The AI Shortcut Delusion

The plan was simple. Two developers. An AI coding tool. A revenue cycle management system built from scratch in weeks.

The executive sponsor leaned back and said: "This is exactly what we need. We own our system. No more vendor dependency. No more delays."

I had been involved with this hospital's digital transformation long enough to know what that room did not want to hear.

Some context.

The institution was under new leadership and impatient for results. The technology had drifted over the years. Multiple systems, partially integrated, none fully owned. The frustration was real.

And into that frustration stepped a belief that many organizations are developing right now. That AI has changed the equation. That what previously required enterprise platforms, implementation cycles, and serious investment can now be done faster and cheaper by a small team with the right tools.

The two IT advisors driving the proposal were not unintelligent. They understood the technology. What they did not have was a working model of what enterprise software in a clinical environment actually requires.

Revenue cycle management in a hospital is not a database with business logic on top. It sits at the intersection of clinical documentation, regulatory compliance, payer contracts, claims processing, and financial reporting. Each layer has its own rules, exceptions, and audit requirements. It touches patient safety directly. When billing records and clinical records are misaligned, care decisions get made on incomplete information.

Vibe-coding an RCM is not a productivity hack. It is a category error.

But the hype had done its work. The sponsor had watched demos, read articles, and concluded that the old rules no longer applied. The same AI that generates a website in an afternoon could surely build a claims management system in weeks.

When the proposal began expanding, with demands that other clinical systems be reshaped to accommodate the new build, the situation clarified completely. This was not a tool decision. It was a reorganization of the institution's digital architecture around an untested belief.

My response was not to argue against AI. That would have been the wrong fight.

Instead I did what the situation required. I mapped the institution's core systems against its actual workflows. Patient admission. Clinical documentation. Billing. Claims. Discharge. I anchored each system to the specific outcomes it was accountable for: patient safety, revenue integrity, regulatory compliance.

What emerged from that mapping was not an argument. It was a picture. One that made clear which system owned which problem, why the boundaries existed, and what the operational consequences of blurring them would be. Not in theory. In terms the sponsor could not dismiss.

The vibe-coded RCM plan slowed. The conversation shifted from what AI could build to what the institution actually needed to function.

That is a harder conversation. It takes longer. It is less exciting than a demo. But it is the conversation that protects patients and revenue. Those are the only two things that matter.

The pattern I watched in that meeting room is not unique to hospitals. It is playing out in boardrooms across sectors right now.

AI has given decision makers a new frame. The bottleneck was always the cost and speed of software development, and that bottleneck is now gone. Therefore the enterprise investments they have been making, in platforms, implementation partners, and governance structures, are suddenly optional.

This is wrong for a specific reason. The bottleneck was never the code. It was always the operating model: the decisions about what the system needs to do, who owns which process, how exceptions are handled, and how outputs connect to real accountability. No AI tool generates that. It has to be designed, governed, and maintained by people who understand the institution.

None of this is an argument that the tools are weak. They are not, and they have only grown more capable since that meeting. That is precisely why the discipline matters more, not less. The easier it becomes to generate a system, the more the real constraint moves to the part no tool supplies: deciding what the institution actually needs, and what it can absorb.

The executives who create real value from AI are not the ones who move fastest. They are the ones who can tell the difference between what AI can generate and what an institution can absorb.

This is the first piece in a series on that gap. I also wrote a companion guide on the same discipline at the level of the individual professional, Context by Design.