What Tasks Should I Give an AI Agent? (The Do-It-Twice Rule)

If you're asking what tasks should I give an AI agent, start with one rule: if you or your team does something twice, an AI agent can do it. Repetitive, process-driven work like reporting, meeting recaps, CRM updates and relationship follow-ups is where an AI agent pays off first, because you already know the steps and you just have to write them down.

I saw this play out on a discovery call this week with a leader who runs a relationship-driven sales team. They had just started a new workflow. At the end of every day, ChatGPT went into their work inbox and their meeting notes and built a summary of everything that happened. They would send that recap to the team, and the team would update the CRM and create follow-up tasks. They were speaking my language. Then I told them, "You shouldn't have to send it to your team."

What Tasks Should I Give an AI Agent First?

Start with the work you never look forward to.

I don't wake up on the first of the month thinking, I can't wait to jump into my Google Analytics. Reporting is the lowest-hanging fruit there is. So I created an analytics bot as a routine that runs the whole review in about six minutes. It has complete visibility into everything.

Good first tasks share three traits:

• You've done it more than twice. You already know what good looks like.
• It's data-based. Analytics, social reporting, CRM activity and email campaigns are all perfect starting points.
• It drains you. If you dread it, an agent should own it.

If you've never set one up, start with my guide on getting started with personal AI agents. The short version is that personal agents finally do what the industry promised for years. For a long time, "AI agents" meant fancy tech automation that didn't work. Over the last month, the personal agents arriving from the biggest AI companies changed everything.

How Do I Know If a Task Is Ready for an AI Agent?

Ask one question: could I write the steps down right now?

If you're already doing it, all you have to do is write down the process and say, "Here's what we do. Can you help me do this?" That's why I lean so hard on documentation. An agent can only run what it can read, and I go deeper on that in this piece on writing documentation for AI agents.

The leader on my call described their own style perfectly. If something is too hard or takes too many steps, they won't do it. They'll go do something else. That instinct is a gift when you're designing agent work. Every step you hate is a step you can hand off.

Which Work Should Stay Human When You Use AI Agents?

I built my whole approach on a principle I call 92/8. About 92% of what you do can be run by AI. The other 8% is you in your zone of genius: the relationships, the soft skills, the energy.

Energy is transferable. It's one of my superpowers. I open every single call with "Tell me something good that's happened," because most people have never been asked. The leader I spoke with this week has sent thousands of personal video messages over the years, just to let people know they're on their mind. That part stays human.

The point of giving an agent the 92% is to give you more time for the 8%.

How Do I Turn a Workflow I Already Do Into an AI Agent Task?

Take the workflow you already have and finish the loop.

Back to that end-of-day recap. The agent was already doing the hard part. The human was still the delivery truck. So the next version looks like this: create the recap, send it to the team, pull out every task, and put those tasks into the CRM or task manager. One instruction covers the whole loop.

The principle underneath it is simple: I'm the roadblock for all my growth. Once you see that in your own business, you see it everywhere. I wrote more about that pattern in how to stop being the bottleneck in your business with AI.

Then you duplicate. One documented workflow becomes another, then another, until you have a library of systems running the work you used to do by hand. If you want those systems to be reliable before they're autonomous, start each one as a routine. I explain why in AI agent vs automation.

What Tasks Should I Give an AI Agent for Client Follow-Ups?

This is where it gets fun for anyone in a relationship business.

I call it One Connected System. Your AI agent can see your email, your CRM and your meeting transcripts. From all of that, it pulls out the key things you know about a person. Then you give it a rule: scan for birthdays, family news and anything life-changing.

The leader on my call described their dream version. The CRM would notice that someone in the database just had a baby, then create a task for someone on the team to send flowers congratulating them. That's a perfect agent task. It's repetitive, it's rule-based, and it makes the human relationship stronger.

Every time you spot a new signal worth tracking, it becomes a new rule. Your team populates it, and the AI scales it.

What If My Tools Won't Connect to an AI Agent?

This is the friction that's going to decide who wins in the AI era.

The leader I spoke with is stuck inside a proprietary CRM that won't connect to their ChatGPT account. They can only use the AI bot built into that CRM, and it can't do much. Eventually that friction gets so big that you look at a competitor who can integrate everything and realize the old way doesn't make sense anymore.

Until then, there's a bandaid. Some AI tools can take over your screen and do the clicking for you. It's a temporary fix, and it shows exactly where things are going.

Your Next Steps

1. List everything you did twice this week. Every item on that list is a candidate.
2. Pick the one you dread most. Reporting is a great default.
3. Write the steps down. If you can explain it to a new hire, you can explain it to an agent.
4. Finish the loop. Make the agent carry the result all the way to the finish line.
5. Add one relationship rule. Birthdays, family news or life changes. Let the agent surface them and tee up the human touch.
6. Protect your 8%. Spend the time you get back on the people.

Once you start asking what tasks you should give an AI agent, you'll find them everywhere. Pick one tonight and let it run tomorrow.