Prompt
Agent behavioral contract
Write the explicit contract for an AI agent: what it will do, what it will never do, when it stops and asks.
By Gus · founder — Agentic Design
The prompt
Write a behavioral contract for [AGENT + DOMAIN]. Constraints: [CONSTRAINTS]. Produce four sections: (1) Will do autonomously — specific actions; (2) Will never do — hard limits, with the reason each exists; (3) Will do only with confirmation — and the exact trigger for each; (4) Escalation — the conditions under which it stops and hands back to a human, with concrete thresholds. Write it so a designer and an engineer would build the same thing from it.
How to adapt
Before you design an agent's screens, design its promises. A behavioral contract is the plain-language spec of what the agent is allowed to do autonomously, what is always off-limits, and what triggers a hand back to the human. It becomes the reference every downstream design decision is checked against.
How to adapt
Swap [AGENT + DOMAIN] for yours and list any hard regulatory or safety limits in [CONSTRAINTS]. The escalation triggers are the highest-value section — push the model to be specific about thresholds, not vague about "risky actions".