Skip to content

IntermediateLesson 2 of 59 min read

Writing a mission: goal, limits, finish line

A course from the Academy: AI agents: handing them a real task

Go to the quiz ↓

An agent does what you ask, including what you phrased badly. With a chat, a vague request gets a vague answer. With an agent, it can mean twenty minutes of useless actions, or an action you didn't want.

The four blocks of a good mission

  • The goal: the expected result, concrete. Not "look at flights to Lisbon", but "find the three cheapest Lyon to Lisbon return flights from 10 to 14 May, cabin bag included".
  • The context: what it needs to know that you already know. Where the files are, what format you prefer, what has already been tried.
  • The limits: what it's allowed to do, and what is forbidden. "You can browse the sites, you don't book anything, you don't create any account."
  • The finish line: how it knows it's done, and in what form it hands over the work. "You stop when you have three checked options; you give me a table with the price, the times and the link."

Planning for surprises

Tell it what to do when it gets stuck: "If a site asks you to log in or pay, stop and ask me. If a piece of information is missing, note it rather than guess." Without this instruction, an agent tends to improvise to get to the end.

A complete example

"Goal: prepare the guest list for the end-of-year dinner. Context: the replies are in the emails received since 1 November, with "Dinner" in the subject. Limits: you read the emails, you don't send any and you don't delete anything. Finish: a table with the name, the answer (yes, no, no reply) and dietary needs; flag the cases where the answer isn't clear."

Analogy. Sending someone to do the shopping. "Get something to eat" doesn't work. "What we need for a ratatouille for six, thirty euros max, and call me if they're out of aubergines": now you know what will come back.

Key takeaways

  • A concrete goal, context, limits, a finish line.
  • Write down in black and white what is forbidden.
  • Tell it to stop and ask when it gets stuck.
  • Specify the form of the output: table, list, draft.

Key concepts

Complete the lesson

A 4-question quiz: 3 correct answers are enough, plus an exercise to practise. It opens in the Academy, on this same lesson.

Take the quiz and complete the lesson Free. With an account, your progress is saved and leads to the course certificate.

The course: AI agents: handing them a real task

  1. An agent is not a chat9 min
  2. Writing a mission: goal, limits, finish line9 min
  3. Monitoring and checking its work8 min
  4. Security: access, data and traps9 min
  5. Your first agent, step by step10 min

The whole Academy →The AI glossary