Software vendors are now shipping AI with job titles. The title is not the problem. The empty chair above it is.

Key Takeaways

• An AI role is only as good as the person who defines it, checks it, and decides when it is wrong.

• The tasks being automated first are the ones where marketers used to learn judgment. Plan now for where your next senior person comes from.

• Most small teams do not have a hands problem. They have a direction problem, and more agents will not fix that.

On August 31, Optimizely introduced “Virtual Teammates”: AI personas with job titles like Chief of Staff, Marketing Analyst and CRO Manager. They run on schedules, hold their own permissions, keep their own audit trails and sit in the same chats as your people. The trade press asked the obvious question within hours. If the digital teammates prove their worth, why would a marketing leader ever add a human again?

That is the wrong question, and it will cost small businesses money if they take it seriously. The right question is older: who is responsible for the result?

The Standing-Role Trap

A standing role is a job that runs whether or not anyone asked for it this week. A weekly performance briefing. A monthly SEO audit. Continuous monitoring of conversion rates. These are exactly the tasks vendors are handing to agents, and for good reason. They are repetitive, they are well defined, and nobody enjoys them.

But a standing role carries a hidden cost. It produces output on a schedule, output on a schedule gets read on a schedule, and before long the business is organized around consuming reports rather than making decisions. You have seen this with human roles. A report nobody acts on is not analysis. It is furniture.

Optimizely’s own survey of more than 2,000 B2B marketing leaders found that 81 percent switch between two or more disconnected AI tools every week. Read that as a description of the average marketing team: plenty of tools, not enough direction. Adding a teammate that produces more on its own does not solve a direction problem. It amplifies one.

The Judgment Layer

To its credit, Optimizely says out loud what the product cannot do. Its product lead told AdExchanger that the teammates cannot decide what your messaging should be or what makes you different from the company down the street. Those remain human decisions.

Notice what that means. The vendor is selling every layer of the marketing function except the one that determines whether the rest of it matters. If the positioning is wrong, the SEO analyst optimizes the wrong pages, the personalization strategist personalizes the wrong offer, and the CRO manager improves conversion on a promise you cannot keep. Faster, cheaper, and at scale.

At Stellant we pair senior people with AI systems for exactly this reason. The systems take the standing roles. The people own the judgment layer: who we are for, what we promise, what we will not do. That layer cannot be scheduled. It has to be led.

Where Juniors Used to Learn

There is a slower problem underneath the fast one. The analyst seat, the reporting seat, the “pull the numbers and tell me what you see” seat were never only about the numbers. They were where people learned to see. A junior marketer who spends two years staring at campaign data develops instincts no course teaches, and those instincts are what you are paying for when you hire a senior one.

If the standing roles go to agents, the apprenticeship goes with them. Nobody is planning for this. The vendor is not responsible for your bench, and the marketing leader who is asked “why would you add a human” has no line item for “so that someone can run this in five years.”

A small business feels this more sharply than a large one because it has fewer seats to begin with. If you have one marketing person and an agent takes the reporting, the question is not whether that person is still needed. It is whether you have given them the one job only they can do.

The Role Audit

Run this before you buy any AI teammate, and run it again for every human role on the team.

Take a sheet of paper and write each role down the left side, human and automated alike. Across the top, three columns: Defines, Decides, Checks. For each role, write the name of the person who defines what good looks like, the person who decides what happens with the output, and the person who checks whether it was right.

Three things will happen. Some rows will have the same name three times, which is fine in a small company as long as it is a person and not the tool. Some rows will have a blank in the Decides column, which means you are producing work nobody uses. And some rows will have a blank in the Checks column, which is where AI output quietly becomes company policy.

Every blank is a leadership job you have not assigned. Fill them before you add another row.

From Insight to Action

1. Name an owner for every automated task, and make the owner a person with a title, not the tool.

2. Cancel one recurring report this month. If nobody asks where it went, you have your answer.

3. Write your positioning in one paragraph and hand it to whoever configures your tools. If they cannot work from it, neither can the tools.

4. Protect one apprenticeship task. Keep one piece of analysis human on purpose, and say why.

5. Set a review rhythm for agent output that is shorter than the schedule the agent runs on.

6. Ask your vendor what their product cannot do. The good ones will tell you. That answer is your job description.

A company that staffs its judgment defaults to direction. A company that automates it defaults to drift.