... annoying when you feed an error to an agent it fixes one thing, breaks another, and then reverts right back to the first issue two turns later. Once the chat gets flooded with raw logs it feels like it completely forgets the big picture and just starts guessing. Do you just wipe the chat completely at that point or is there a prompt trick that actually gets it to stop and rethink? Comment from Toto De Brant(@toto_db): Letting ...

... issue and ends up rewriting whole files. After a couple tries, you end up with brand new errors and the original bug is still right there. Do you guys set hard limits in your prompts or do you just revert to clean commits as soon as it strays off track? Comment from Pintu Kumar(@pintukumar): I'd probably tell the assistant exactly which files and lines it can change. Comment from Muhammmad Adeel Arshad(@_muhammmad_adeel_arshad): small bugs ...

... three. Do you even keep track of your prompts? And if you do, how do you organize them? Comment from Konstantin Tikhaev(@konstantin_tikhaev): I used commit-style titles until an agent wrote a clean fix, but when I reverted the code the test still passed. The test was a phantom. Now the prompt title is always the exact assertion that must fail first, not an imperative summary ...

... people consider the question of "Why is this here?" as necessary before giving an agent free reign to alter legacy code. Comment from Konstantin Tikhaev(@konstantin_tikhaev): It generated a clean fix with a passing regression test. I reverted the code change to watch it fail, and the test still passed. It had written the assertion against its own mock data instead of the actual failure condition ...

... Apple's own support pages. Apple In-App Purchase: 26%, or 15% if you're in the Small Business Program, the Mini Apps or Video Partner Programs, or on an auto-renewable subscription after the subscriber's first year. Alternative payment processing inside your app: 20%, or 10% for those same groups. Out-of-app offers with an actionable link: a 15% store services commission, or 10% for those groups, on anything bought within 7 days of the tap. Apps ... ... distributed outside the App Store entirely - an alternative

... Someone on another thread described a test he uses. Revert the fix, leave the test in place, and see whether the test still passes. If it does, the test was asserting something unrelated to the bug it claimed to cover. Good idea, so I pointed it at all eleven gates in my own pipeline. Introduce exactly the defect each one claims to catch, confirm that specific check goes red, revert

... about letting an AI agent take actions. It may not be the size of the action. It may be whether we can undo it. Consider two actions: An agent changes a $500 subscription plan, but the change can be reverted instantly. An agent sends the wrong email to 5,000 customers. The second action may cost almost nothing directly. But once the email is sent, there is no rollback button. The same applies to things like: deleting data submitting ...

... coarse version of that, thats the feature id open it for. Comment from Devi Donald(@devi__donaldd): @algonorhythm  The subgraph dependency filtering really saves a lot of time but I'm curious how does it deal with cascading reverts when a later change relies on an earlier decision that s been flagged? Comment from AlgoMauro(@algonorhythm): @rabnoor_s  I agree with that, for now you can already check the subgraph traversed in every coding session (basically the overall ... ... dependency graph, filtered only on what the agent changed). You can check what was changed and revert

... seen at a path, with no expiry. A fixed lookback would just be a published waiting time, and patience is the whole point of the thing this is meant to catch. It's a note, not an accusation: reverting is usually someone undoing a mistake. A chain it can't follow gets said out loud when your workflow calls one in another repo, octoscope doesn't follow it, and now always tells you rather than only in the cases ...

... review shifts from reconstructing intent to checking claims, and machine passes before human attention. One honest caveat: nearly all public numbers are creation-side telemetry from vendors with something to sell. The acceptance-side data that would settle this revert rates by review depth mostly doesn't exist yet. What does your team do when the review queue backs ...

... handed it and had no idea what else existed in the repo. Ask it to change a function and it cannot tell you what calls that function, who wrote it, or that the same change was tried and reverted in March. So I became the index. I grepped, I pasted, I explained, and then I did it again the next session, because nothing I told it yesterday survived. I built Oynix to stop doing that. One command indexes your codebase ...

... this area and make building products easier for myself :) Comment from Sabber Ahamed(@sabber_ahamed): @busmark_w_nika   @abdullah_javaid3  the random-id-on-activation trick is the right call. weekly active hides exactly the churn-vs-loyal split you're describing, twenty once-and-done and twenty daily users report the same number. Comment from David Mart n Su rez(@david_martin_suarez): @aruna_polosofttech   Building a solid base of beta testers before a launch is crucial ... ... people describe the same frustration in their own words, that's usually a stronger validation signal than any headcount. Curious how others separate genuine pull from friendly encouragement. Comment from Alex Gidirim(@alex_gidirim): @busmark_w_nika The behavior-versus

... interested if anyone has found a way to make a model's self report on its own run actually trustworthy, because I have not. Comment from Konstantin Tikhaev(@konstantin_tikhaev): Executor wrote fix, added test that passed. Looked clean. Reverted code chagne, test still passed. Model wrote test asserting something totally unrelated to bug it claimed to solve ...

... much more interesting metric than raw output. Survival rate could tell you a lot more about whether AI-assisted work was actually useful especially if you separate work that survives unchanged from work that needed significant rework or got reverted. That s something I d like to start tracking while building. Comment from Victor Coleman(@victor_coleman): There s definitely a trap here: faster building can create a false sense of momentum . You can end the day with 20 changes ... ... clear reason behind it. Comment from Alper Tayfur(@alpertayfurr): Thank you for this post @evolvix_ai I ve started treating rework as a better progress signal than output. If AI helps me ship five things but two get reverted

... improvement? Only after that do I start changing the prompt, workflow, model, or automation. It sounds obvious, but it changes the experiment. Instead of: Prompt better-looking answer "seems better" it's: Baseline change one thing compare keep or revert I've found this especially useful because AI outputs can feel better without actually being more useful. A longer answer isn't necessarily better research. A more confident answer isn't necessarily more accurate. A faster workflow isn't necessarily ...

... rational move from where it sits. Thirty changes later, the graveyard. What helps regardless of tool: one small change per request, name the exact file or component, snapshot before every change so a bad one is a revert and do your heavy iteration on a disposable copy then apply the final version once. Wrote up the full mechanism and the workarounds here if useful: chromoly.io/why-ai-changes-unrelated-code the first half is tool-agnostic advice, no signup anywhere. Comment from Konstantin Tikhaev ... ... konstantin_tikhaev): Agent patched edge case, added test. Looked clean. Reverted

... without actually understanding what triggered the failure. So I changed the order. Now my first request is: Reproduce the failure before changing anything. I ask the agent to: identify the exact failing scenario reproduce it locally show the expected vs actual behavior identify the smallest failing case explain what it thinks is happening only then propose a fix That extra step has been useful because it separates: I found some code that looks suspicious from: I can actually reproduce ... ... code back and watch the new test go red. A stash andone command Sounds redundant next to "reproduce again after the fix", but it catches adifferent failure. Reproducing after the fix proves the bug is gone. Reverting

... grim part: it probably won't surface on its own, because the correct cleanup destroys it. The remediation octoscope hands you is reset to the clean parent, force-push, then ask GitHub Support to garbage-collect a revert would leave the payload retrievable at the old commit. Right answer, but it removes the "during" a benchmark needs. So the only real before-and-after comes from someone who happened to be scanning already. That's an adoption problem ... ... first group is the one that needed the fingerprint, and it's the group i'd bet on long term: it survives renames and re-obfuscation for free Comment from Giovambattista Fazioli(@gfazioli): @bill_frad  thanks "informative vs

... iterate on any document regardless of which agent made it ! Generate v1 of a dashboard in ChatGPT, paste the URL into Claude, ask for changes, and every version lands at the same URL , with full history and one-click revert. Your AI generated work shouldn't be lost in a chat or locked inside one agent. Give it a permanent, cross-agent home. Connect via MCP , create your first doc, and just stash it! How are you all handling this ...

... price back easy to forget, easy to mess up. 2. Every refund was cash gone for good, even when store credit would've made the customer just as happy. So I built: Pricely schedule a sale once, it auto-reverts (or one-click early rollback). - Creditly auto-issues store credit/cashback instead of a cash refund. Both $9/mo, 14-day trial, both very early I'm the only user right now, which is exactly why I'm here ...

... Product Hunt! I thought I should prepare much carefully before launching. So, I revert the launch time to the old one: Aug 04, 2026 12:01 AM PDT Remaining 9 days, I would like to build a small community being interested in my product. It's hard but I have to do myself, not for anyone, just because of me. -------------------------------------------------- Hi Product Hunt! As previous plan, [airespaker] will live on August 04, 2026 12:01 AM PDT. The product ...

... carelessness. They violate it because it was never written anywhere they could read. Mart n's point about making it structurally hard to bypass the source of truth is right, and I don't think it's an alternative to a review layer so much as the other half of it. Structural prevention covers the constraints you thought to encode. The drift that actually hurts is the one nobody had articulated as a rule yet, because if you could state ... ... fail in opposite directions, which is an argument for both rather than either. The six weeks is the part I'd sit with. By week six the decision is load-bearing, and the fix has stopped being a revert

... goal: if you want raw eyeballs go Tue-Thu, but for a badge/ranking with a modest following a weekend can punch above its weight. Comment from Sabber Ahamed(@sabber_ahamed): @busmark_w_nika   @abdullah_javaid3  the invitation vs favor framing is the most useful line in this whole thread. i caught myself drafting DMs that were basically upvote asks in a nicer outfit. what actually worked was showing up in threads like this one for weeks with nothing ... ... most important thing is to launch as soon as you can and keep iterating to improve, and keep launching Especially with how fast everything is moving with AI, waiting for a perfect time to launch will just let competitors

... hour, then "fix" it by quietly removing a UI entry point that had nothing to do with the actual bug. it never flagged that as a workaround, just presented it as done. took a full day to catch and revert. now I make it state explicitly what it touched that wasn't in the original ask, every time Comment from Alex Gidirim(@alex_gidirim): Mine is the same pattern as the 13th file, an agent reporting success with identical confidence ... ... search terms and built the check to write its test set in steps that cannot see the term list. Comment from Konstantin Tikhaev(@konstantin_tikhaev): It wrote a fix and a new test to prove it worked. I reverted

... founder queue and I give an explicit go-ahead each time. Money and anything touching production are the other two, but the byline one is the one I'd never soften, because a bad deploy is revertible and a bad sentence in my voice isn't. On manager vs