar Paper into public open beta. It s a self-hosted platform with 19 applications covering project planning, forms, automation, dashboards, documents and creative tools. Most support live collaboration ...

... real provider and at our sandbox and diffs the result. It's already caught four real drifts in our own simulation. @vansh36 flagged the favicon was still the Next.js default. Fixed. Half done: @maschiojv , @hi_i_am_mimo , @omri_ben_shoham1 and @priya_kushwaha1 , out-of-order events. The permutation fuzzer is built and Paddle has the patterns. What isn't wired yet is the end-to-end path where you describe the ordering bug and it reproduces ... ... demand. Not going to say it's done when I just checked and it isn't. Not done : @sabber_ahamed , per-event latency jitter. @omri_ben

... other makers too, or if I m just unusually deep into AI tools. For those building SaaS products: are you seeing people discover you through AI yet? Or is Google still doing most of the work? Comment from Ben Cohen(@bencohen): @rabnoor_s   @oshylabs  @oshylabs the signup wording is a pretty interesting clue, even if it s messy to prove. Makes me wonder what people would write if we asked how was the product described to you? instead ... ... fair point on llms.txt. I probably gave it too much credit in my earlier reply. I d want to see the same site before and after adding it, since comparing different products would be hard to untangle. Comment from Ben

... purpose, and write down why. Not because Cloudflare's defaults are wrong, mostly they're reasonable, but because a default you never chose becomes a strategy by attrition. This one has a date on it. Comment from Omri Ben-Shoham(@omri_ben

... other makers and users think about this. Comment from Adana Marukhyan(@adana): @kshitij_mishra4  As I understand it, you think that having everything in one environment could reduce your problems ? =) Comment from Philip Roigard(@philip_roigard): @omri_ben_shoham1  I'm finding I don't want to learn or use UI anymore I reach for connecting my agent and having it do it for me. I still build the UI for humans but I feel having an agent ... ... setup time but doesn't actually delivery the automation after. Curious though, do you think most users can tell the difference upfront, or does it only become obvious once they're already invested in the setup? Comment from Omri Ben

... logsheet.ch and the app is LogSheet on the App Store. Happy to put whatever numbers I have in the comments including the unflattering ones. If you want to see the app and leave a score! apps.apple.com/app/id6795027347 Thanks Ben ...

... Comment from Evelyn Clark(@evelyn_clark1): Cool story. I'd bet it's user-base dependent too different crowd here, technical folks lean into raw/debug tools more than a general audience would. Comment from Venkat A(@venkat_a2): @omri_ben_shoham1 Mine is a feature that refuses to do the thing the product is sold on. We turn a customer interview into a written case study. Somewhere in there I added a check that judges whether the conversation actually produced ... ... hole was worth more than the compliment. My theory on why this happens: the safeguards are the only claims a stranger can actually check. Everything else is a capability, and capabilities all sound the same. Comment from Omri Ben

... everyone, I m Ben. I run two businesses, and somewhere between meetings, driving, client work, and everyday life, I m constantly having ideas or remembering things I need to do. Writing something down was never really the problem. The problem was what happened afterward. My notes ended up scattered across different apps, transcripts, documents, and random lists. Even when an AI tool cleaned up or summarized a recording, I was usually left with another large block of text that ... ... regularly You can see it under my launched apps I m happy to answer questions about the product, how I built it, or what I ve learned trying to launch it while running two other businesses. Comment from Ben

... Hermes. I m curious. Which agent harnesses are you currently using? How do you use them differently? Am I missing any that you would strongly recommend? Looking forward to hearing everyone s thoughts! Comment from Kuanze Ma(@kuanzema): @omri_ben_shoham1  I like its heartbeat and scheduled execution. It s useful for routine operational work like monitoring and summarizing the news without needing constant supervision. That said, I wouldn t give it tasks where mistakes are unacceptable or irreversible ... ... Comment from Akarsh Hegde(@akarsh_hegde): @kuanzema  Storing skills and decisions in Git makes sense, especially when switching between harnesses. How do you prevent stale decisions from being reused after the codebase or requirements change? Comment from Omri Ben

... Jake Moshel(@jake_moshel): Disclosure first: I build getnovix.ai which is an agent product we always end with a PR. Done. So it has to be approved and merged by a human, simple as that Comment from Omri Ben-Shoham(@omri_ben

... first. we now throw every ordering + every duplicate at your handler and hand back the exact sequence that broke it. which of these has bitten you? (i've shipped #1 myself. be honest ...) Comment from Raj Nagulapalle(@rnagulapalle): @omri_ben_shoham1  yeah that's exactly the wall i keep running into when people ask about live calls. everything i do assumes i can re-fire the exact webhook ordering because it was captured as typed events in the first ... ... interaction was capturable as replayable state to begin with. where it isn't, the leverage moves upstream to live capture. genuinely different problem. feels like our two use cases meet right at that line Comment from Omri Ben

... point your AI agent at to see if it catches it. ~20 min, no accounts or keys. https://github.com/fetchsandbox/... I want the "this didn't work" reports as much as the wins. Ask me anything. Comment from Omri Ben-Shoham(@omri_ben

... find engaging in threads like this actually translated into day-one support, or is it more about the DMs/relationships that come out of it?" Comment from (@new_user___26020267b63ea4605738cbb): Launching BillFast today has reinforced the ready enough vs. finished distinction for me: the core flow (create an invoice, preview it, download a client-ready PDF) is narrow and usable, while feedback is now surfacing the edges currencies, taxes, templates, and payment instructions. For a small product, that feels ... ... would be: A stable core product A clear landing page and positioning Real users who have used the product before launch Basic analytics to measure what happens after launch A strong reason for people to choose this product over alternatives

... created a couple of agents with claude code and created a couple of agents with codex to cross check, review, continue and approve the work. Only tag me for critical decisions or blockers Comment from Kidder(@hellokidder01): @omri_ben_shoham1  Yeah, the "thinking vs

... feature reason before I'll even let the agent near it, tests or not - because the failure mode isn't "tests fail," it's "tests still pass but the edge case that mattered isn't covered." Comment from Omri Ben-Shoham(@omri_ben

... True. As much as I'd like to imagine these people don't understand the problem/solution and it falls into communication, the truth is just that haters are gunna hate sometimes. Comment from Nika(@busmark_w_nika): @omri_ben_shoham1  That's why I am trying to ask satisfied users for reviewing my solution :D Comment from Nancy Chu(@nancy_2choo): As a small team, we've started looking at comments a little differently. Instead of asking whether they ... ... This is garbage, it hung up on my mother" is rude and actionable. "Love the vision, excited to see where this goes!" is lovely and worth nothing. Second, the distinction that changed how we respond: hate at your product versus

... needed. The rough stuff is the useful stuff, and we read all of it. And if you're just finding us today: ask us anything in here. Omar and Ahmad (CEO/CTO) are both in the thread. Comment from Omri Ben-Shoham(@omri_ben

... teaching me that you don t need to understand your own codebase deeply to build something people love. You just need to understand the problem. Anyone else navigating this weird middle ground? Comment from Mona Truong(@monatruong_murror): @omri_ben_shoham1   yes, caught it once so far. the agent described a flow where data went through a validation step before saving, and it sounded completely reasonable. what tipped me off was the UI the error message I expected from ... ... your job, and a colleague is doing theirs. In your case, it could just as well have been a freelancer, and that would instead be considered a compliment to your management skills. Comment from Mona Truong(@monatruong_murror): @omri_ben

... level but I'll be amazed if there's ever a 100% watertight solution to this that doesn't just remember, but uses the context sensibly too (like knowing exactly what to degrade over time etc.) Comment from Omri Ben-Shoham(@omri_ben

... report, it's a support ticket and a refund. what's yours - the one category of change you still gate by hand even after trusting the agent with everything else around it? Comment from Sabber Ahamed(@sabber_ahamed): @omri_ben_shoham1  same silent-failure logic is why anything writing to a health record gets a human pass here too - a wrong value in a checkin log doesn't throw, it just sits there quietly wrong until someone acts ... ... apart. In general, I think it's really important to build a system where the human in the loop shows up in the right, necessary, and important places, instead of handing everything over to the agent. Comment from Omri Ben

... explain when something went wrong? - Has this changed how thoroughly you review AI output before it ships, or has trust just gone up over time? Comment from Gal Dayan(@galdayan): I say "still investigating" openly, every time. tried the alternative once early on, waiting until I had the full trace before saying anything, and it backfired worse. the silence reads as either not knowing or hiding something, and with a non-technical stakeholder both look the same from the outside ... ... here's what I found after digging" becomes the honest answer, and that's exactly where I've started forcing smaller PRs regardless of how fast the agent could theoretically ship the whole thing at once. Comment from Omri Ben

... what it actually does: It sees your screen and acts on it Not summarizes. Not suggests. Clicks, scrolls, fills forms, and navigates any page, on its own. It builds Google Sheets and Docs from scratch "Make me a competitor pricing table with graphs." It opens the tabs, pulls the data, formats the sheet, builds the charts. You just watch. It runs WhatsApp & Telegram like a personal assistant Broadcast to 50 contacts at once. Send follow-ups with Word docs ... ... attached. Message from your phone, get a finished file back. Nobody else does this. It monitors pages and pushes you alerts Price dropped? Ticket available? Competitor

... audit and review it and I analyse the review output. It's definitely easy to get comfortable especially as these models get better and better but knowing how to code is a valuable skill to have. Comment from Omri Ben-Shoham(@omri_ben

... working on exposing an eval environment configuration. The use cases we discovered The ones we expected: simulation for learners, architecture validation, cost estimation. The ones the community surfaced: Architecture reviews before anyone writes an infrastructure ticket. Omri Ben-Shoham @omri_ben

... into the next. Curious if others have hit a similar wall. What was the first thing in your vibe-coded project that the AI couldn't cleanly fix, and how did you actually handle it? Comment from Omri Ben-Shoham(@omri_ben