3 ms·
Having the agent autonomously perform the plan stage is fine; that’s not destructive. It’s the blind application stage without human validation or other safety
by otterley 8mo ago
Having the agent autonomously perform the plan stage is fine; that’s not destructive. It’s the blind application stage without human validation or other safety checks that is the problem.
- stackskipton 8mo agoI mean, apply is not destructive without human in the loop if you don't pass in -auto-approve. In any case, I think spending few seconds typing into your terminal and get yourself in human review mode is mindset improvement if it's not 100% speed optimal.
- otterley 8mo agoAgents are perfectly capable of responding to confirmation prompts. The auto approve flag requirement won’t stop a determined agent if it concludes that’s what the principal desires.