So, Tempus, a well-known techbio company, is the latest to bet that AI agents could replace (at least some) data scientists and analysts in drug development...

The new version of its platform Lens, out today, works like this: a scientist describes a biological hypothesis in plain language, and the platform's agents draft an analysis plan, write the code, and run it against Tempus' real-world dataset. They cite more than 8.5 million de-identified patient records, by the way.

Results come back as interactive reports, with a toggle to view the underlying code. There are specialized agents for specific jobs like biomarker validation and trial design.

What used to be the workflow (i.e., a scientist frames the question, hands it to a data science team, waits weeks for the analysis, etc) now can theoretically be compressed into a conversation with an agent.

Agenticizing the workflow is becoming the default move across the data and tools layer right now. The pattern is the same wherever you look: take a process that requires a specialist intermediary, put a natural-language agent in front of it, and let the domain expert drive directly.

Clinical data, lab automation, regulatory work, literature review... same shape.

Tempus is doing it with one of the larger proprietary oncology datasets behind the agent, which is what makes their version worth watching. But the architecture is fast becoming a kind of new normal.

A lot of techbio companies, if not all, are implementing agentic workflows in one way or another, or piloting them internally. But I have a question for you guys, what about cost?

More and more I see the emerging challenges with the economics of frontier model token usage... how does it relate to the rise of agents in drug discovery/biotech?

I would be glad to hear your thoughts about it, but IMO, we are going to see "the new shade of exciting" about the agentic AI trend, quite soon...

Image credit: Tempus