Which Marketing Tasks Belong On An AI Agent?
The marketing tasks that belong on an AI agent are the ones that are a rule rather than a judgment call, and repeatable rather than relational: transcribing calls, sifting transcripts for hooks, drafting magnets from proven structures, launching and monitoring the $100 test, iterating ad variants, and producing the morning verdict. The owner keeps the math, the strategic choice of which hook to pursue, the offer, and the relationships. Here is the full split and the two tests that produce it.
The two tests
I use the same two questions for every task in the marketing stack, and they settle almost every argument.
Is it a rule or a judgment call? If you can write down what good looks like and what to do when a number crosses a line, it is a rule, and rules go to an agent. If deciding requires knowing where the business is going this year, who you want as a client, and what you are willing to be known for, it is judgment, and judgment stays with the owner. The trap is getting this wrong in either direction: handing judgment to a machine that will confidently optimize the wrong thing, or hoarding rules so you stay the bottleneck on work that does not need you.
Is it repeatable or relational? If the task happens every week in roughly the same way, it is repeatable and belongs on a schedule the agent runs. If the task is a human deciding to trust you with money, it is relational, and no amount of agent capability changes who should be in that room.
Run those two tests across the seven lessons of the crash course and the split falls out cleanly.
The split, task by task
| Task | Agent or owner | Why |
|---|---|---|
| Record and transcribe every call | Agent | Pure rule. Happens every day. Zero judgment. |
| Sift transcripts for the five lean-in moments | Agent | Repeatable with a clear definition of done. The owner reviews the list, not the dirt. |
| Choose which nugget becomes this week's magnet | Owner | Judgment. Which hook fits where the business is going is a strategic call. |
| Research what is already winning in the category | Agent | Ads running three to six months are winners, and that is public. Collecting it is a rule. |
| Draft the magnet from a proven structure | Agent | Model winners, don't invent. Swapping your nugget into a working structure is repeatable. |
| Set target CPA and kill thresholds | Owner | The math. The agent optimizes toward whatever number you give it, so the number must be yours. |
| Build the page and launch the $100 test | Agent | Same-day execution through the Meta developer platform. Rules all the way down. |
| Monitor spend, pause losers, iterate variants | Agent | Thresholds are rules. If cost per lead crosses the line, pause. If a hook wins, test the next variant. |
| Produce the morning verdict | Agent | Repeatable summary against the owner's numbers. |
| Decide kill, rerun, or scale | Owner | Five minutes, but it is judgment. Cheap leads that will never buy look great on a dashboard. |
| Write and schedule the trust-building hub content | Agent drafts, owner approves | The hub builds trust while you sleep. Drafting is repeatable; what you are willing to say in your name is not. |
| Define the offer and the invitation | Owner | Relational and strategic. The agent can draft and schedule it; you decide what you are selling and to whom. |
| The sales conversation | Owner (or your closer) | Relational. Someone deciding to pay you wants a human. |
Why the owner side is smaller than it feels
Look at the owner column. Four or five lines, and most of them take minutes, not days. Set the math once a quarter. Pick the nugget once a week. Make the kill-or-scale call every morning over coffee. Define the offer when it changes. Have the sales conversations you were always going to have. That is the whole job. Everything that used to eat your marketing team's week (the brainstorming, the drafting, the launch logistics, the ad babysitting) is in the agent column, because every one of those tasks is a rule you can write down and a thing that repeats.
This is what lesson seven means by "you architect." The architect does not lay bricks. The architect decides what gets built and checks the work. The practical workflow for the agent side is in how to automate the marketing that runs overnight.
The tasks owners hand over too early
Two of them, and they are the expensive ones.
The math. Owners will say "just optimize for leads" because setting a real target CPA requires admitting what a customer is actually worth and what the close rate actually is. The agent takes the instruction literally and buys the cheapest leads available. Three weeks later the dashboard looks wonderful and the calendar is empty. The course is blunt about this: "It doesn't matter how good you get at AI if you don't understand the math of marketing." The math is yours. Do not delegate it.
The nugget choice. It is tempting to let the agent pick whichever hook tested cheapest and run with it. Cheap is not the same as right. A hook that pulls in bargain hunters when you are moving upmarket is a win on the spreadsheet and a loss for the business. The agent surfaces the list and the numbers; you pick.
The tasks owners hold onto too long
Also two.
Drafting. "Nobody can write it like I do" is the sentence that keeps marketing at one post a month. The words should come out of your market's mouth anyway; that is the whole point of sifting transcripts. The agent packages what your buyers already said. You approve what goes out in your name. That is a five-minute review, not a writing session.
Ad management. Clicking through the Ads Manager every morning is a rule-following task that an agent now does through the Meta developer platform and the Ads MCP. Thresholds, pauses, variants, reports. If you are still doing this by hand, you are paying an owner's hourly rate for work that has a written definition of done.
How to run the split on your own stack
- List every marketing task your team touched in the last month. Be specific.
- For each one, ask: rule or judgment? Write down what "good" looks like. If you can, it is a rule.
- Ask: repeatable or relational? If it happens weekly in the same shape, it is repeatable.
- Rule plus repeatable goes to the agent with a schedule and a definition of done. Judgment or relational stays with you.
- Anything in between (drafting in your voice, hub content) is agent drafts, owner approves.
Most owners who do this exercise find that the agent column is much longer than they expected and the owner column is shorter and more valuable. That is the point. If you want to see where owners get the split wrong in practice, read the seven overnight marketing mistakes.
FAQ
Should an AI agent write my marketing copy?
It should draft it, from your own transcripts and from structures already winning in your category, and you should approve what goes out in your name. The words come out of your market's mouth; the agent packages them. Approval is a five-minute review, not a writing session.
Can an agent decide which lead magnet to run?
It can surface the list of hooks and the numbers from each test, but choosing which hook to pursue is a strategic call about where the business is going. Cheap leads are not automatically the right leads. The agent recommends; the owner picks.
What is the fastest way to tell if a task belongs on an agent?
Ask whether you can write down what good looks like and what to do when a number crosses a line. If yes, it is a rule and it belongs on an agent. If deciding requires knowing your strategy or involves a human choosing to trust you with money, it stays with you.