AI is the car. The driver is everything.

Every IVD sponsor we talk to in 2026 has questions about AI. Some are eager to know how we use it. Some are concerned about how we might use it. Some have heard from another CRO that AI is going to compress validation timelines or replace medical writers or auto-generate CRFs — and they want to know whether we can match those claims, or whether we are about to be left behind by the CROs that are leaning into the technology hardest.

I have been around AI long enough to know how to answer the question honestly, which is that the question is the wrong one. The right question is not whether the CRO uses AI. The right question is who is driving.

A Ferrari in the hands of someone who has never driven a manual transmission is not a sports car. It is an expensive way to grind a clutch. A Ferrari in the hands of someone who is texting on their phone in the left lane of the freeway is not a sports car either — it is an upgrade kit for the same accident they would have had in a Honda. The car does not produce the outcome. The driver does, and the car is a multiplier on whatever the driver was already going to do.

AI in clinical operations works the same way. Used by a project manager who already knows what site engagement requires, AI compresses the time it takes to draft the weekly site update from forty minutes to ten — and the PM uses the thirty minutes saved to call the site about the freezer that has been on a slow leak since November. Used by a project manager who does not know that the freezer matters, AI generates a weekly update faster, and the freezer goes uncontacted, and the study slips. The tool changed the time. It did not change the outcome.

The same dynamic governs every piece of the operation. A medical writer with deep IVD experience using AI produces sharper, faster protocol drafts. A medical writer without that experience using AI produces the same shallow protocol drafts they would have produced manually, just at higher volume. A data manager who already knows what queries matter uses AI to surface them earlier; a data manager who does not, uses AI to generate query suggestions they cannot evaluate. The tool amplifies what the operator brings. It does not substitute for what the operator is missing.

This is the part of the AI conversation that the industry's marketing has gotten wrong. The pitch is that AI lets a CRO do more with less — fewer experienced operators, faster timelines, lower cost. The empirical reality is the opposite: AI raises the value of experienced operators because it raises the leverage of every decision they make. The PM whose hour is worth ten of someone else's now produces twenty hours of impact in their hour. The medical writer who could carry the protocol context in their head writes ten times faster. The advantage of having the right people compounds, not flattens.

The CRO that hands AI to a junior operator who has not yet learned the work and tells them to produce protocols at scale is not building leverage. They are scaling the absence of judgment. The output looks like work. It is not work. Sponsors who receive that output discover the absence of judgment three months later, when the protocol the AI produced does not survive its first IRB review or when the data manager's auto-suggested queries miss the deviations that actually matter.

What we do at RDI is the opposite of the conventional pitch. We use AI extensively — for drafting, for analysis, for first-pass review of long documents, for surfacing patterns in monitoring data. We hand it to people who are sharp, driven, and care about getting the trial right. They use it to do more of the work they were already doing well, faster, and with less of the rote effort that does not require their judgment. The results are everything we hoped AI would deliver, because the people using it brought the judgment AI cannot bring.

What we do not do is hand AI to anyone who would otherwise be on a learning curve and ask them to skip the curve. The judgment that AI amplifies has to come from somewhere, and the somewhere is years of doing the work without the tool. A PM who has never personally found dry ice for a site at six in the morning does not know what to ask AI to optimize. A medical writer who has never had a protocol bounce back from IRB does not know what to make AI catch. The driver makes the car, and the driver is made by the road.

The honest version of the AI conversation, for sponsors evaluating CROs in 2026, is therefore not about which CRO uses AI most. It is about which CRO has the drivers. Ask the question that way, and the answer becomes legible. Look at the people, not the tools. The CRO whose project managers have ten years of running IVD studies and use AI to multiply that experience will produce the study you wanted. The CRO whose project managers have six months of experience and use AI to look like they have ten years will not — and the gap will surface, as it always does, in the field.