Publication

The most useful thing an AI tool can tell me isn't always 'Done.' Sometimes it's: 'Here's what I couldn't verify.' #AIAgents #AICoding

@ganesh2412 Show the failed checks.

Good point, Karina! 😄 And not just “3 checks failed.” I'd want to see which checks failed, what was expected vs. actual, and the supporting logs or screenshots. Also, there's a difference between a check that failed and one that couldn't run at all. That's an important distinction I'm exploring with AgentProof. A useful AI tool should explain what went wrong, not just announce that something went wrong!

@ganesh2412 Especially when the code changed, but the actual user flow hasn't been checked yet.

Exactly, Matoto! 😄 A code change being completed doesn't mean the feature has been verified. The real question is whether the user can still complete the journey successfully, especially after changes to authentication, permissions or data handling. That's one of the things I'm exploring with AgentProof: making the gap between “code changed” and “behavior verified” visible. Until the actual flow is tested, I'd rather see NEEDS REVIEW than a misleading green checkmark!

I'm building AgentProof around a question that keeps coming up in AI-assisted development: Who independently checks whether the result matches what was actually requested? The direction: Clear requirements. Human approval. Independent verification. Evidence-backed defects. Still early. Building and testing one piece at a time. #BuildInPublic #AICoding