How to Set Goals for AI Agents (The Goalkeeper Bot That Gives Every Agent a Report Card)

If you run AI agents and you want to know how to set goals for AI agents so they stay on track, start at the end. Write the outcome each agent owns, document it where the agent can read it, and have that agent report against the outcome every week. Those three moves turn a helpful bot into a teammate you can hold accountable.

I learned this recently. I saw someone on X who had created a bot called Goalkeeper. Its whole job is making sure every other bot has a goal and stays accountable to it. I went through my entire team one bot at a time, restated the intention of each one, and had it write the clear goal that bot would report on every week.

I did it in a small pocket of time over the weekend.

Why Do AI Agents Drift Without Clear Goals?

An agent can only aim at what you gave it. Tell it to go do your reporting and it will produce a report. What is that report supposed to change?

I teach outcome based prompting. Start at the end and let the AI work backwards. The same lesson applies once you have a team of agents. How explicit are you on the actual goals and deliverables of each one?

Think about how you would treat a human teammate. You would never hand someone a job with no target and expect them to know what winning looks like. Fuzzy goals are easy to create inside of AI, and very few people are clear on this. When the goal is missing, the agent drifts, and you only notice after the work has gone sideways.

How to Set Goals for AI Agents: What Should Each Goal Include?

A good goal answers one question: what is the goal of running this?

I worked through this live with a founder I coach who wanted an analytics bot for their ads. It would be easy to tell that bot to go run my analytics.

So I asked what the goal of running it was. Their answer gave the bot a real job:

• Kill the ads that are not producing the type of leads we want.
• Maximize the ones that are.
• Add money to the ones clearly doing better.

Now the agent has a job with a finish line. Each goal you write should carry the same three pieces:

1. The outcome. What changes in the business when this agent does its job well?
2. The measure. What will the agent report on so you can see progress?
3. The home. Where is the goal documented so the agent reads it every run? Mine live in Notion.

Once a goal has those three pieces, picking the work gets easier too. If you are still deciding what belongs on an agent's plate, start with the tasks you already do on repeat.

What Is a Goalkeeper Bot?

A goalkeeper bot is an agent whose goal is making sure all the other agents have goals. It works the same way a watchdog watches over your other bots, except it watches the target instead of the task.

It goes through your team bot by bot and asks what each one is for. If an agent is missing a documented goal, the goalkeeper catches it, because a bot with no documented goal is free to drift.

Mine went through my team one by one. I restated the intention of each bot, it created the clear goal, and that goal is what gets reported back every week.

How to Set Goals for AI Agents Across a Whole Team: Where Do You Start?

Go one agent at a time. State the intention, let the AI write the goal, and read it back.

You can start with whichever bot matters most to the business, then work down the list. Sometimes you just find five to seven minutes. Those windows show up while you are watching football.

If you only have one or two agents, that is an even better time to do this. Every new bot you add should get its goal on day one, the same way it gets its standards when you set it up.

How Often Should You Review AI Agent Goals?

Weekly is what I am running. When I finished the walkthrough, my goalkeeper told me it would give me a report card on every bot, every week. Report card was its term. I liked it right away.

You inspect what you expect. A weekly report card puts that principle on a calendar, and it gives you one place to review agent output without digging through every run.

I will be candid about where this stands. I made this change days before that coaching call, so the first report cards are still coming in. I am calling it now anyway. I believe this will be the biggest upgrade I have made since I started running agents, because it forces clarity on the one thing every other result depends on.

Pair the report card with a look at whether the agent is saving you time. One tells you if the agent is hitting its target. The other tells you if the target was worth the effort.

What Should You Do First to Set Goals for AI Agents?

Pick a few small pockets of time this week and run this:

1. List every agent you have running, even the small ones.
2. For each one, state why you built it.
3. Ask your AI to turn that intention into one clear goal with a measure attached.
4. Document the goal where the agent will read it.
5. Have each agent report against its goal once a week.
6. When you are ready, hand the whole job to a goalkeeper bot of your own.

As you go, protect your own focus. I have a rule at the gym: no AI on my phone beyond logging my sets and playing music. This work can pull you in for hours, because every system you build makes you want to build the next one. Give your agents clear goals so you can step away from the desk and get it in.

Your agents will work toward whatever you make clear. Decide what winning looks like for each one, write it down, and check the score every week.