Most AI agent teams keep the output and delete the part that actually matters: the trajectory. Every tool call, dead end, and piece of reasoning behind an agent's answer is more valuable than the answer itself for monitoring, security, audits, and training data.
New piece from
