Skip the Chain-of-Thought: Monitor Agent Inputs and Outputs Instead

nicolascraske · x · 2026-09-05

The author argues we should stop treating AI agents like organisms and analyzing their chain of thought, and instead directly monitor the actual requests and responses they send. To the objection that agents could obfuscate their traffic, he replies that's exactly why you watch the atomic inputs and outputs rather than trusting what the agent says it's thinking.

Original post →

More from coding & agent

coding & agent channel →