When an agent's instructions run out, it fills the gap with its best guess. This is the version of my intent engineering article you fill in for your own agent, so it has less to guess.
Get the framework
Put in your email and I'll send you the framework: a one-page version to read and a file you can load into an AI.
You'll also get Strategic Thinking Weekly: one email when a new issue comes out, sent by me. No drip sequence. Reply "stop" anytime and you're off. I don't share your address.
What you get
- The seven components, each with the question to answer and what a good answer looks like.
- One test that sorts every rule into guidance or a hard limit, and five ways to enforce a hard limit outside the model.
- A score out of 21, with what each level means, so you know which part to fix first.
- A worked example (a support agent that can issue refunds) and a blank template for your own agent.
- The same framework as a JSON file you can load into Claude, ChatGPT or your own agent.
The seven questions
Here they are. The download adds the levels, the test, the example and the template.
- ObjectiveWhat problem must the agent solve, and why does it matter?
- Desired outcomesHow will we know the agent succeeded, without taking its word for it?
- Health metricsWhat must not degrade while the agent operates?
- Strategic contextWhat context shapes a good decision here, and what wins when goals conflict?
- ConstraintsWhich rules are guidance, and which can never be broken? What enforces each one?
- Decision typesWhich decisions can the agent make alone, and which must it hand up?
- Stop rulesWhen must the agent stop, escalate or call it done?
The one most people skip is the fifth. A constraint that matters can't live only in a prompt. The framework shows you how to tell which of your rules need more than that.
Rather not give an email?
Join the Strategic Thinking Academy instead. It's my free community on Skool, where we talk about building frameworks like this one.
Join the free communityNew to intent engineering? Start with the article: Intent Engineering for AI Agents.