Would a separate verification agent actually solve this problem?
retained text
I'm building trading bots and data monitoring tools with AI assistance and keep running into the same issue. I give the model API documentation, we implement something, test it, fix issues, and eventually it tells me everything is working and we can move on. Then several tasks later something breaks and I discover that an earlier step was never properly verified. Sometimes it's something basic like incomplete API data, pagination issues, missing records, etc. So I'm wondering if using the same agent/context for both implementation and verification is the mistake. Has anyone had better results with something like: Developer agent → implements the task Reviewer agent → independently checks the implementation against the docs Test agent → tries to break it and verifies actual outputs And only then is the task considered complete? I'm curious whether people are actually getting reliable results from setups like this, or if multiple agents just end up confidently agreeing with each other's mistakes.
document fetched 2026-09-30 00:19 · canonical https://reddit.com/r/AI_Agents/comments/1wt68ve/would_a_separate_verification_agent_actually