Counting agents

IQVIA's latest earnings call had a number that's going to get repeated a lot: 415 AI agents deployed, up from 192 in Q1. They also credited those agents with lifting win rates by two to three points. That's real traction, and it's tied to bookings, not just a demo count.

But agent count is a measure of tools, and the rest of the industry is about to start reporting it as a measure of progress. I'd rather know a different number: how many workflows have been redesigned so they can't run without AI?

The more interesting line on the call was a smaller one. Ari Bousbib described layering agents to bring clinical delivery, lab services, and electronic patient assessments "under 1 model" so issues surface earlier. That's not an agent added to a step. That's an attempt to remove the handoffs between steps, and that's where the big gains in trials are hiding.

He also called what the frontier AI labs provide "generic tools," arguing the value comes from life sciences data and domain context. I think that's right, and it's the most useful build-vs-buy lesson in the whole call: the model is the part you can buy. The context, and the redesigned workflow around it, is the part you have to own.

So if you're tracking AI progress at your own organization, track two numbers: how many AI tools you've deployed, and how many workflows you've actually redesigned. The gap between them tells you whether you're getting faster or becoming AI-native.