Mission too broad: how to break it down for your AI agent

You've created your AI agent, you've given it a mission, and yet it's going in circles. Its responses are vague, it forgets elements, or it does one thing while neglecting others. The problem rarely comes from the agent. It almost always comes from the mission itself: too broad, too vague, impossible to express in a single clear sentence.

IWhy a Mission That's Too Broad Paralyzes an AI Agent

An AI agent is not an employee who guesses what you expect. It takes your mission literally and tries to do everything at once. When you ask it to process orders, follow up with late-paying customers, and archive correspondence, it doesn't know where to start. It also doesn't know when one task ends and another begins. Result: it does a little of each, poorly, or it gets stuck waiting for clarification.

IIThe breathing test: is your mission too broad?

Take the mission you wrote for your agent. Read it aloud, all at once, without stopping. If you need to catch your breath between two ideas, that's a sign you've put multiple missions into one. Each natural pause marks a boundary between two distinct tasks.

IIIConcrete example: the impossible e-commerce mission

An online store manager writes this mission: "Process new orders by checking inventory, follow up with customers who haven't paid for more than seven days, and archive all correspondence in the month's folder." Three action verbs, three different contexts, three different work rhythms. The agent doesn't know if an urgent order takes priority over a payment follow-up, or when to archive.

Single missionThree separate missions

Process orders + follow up + archive

Agent 1: process orders

The agent hesitates over priorities

Agent 2: follow up on unpaid invoices at day 7

Inconsistent results

Agent 3: archive emails each evening

By separating, each agent has a single responsibility. The orders agent doesn't worry about follow-ups. The follow-up agent doesn't touch archives. All three can work on the same account, with the same token pool, but each knows exactly what they need to do.

IVThe Method for Breaking Down a Mission That's Too Broad

1. 1
   Write the raw mission

   Write down everything you expect from the agent, without holding back. Let the complete list come out, even if it takes ten lines.

2. 2
   Read aloud

   Read straight through. Mark each place where you catch your breath or change your mind. These are your cut points.

3. 3
   Isolate each action

   Each main verb becomes a candidate mission. Process, follow up, archive: three verbs, three possible missions.

4. 4
   Rephrase in simple sentences

   Each mission fits in a sentence of fewer than twenty words. If you can't manage it, break it down further.

VShould I create multiple agents or a single one with multiple missions?

The answer depends on the link between the tasks. If they involve the same data and the same rhythm, a single agent with well-separated missions may be enough. If the rhythms differ or the contexts are distinct, multiple specialized agents work better. An agent that processes orders every hour and an agent that follows up on unpaid invoices once a week don't need to be the same entity.

VIPractical case: an accounting firm that wants to automate everything

An accountant writes: "Retrieve customer invoices, sort them by date, verify amounts, follow up with customers who haven't sent their documents, and prepare a monthly summary." Five actions, five different moments. Reading aloud, they stop after "date," after "amounts," after "documents." Three cut points. They eventually create two agents: one that sorts received invoices and verifies amounts, another that follows up with those who are late. The monthly summary, they keep for themselves, because they want to add their own analysis.

VIIHow to Know if the Breakdown Is Sufficient

• It fits in a sentence of fewer than twenty words
• It contains a single main action verb
• You can read it without catching your breath
• A colleague understands what the agent does in five seconds
• The trigger is clear: an email received, a specific time, an event

Check what applies to you: the list remembers your answers on this device.

If you check all five points, your mission is ready. If you hesitate on even one, read aloud again and find where to cut. The clarity of the instruction determines the quality of the agent's work.

VIIIWhat You Gain by Breaking Down Your Mission

An agent with a clear mission works faster, consumes fewer tokens, and makes fewer mistakes. When it makes a mistake, you know exactly where to intervene. You can adjust one mission without touching the others. And most importantly, you can pause one agent without stopping the others. That's the difference between a tool you control and a black box doing things you don't understand.

?Questions we're asked

How many agents can I create with a single subscription?

Starting from the $19 excl. VAT per month plan, you can create multiple agents. It's the token pool, shared by all agents on the account, that limits total activity. Three agents that work little consume less than one that works a lot.

Will my three agents step on each other's toes if they work on the same mailbox?

No, if each agent has a distinct mission. The orders agent processes order emails, the follow-up agent handles unpaid invoices. They don't touch the same messages. Clear missions prevent conflicts.

Can I merge two agents later if I realize they're doing the same thing?

Yes. You can pause an agent and transfer its mission to another. Each agent's history remains accessible. Start by splitting, you can merge later if the need arises.

Does a broken-down mission consume more tokens than a broad mission?

Not necessarily. A vague mission generates back-and-forth, errors to correct, reformulations. A clear mission goes straight to the point. Breaking down often makes consumption more predictable, not higher.