Why AI agents are a race to the bottom

There is a version of my business I could be running right now, and I decided not to.

It looks like this. I put "we build AI agents" on the website, I price a bot at a few thousand dollars, and I compete with every other shop that showed up in the last eighteen months saying the exact same thing. It is a real market. There is real money in it this quarter. And it is a race to the bottom, so I am not going to run it.

Here is why the floor keeps dropping out from under that business.

The thing you are selling gets cheaper and better every quarter without you doing anything. The models underneath these agents improve on a schedule you do not control, and their price falls at the same time. Whatever clever thing your agent did last spring is a checkbox in someone's platform by fall. When the core capability is commoditizing that fast, the wrapper around it cannot hold a premium. You are selling a thing whose value is deflating in real time.

The barrier to entry is a weekend. The same tools that let me build a working agent let anyone build one. That is genuinely great for the world, and it is terrible for anyone whose entire pitch is "I can build the agent." If your only edge is access to the tool, you have no edge, because the tool is the one thing everyone has.

And the buyer has already been burned. Most owners I talk to have paid for at least one AI pilot that demoed beautifully and then quietly died, because nobody owned what happened after the demo. So when the next person walks in leading with "AI agents," the owner is not excited. They are tired. Leading with the technology now works against you.

Put those three together and you get a market where everyone is selling the same deflating thing, anyone can enter it, and the buyer is skeptical before you open your mouth. The only lever left is price. That is the race, and the prize for winning it is a smaller and smaller margin on a product that matters less every month.

So I stopped selling the agent.

Because the agent was never the point. The problem in a business is almost never "we do not have enough software." The problem is a quote that sits for three days while the owner is in the field. It is a good lead that never gets a second call. It is a company that is quietly the best in its market and is invisible to the buyers who would pay a premium for it. It is a number that nobody in the building can actually see. None of those are software problems. They are operating problems, and software is one of several tools you might use to fix them.

That distinction sounds academic until you watch it play out.

I worked with a landscaping company where the office was the bottleneck. Every field decision waited on someone to dig through old invoices, the inventory lived in people's heads, and the owner was the only person who really understood how the whole thing ran. We did deploy software so anyone can now pull years of job history in plain language. But we also built a succession plan so the business no longer depends on one person, stood up training so the crews run the work the same way every time, and got pricing and overhead onto a consistent footing so the margins actually hold. The result was fifteen office hours a week back, and an operation that runs without a single person as the bottleneck. The agent was maybe a quarter of that. The judgment was the rest.

I worked with an industrial cleaning company that had thirty years of quality and a real story, stuck in the crowded middle of its market where everyone competes on price. The fix was not a bot. The fix was a decision: stop selling to the wrong tier. We rewrote who the ideal customer is, moved the whole target list up-market toward Class A property management and specialty industrial sites, repositioned the brand around the things the national giants cannot copy, and yes, built an AI engine so the founder can put her own voice in front of those buyers consistently. The AI made the new strategy scalable. It did not choose the strategy. That is the part you cannot download.

This is the difference between hiring a vendor and hiring an operator. A vendor ships you a system and leaves. Whether it moves the number is your problem. An operator sits in the seat, owns the number, and uses whatever it takes to move it, including AI, but starting from a read on the business that no tool can give you.

I am not against AI agents. I build them, constantly, and they are doing real work for the companies I operate inside. But I will not sell you one as if it were the answer, because I have watched too many owners buy the tool and still have the problem.

If your business is stuck, do not start by asking which AI to buy. Start by asking where the money is actually leaking, and then bring in someone who will own closing the gap. Sometimes the answer is an agent. Usually it is bigger than that.

That is the work. The AI is just how I deliver it.