COO, AG Mednet
Insilico just signed another AI-discovery deal. IGC Pharma cut Alzheimer’s data prep from 28 hours to two and a half. Snowflake and NVIDIA are putting agentic AI directly into life-sciences data platforms. If you only read the headlines, you’d think the story is that AI is getting faster at finding drugs.
The more interesting story is where the constraint went. For a decade, the rate-limiting step in drug development was discovery — finding a viable molecule. AI has genuinely compressed that. Generative models now surface candidates in a fraction of the time, and the deals reflect it, moving from science-project pilots into structured, milestone-driven pharma partnerships.
The constraint moved into trial operations and data management
But molecules don’t generate value. Approved drugs do. And between a promising candidate and an approval sits the part of the industry that AI has been slowest to fix:
- Running the trial
- Reconciling data
- Coordinating sites
- Keeping the whole apparatus auditable enough to survive regulatory scrutiny
So as discovery speeds up, the bottleneck doesn’t disappear. It moves downstream — into trial operations and data management. More candidates, reaching the clinic faster, all funneling into an operational layer that hasn’t scaled at the same rate. That’s not a discovery problem anymore. It’s an orchestration problem.
In a regulated trial, speed isn’t the number that matters
Here’s the part that gets underappreciated. IGC’s 90% time reduction is real and impressive — but the number that actually matters in a regulated setting isn’t speed, it’s whether the output is reproducible and auditable. A harmonization that’s fast and unverifiable is worthless to a sponsor who has to defend the dataset to the FDA. Snowflake understood this; their entire pitch is governed agents acting on data that never leaves a controlled environment.
That tells you where the value really sits. Not in the model. In the governance around the model. This is the layer most organizations underinvest in, because it’s less glamorous than a new algorithm and harder to demo. It’s also the layer that determines whether agentic AI is deployable in a GxP environment at all. You can have the best model in the world, but if you can’t show your work, you can’t use it in a trial.
The value is in the governance, not the model
This is what we’ve been building toward at AG Mednet — treating AI not as a feature to bolt on, but as something that has to operate inside a workflow that is accountable by design. Augmenting how trials run, not replacing the people and controls that make the data trustworthy.
The companies that win the next phase won’t be the ones with the fastest model. They’ll be the ones who made AI’s speed auditable.
Where do you think the next bottleneck moves after operations?
John Paul (JP) Lee, COO, AG Mednet. McKinsey alum and Kellogg MBA, JP drives operational strategy for Judi across regulated clinical trial environments.
