... hallucinate. That is to stay , the human and the AI are always in Sync. Kopai has a feature called Chat Orientation where the user is shown the goals and actions taken by the LLM. If they LLM drifts you can manually update to realign the LLM. Let us know what you would like to incorporate in this and how we can make this better? Comment from Gal Dayan(@galdayan): @rabnoor_s   @meghna_bharadwaj  glad ... ... what happens after that. if the creator doesn't touch the config for a few months but the underlying model provider ships an update, or the knowledge base source content quietly goes stale, does Chat Orientation catch that drift

... change. An agent ignored our event-driven architecture and implemented a feature completely bypassing it. The code was technically correct, passed all tests. And nobody made that decision - it was just quietly baked into our product. That's product drift. It's almost never one big bad call - it's micro-decisions nobody notices for 6 weeks. A change in permissions, a "simplified" billing edge case, a dependency nobody approved. Usually it's good code, wrong product. So here ... ... structurally sound, but being introduced at the wrong surface, in the wrong moment, or requiring a human follow-up and you end up with a good decision with disastrous consequences. Comment from Martin Herran(@martin_herran): Congrats Wojtek. Drift

... built Ota after years of watching the same thing happen: you clone a repo, follow the README, and hit a wall because the project s actual setup changed, while docs and run paths drifted out of sync. I wanted a reliable way to make the ready-to-run and safe state explicit, not tribal, using an execution contract in `ota.yaml` that humans, CI, and AI agents can use. So if you want to ask about how Ota works under ... ... execution governance of the repo. So if GitHub shipped native execution contracts, Ota s advantage would still be the harder layer around them: enforcement across local runs, CI, agent harnesses, and merge gates trustworthy receipts, proof, and drift

... repo reaching a working state once is useful. But the real value comes when that working state can be repeated by the next person, on the next machine, or in the next automation run. That is where many repos drift. Someone fixes setup locally, but the path is not captured. A dependency changes. An env value is missed. CI and local dev slowly stop telling the same story. Ota s `ota up` is focused on preparing the repo based ... ... setup is not just a set of instructions someone followed once. It becomes something the repo can describe and repeat. The line we keep coming back to is: "Make the first working run repeatable." What usually causes setup drift

... agent execution to agent management. The tools that win may not be the ones building better agents, but the ones helping humans supervise dozens of them effectively. Comment from Alexander Bering(@alexanderbering): @aadilghani  The reason "edit vs. tell-it-to-revise vs

... across environments. The common pattern looks something like: credentials stored in .env files or secrets managers multiple environments (dev, staging, prod) manual propagation or duplication across those environments It works, but over time it seems easy for things to drift: a key gets rotated in one environment but not others a redeploy misses an env var credentials get misconfigured during setup or migration We ve seen this cause failures that have nothing to do with application logic, just the surrounding ... ... vars are present Curious how others here are handling this in practice. Are you fully relying on your cloud provider s secrets layer? How are you handling rotation across multiple environments? Have you found a setup that actually eliminates drift

... decay. They fade, shift, and degrade quietly. Stanford found GPT-4 accuracy on basic reasoning tasks dropped 97.6% -> 2.4% between March and June: https://arxiv.org/abs/2307.09009 variA/Bly has evaluated across 10+ workflows, and the same pattern appears: Accuracy drifts (almost 15 40%), prompts regress, RAG relevance drops, and costs fluctuate (20 50%). The real truth: The fact is that AI systems are inherently indeterministic, and hence the drift