One of the most exciting discoveries during my MBA projects was the possibility of creating an AI agent to represent a buyer persona.

Rather than just leaving the persona on a slide, we could actually talk to it. We could show it a product idea, an ad, or a visual and ask, "What do you think?" or "What feels unclear?" or "What would make you interested?"

Picture giving the agent this description: “Your name is John Smith. You’re 38, live in San Francisco, and work as a professional photographer.”

Suddenly, there is someone to ask questions. It seems like a fast and easy way to try out ideas before investing time and money in testing.

But this also brings up a question: whose perspective are we actually hearing?

The assumptions inside the persona

Even before the agent says anything, we have already made decisions about the person it represents.

We chose John’s age, where he lives, and his job. Maybe we also decided he cares about quality, keeps up with tech trends, and is willing to pay extra for reliable gear.

However, knowing John is a photographer doesn't tell us whether he likes using familiar tools or prefers to experiment. Living in San Francisco does not reveal his budget. His age does not show how he decides what to buy.

When a conversation feels like validation

Let’s say we ask John if he would buy our product. He says yes and explains his reasons.

That answer feels encouraging. It might even sound like the kind of customer feedback we want. But we still do not know if a real photographer would understand the offer, trust it, or actually buy it.

The agent has answered our question, but no real purchase decision has happened.

This is where I see the biggest risk. A simulated conversation can make an assumption seem tested. If we treat that answer as proof, we might feel more confident without learning anything new about real customers.

What the stereotype can hide

A narrow persona can cause us to miss people whose needs do not match the profile.

We might focus on a photographer who wants advanced features and miss someone who just wants things to be simpler. We might think premium quality means a higher price is fine, but overlook someone who cares about reliability and has a tight budget. It can shape the features we prioritize, the language we use, and the customers we decide to pursue.

Making several agents does not always fix this problem. If all their profiles are based on the same ideas, we just hear different versions of the same made-up customer.

How I would use a persona agent

I still think these tools are exciting. I would use them to look for possible objections, test a message, and come up with better questions for customer research.

But I would make sure to show where there is uncertainty:

· Which persona details come from research?

· Which are assumptions we need to check?

· What changes if we alter the person’s budget, experience, or priorities?

· Which responses should we investigate with real people?

Instead of just asking “Would you buy this?”, I would also ask “What would make this not matter to you?” or “What would you need to know before thinking about it?”

These answers can give us ideas to explore. Talking to customers and watching what they do can help us figure out if those ideas are important.