Most agent prompts are instructions. Do this. Then do that. If X, do Y.

The problem: Instructions break the moment reality diverges from what you anticipated when you wrote them. And in production, reality diverges constantly.

The shift that changed how we build agents at rp1 : stop writing instructions, start encoding reasoning frameworks. Don't tell the agent what to do in situation A — teach it how to think about situations like A, so it can handle A, B, and the edge case you didn't name.

This is the core idea behind constitutional prompting . Instead of a procedure, you give the agent:

A typed contract for its outputs

Explicit principles for how to reason under uncertainty

Anti-patterns it must recognise and avoid

A clear definition of when to pause vs. proceed

The result isn't just more reliable agents.…