What twenty years in healthcare ops taught me about AI agents.
Twenty years in corporate healthcare operations taught me one thing about technology that most AI consultants will not tell you. The tool is never the issue. The process underneath is the issue. The tool just makes the process visible whether you want it visible or not.
That lesson did not come from watching other companies. It came from being in the work, leading large projects inside a healthcare company. When those projects were poorly executed, the cost was millions in revenue and membership losses in the six figures. SOPs and governing documents cannot be updated while a project depends on them. The documents have to be finished and stable before the project starts. The project depends on the documentation. Not the other way around.
Lesson one. If the process is not documented, the automation will not save you, it will embarrass you. In healthcare we called this surfacing the workarounds. The system would do what it was told to do, and what it had been told to do was a fantasy version of the actual workflow. In a small business, the AI agent will follow whatever it is given. If what you give it is a hope and a vibe, you will get hopeful, vibey output that has nothing to do with how the business actually runs.
Lesson two. If nobody owns the process, nobody will own the AI. Every system I watched fail in healthcare failed because the owner was a committee, and a committee is the same thing as nobody. AI agents need an owner. One person who can answer when it produces something wrong, who can update the prompts, who can decide what the agent should and should not touch. Without that person, the agent drifts. Drift in healthcare killed projects. Drift in a small business kills client trust.
Lesson three. If you cannot measure the process before the tool, you cannot measure the tool. Healthcare operations live and die by baseline metrics. How long did this take before. How long does it take now. What was the error rate before. What is the error rate now. Small businesses skip this part and then wonder if the AI agent is working. You cannot know if it is working if you never knew what working looked like in the first place.
Lesson four. The tool exposes the gap. It does not close the gap. This is the lesson I repeat most often because it is the one owners want least to hear. AI agents are not going to fix what the business has not fixed for itself. They will only make the gap obvious, in public, in front of clients, in writing. That is what a poorly deployed agent does. It tells on the business.
The good news is the inverse is also true. A business with a documented process, a single owner, and a baseline number can plug in an AI agent and see real lift inside thirty days. Same tool that embarrassed the unprepared business now performs for the prepared one. The difference is never the technology. The difference is the work that was done before the technology arrived.
Two decades of watching this play out in healthcare made one thing clear. Deploy the tool last. Document, own, measure, fix, then automate. That order has not changed in twenty years, and AI agents are not going to be the exception.