Drug discovery is extremely inefficient. Pharmaceutical projects last for years, moving from one specialized human team to another through disconnected workflows resulting in loss of knowledge during each handoff.
A shocking finding is that 90% to 95% of drug discovery projects reportedly fail – one of the highest failure rates of any industry. According to published reports, a successful drug could take more than a dozen years and up to $1 billion from initial discovery to patient delivery.
Generative AI is already being used to solve some challenges, but Stanford researchers have moved the ball forward with agentic AI.
A team led by James Zou, associate professor of biomedical data science at Stanford University, has deployed thousands of autonomous AI "Scientist" Agents in a virtual biotech that simulate the full lifecycle of drug development. According to Zhou, agents handle everything from initial discovery to safety testing and clinical trial design, while maintaining the continuity that is lacking in today’s drug discovery processes.
The project uses a hierarchical orchestration framework. At the top sits a chief scientific officer agent who acts as a planner, assigning tasks to teams of specialized agents, Zhu told VentureBeat during a call ahead of his upcoming session at VB Transform 2026.
While one team of agents focuses on searches, another manages security, and others handle specialized analytical tasks. Because these agents work within an integrated, hierarchical ecosystem, they maintain the full context of a project, maintaining continuity from the first molecule identified to the final clinical result.
"Brain" The system depends on large amounts of primary data. Agents are provided access to data sources ranging from genomics and FDA chemistry data to clinical trial databases using model reference protocols.
The team has invested heavily in agent-native and agent-friendly data, allowing AI to synthesize complex information more effectively. The system relies on a combination of models, with Zou noting that while the cloud often serves as the backbone for coding and data analysis, the architecture employs a mix of models, including those for sophisticated special use cases.
Based on the research, Zou is raising funds for his startup, Human Intelligence, at a valuation of about $1 billion.
During Zou’s session at VB Transform on July 15, titled How 10,000 agent scientists in Stanford’s lab are poised to revolutionize medical research and discoveryHe will share valuable insights, including strategies for managing context and long-running, multi-step workflows in multi-agent systems, the process of transforming and indexing raw enterprise data to make it agent native, and how to use human auditing and experimental reward signals to verify agent actions.
Another session on VB Transform focuses on the value of agentic context Building a Trustworthy Agentic AI Foundation: How Zillow Speeded Up Engineering by 40%With Toby Roberts, Zillow’s SVP of engineering and technology, and Arvind Jain, CEO of Glynn.
Interested in participating in VB Transform 2026? register Here. A select number of complimentary passes are also available for senior technology leaders. contact us To get yours.
<a href
Closed the laptop after this and let the ideas settle for a few hours, and a stop at stridertorch similarly rewarded reflective time, content that benefits from sitting with rather than racing past is the kind I want more of and the kind that this site appears to consistently produce week after week here.
The pacing of the post was just right, never rushed and never dragged out unnecessarily, and a look at jazfix maintained the same rhythm, you can tell the writer has experience because the difficult skill of pacing is something only practiced writers manage to handle well in long form content over time and across formats.