The People Buying AI Agents Are Mostly Other Builders

Agents are now commissioned almost as often as mobile apps. We wrote about that, and ended on the customer arriving: a business with a manual process, asking for the agent that runs it.

The obvious next question is which processes those buyers want run.

We went looking. The answer depends entirely on who is asking, and the who turns out to be the story.

The appetite ladder

Measure every type of buyer by its own behaviour: out of everything that buyer asks for, how often is it an agent?

For startups and software companies, one ask in every twenty. For enterprises, one in forty. For small businesses, one in a hundred and sixty.

That is a ladder with a factor of eight between top and bottom, and the top rung is not who the agent hype is aimed at.

The most agent-hungry buyers in the market are other builders.

Same word, two different products

When a builder commissions an agent, four times out of five it goes into engineering or product. Into the thing they sell.

When a small business commissions one, the proportions flip almost exactly. Four times out of five, it goes into running the business.

So "AI agent" currently means two different products. For builders, a component, bought the way payments and authentication were bought last era: capability that disappears into someone else's product. For everyone else, a worker-shaped thing that does a process.

The component market is the one that is large today. That is the picks-and-shovels phase, and it is always underestimated, because end customers cannot see it happening.

Everyone enters through the same door

When buyers name the process, one answer leads for builders and small businesses alike: customer support. Behind it, lead handling and data analysis. Enterprises are the exception, leading with data analysis, support just behind.

Notice what that means. The process small businesses most want an agent for is support. The agent builders most often put into their products is also support.

The queue is forming on both sides of the same door.

The price already tells you the rest

An enterprise sizes an agent at roughly five times what a builder does, and around eighteen times what a small business does. And enterprises now commission agents about three times as often as they commission mobile apps.

For that buyer, the agent has already replaced the app as the default thing to want, budgeted like real software, not like an experiment. The small business wants the same shape sized like a subscription. Not a project. A product.

Three waves, three speeds. What to do depends on which one you surf:

• Selling to builders? The market is open now, and it is the big one. They need capability, evaluation, safety and plumbing more than another finished agent.
• Selling to enterprises? Lead with data and analysis, price with confidence, and know you are replacing the app on their roadmap, not supplementing it.
• Selling to small businesses? Build the support agent as a product with a subscription price. The demand is real, and it will not fund custom work.

One mistake covers all three: building for the wave you keep reading about instead of the one that is actually here.

This reading comes from millions of demand signals, collected continuously. Not headlines and not surveys, but real buying activity, the same intelligence the product runs on. We keep saying the market moves at the speed of AI. It also moves in a particular order, and knowing where the wave actually is beats knowing that it is coming.