Half of what gets called "build an AI agency" is a course selling the idea of a business, not the business. Here is how to tell which one you are looking at.
A large share of what shows up under "build an AI agency" or "make money with AI agents" is not a business model. It is a lead magnet — content whose actual product is the audience it gathers, monetised later through a course, a community, or a coaching call, and the "agency" framing exists to make the audience feel like they are one purchase away from replicating a business rather than one purchase away from a slide deck. I think the two things are genuinely different, and conflating them is why so many people who try this come away with neither a business nor a refund.
The tell is what the pitch asks you to believe is hard. A real service business built around AI tooling has an unglamorous hard part: finding clients who have a specific, recurring, painful problem, and being reliably good enough at solving it that they keep paying. The lead-magnet version skips straight past that and treats "set up the AI workflow" as the hard part, because that is the part a course can package and sell. It is not the hard part. Wiring an agent to summarise inbound leads or draft first-pass replies is a genuinely small amount of work compared to the actual business of acquiring and keeping a client, and any pitch that spends ninety percent of its time on the tooling and ten percent on the client relationship has the ratio backwards.
The second tell is who the numbers describe. A real agency's numbers are about client outcomes — retention, turnaround time, the thing the client is actually paying to have solved. A lead magnet's numbers are about the seller's own growth — audience size, "students," testimonials from people who bought the course, not from people who bought the service the course claims to teach. When the proof offered is entirely about the teacher's success and never about a client's, that is the business model showing through the framing.
None of this means the underlying pattern is fake. Small, real service businesses built around AI-assisted delivery exist and work, in the same way small service businesses built around any efficient tool have always worked — a smaller team doing what used to need a bigger one, at a price that still beats the client doing it themselves. What makes them work is the same thing that has always made a service business work: a specific client problem, a repeatable way of solving it, and a relationship worth renewing. The AI part changes the cost of delivery. It does not remove the need for the rest of it, and no course can sell you around that requirement because it is not a workflow problem.
What I would actually look for, if evaluating one of these from the outside, is whether the pitch can describe a specific client and a specific problem without mentioning the AI tooling at all. If the answer collapses the moment you remove the tool — if there is no business there without the workflow, only the workflow itself being resold as a business — that is the lead magnet, however it is titled. If the tooling is clearly one input into a service that would still make sense without it, just slower or more expensive to deliver, that is closer to a real agency.
I would rather see fewer of these built loudly in public and more built quietly around an actual client, because the quiet version is the one that survives past the first cohort of the course. The loud version needs a constant stream of new believers to stay funded. The quiet version needs one client who is satisfied enough to renew, which is a much lower bar and a much more durable one.