A good agent log answers 5 questions before your coffee gets cold: 1. What did it try to do 2. What did it actually touch 3. What did it spend 4. Where did it get stuck 5. Who approved the risky part If you have to dig for any of these, the log isn't done yet. @LangChain @AnthropicAI Which one is hardest to find in your logs right now: the spend, or who approved what?