For builders and agencies
How to write a mission for an AI agent
Most AI agents do not fail because the model is weak. They fail because nobody wrote down what "good work today" means. A mission fixes that: a short brief the agent reads every round, with one goal, clear daily tasks, hard limits and proof. Here are the five parts, and an example you can copy the shape of.
The five parts of a good mission
- One goal, in one sentence. Written so an outsider understands it: "Every new lead for the webinar gets a personal reply within the hour." Not three goals. If you need three, you need three agents.
- Daily tasks you can count. "Reply to every new lead" or "one post live" can be checked at the end of the day. "Grow the brand" cannot. The agent should know exactly when today is done.
- Hard limits. Which accounts it may use, how many messages or posts per day, what it may never do (pay, delete, promise discounts) and when it must stop and ask a person.
- Proof for every step. Each finished task leaves a link, a screenshot or a file. "Done" without proof does not count, for the agent and for you.
- A checklist. A short list of open and finished items the agent ticks off. It keeps long-running work on track across days and shows you at a glance where it stands.
An example mission
- Goal: every person who signs up for the free training gets a friendly, personal reply and a reminder before it starts.
- Daily task: all sign-ups of today answered, with the sent message saved as proof.
- Limits: only the company inbox, at most a set number of mails per day, no prices or promises that are not on the website, hand angry or unclear messages to a person.
- Proof: a log line per reply with the time and a link to the sent message.
- Checklist: reply template approved, inbox connected, test round passed, first day watched.
FREELearn to build & sell AI agents with Jarvy
Reading is step one. Jarvy's free course turns it into action: every lesson ends with one concrete task.
- Pick one boring task to automate
- Build your first agent, no code
- Answer the 4 objections clients raise
- Run a paid 14-day pilot
Common mistakes when writing a mission
- Writing steps instead of a goal. If you list every click, the agent breaks the moment a page changes. Give the goal and the limits; let it choose the steps.
- Goals you cannot measure. "Be more visible" gives the agent nothing to aim at and you nothing to check.
- No stop rule. Without "when in doubt, ask", an agent guesses. Guessing with a customer is expensive.
- Never updating it. Every time you correct the agent, put the correction in the mission so it holds tomorrow too. More in 7 mistakes when building AI agents.
From mission to live agent
- Write the mission and the daily task first, before any tool or login.
- Run one test round where nothing is sent, and compare the result with the mission. See how to test an AI agent before it goes live.
- Go live with low limits and check the daily task with proof at the end of the first day.
Questions
What is a mission for an AI agent?
A short written brief that tells the agent what it is for, what counts as done each day, what it may and may not do, and how it proves its work. The agent reads it at the start of every round.
How long should an AI agent mission be?
Short enough to read in two minutes. One goal, one or two daily tasks, a handful of rules and a checklist. Long missions get skimmed by people and diluted for the agent.
What is the difference between a mission and a prompt?
A prompt is one instruction for one answer. A mission stays the same across many rounds and days: it sets the goal, the limits and the proof, while the agent decides the steps inside it.
See agents with real missions at work in the free training
The training shows agents with missions, daily goals, proof and dashboards, and how to package them for clients.