For builders, the working assumption has been one engineer, one agent — the mannequin Claude Code and related instruments. At VB Transform 2026, James Zou, affiliate professor of biomedical knowledge science at Stanford College, argued that assumption is about to interrupt: the subsequent frontier isn't a single, extra succesful agent, it's tens of hundreds of them collaborating.
For builders and product builders, probably the most essential takeaway from Zou’s presentation is how these large techniques are orchestrated. His workforce's analysis affords a sensible blueprint for connecting legacy databases to AI orchestration layers and designing environments that allow hundreds of brokers to collaborate.
Emulating the group — the digital biotech
Zou’s undertaking started as a "Digital Lab" consisting of 5 to eight brokers structured to reflect his bodily Stanford lab. The setup included an AI professor performing because the principal investigator and AI college students with distinct specialties holding common group conferences.
"We additionally created for the brokers a duplicate of Stanford, an agent college, the place the brokers can truly go to the varsity and do supervised fine-tuning to enhance their experience of their particular domains," Zou famous.
The digital lab efficiently designed new nanobody proteins for latest COVID variants.
"What is absolutely thrilling to us is that these AI-designed nanobody proteins truly labored significantly better than the earlier human-designed nanobodies by way of binding to the latest totally different viruses," Zou stated.
Following this wet-lab validation, the workforce expanded their ambition. They transitioned from emulating a single analysis workforce to modeling a large company construction.
The ensuing system, dubbed the Virtual Biotech, includes tens of hundreds of specialised AI brokers overseen by a Chief Scientific Officer (CSO) agent. It operates by distinct company divisions, similar to goal discovery, molecule design, and medical trials.
"Working with the CSO agent are totally different divisions that mirror the divisions present in a human biotech or pharma firm," Zou defined — one centered on figuring out drug targets, one other on designing molecules, a 3rd on security and medical trials. Particular person brokers specialize additional inside a division, he stated. "Beneath the goal discovery division, we'll have one agent that focuses on all of the genetics knowledge, one other agent that appears in any respect the genomics knowledge and single-cell knowledge, and so forth."
The multi-agent benefit
As basis fashions develop extra succesful, builders face a core architectural dilemma: Why distribute workloads throughout tens of hundreds of specialised brokers as a substitute of channeling all computing assets right into a single, omniscient mannequin?
Zou's workforce ran a head-to-head comparability of a multi-agent workforce towards a single agent tasked with the identical scientific problem. The multi-agent ecosystem created friction and interplay that produced higher options that have been extra resilient towards compounding errors.
"In these scientific digital labs, the brokers truly get into debates and disagreements. They should persuade the opposite AI scientists [of] their concepts, and all of that elicits far more inventive and strong reasoning in comparison with when you have a single mannequin making an attempt to do the issue by itself from scratch," Zou stated.
The orchestration bottleneck
When scaling to tens of hundreds of brokers, orchestration turns into the first bottleneck. The system requires a unified context layer that permits brokers to synthesize information from numerous instruments, datasets, and historic information.
Many enterprise groups try to unravel knowledge integration by wrapping present databases with an MCP. Nevertheless, legacy techniques should not very pleasant to brokers. For example, dropping a PDF of a analysis paper into an agent's context window is inefficient, and commonplace textual content fashions battle to interpret advanced figures and tables, resulting in hallucinations.
"Even should you wrap an MCP across the present databases and APIs, that doesn't remedy the underlying downside: the interface and APIs should not appropriate for brokers," Zou stated. He added that present databases are designed to be consumed by people or pre-AI algorithms.
To resolve this, Zou's workforce created Paperclip. The platform depends on a core power of recent LLMs: their skill to put in writing code and navigate file techniques. As an alternative of forcing brokers to question brittle, database-specific APIs, Paperclip digitizes unstructured knowledge and maps disparate databases right into a unified, AI-native digital file system.
This construction permits brokers to entry information from thousands and thousands of papers utilizing commonplace file-system operations.
"This mainly exhibits that we will get significantly better accuracy should you use Paperclip, and we will scale back the time and the fee by over an order of magnitude in comparison with should you use brokers with out these AI-native scientific infrastructures," Zou acknowledged.
Actual-world validation
To check the sensible output of this structure, Digital Biotech spun up 37,000 "medical trial brokers" to synthesize fragmented trial knowledge. These brokers recognized single-cell options that predict trial success — drug targets supported by these options have been about 50% extra prone to attain market than comparable medicine with out them.
The system then autonomously designed an antibody-drug conjugate (ADC) concentrating on the CD276 protein for lung most cancers. The brokers accomplished this design autonomously, relying solely on knowledge printed previous to January 2025.
A number of months later, Zou stated, pharmaceutical firm Merck independently developed and validated the identical therapeutic design — which went on to obtain breakthrough designation from the FDA. He characterised this as "a third-party exterior validation of the therapeutic design supplied by the digital biotech brokers."
Designing ecosystems, not workflows
As multi-agent techniques scale, leaders should rethink how they handle these digital workforces. Zou advocated for shifting from designing inflexible workflows to creating open environments. Workflows dictate the precise steps an agent ought to take, just like managing a junior worker. Environments present the infrastructure, guardrails, and incentives for brokers to collaborate on open-ended issues.
"In workflows, we're making an attempt to inform brokers what to do and find out how to do their job. However in environments, we're offering the infrastructures, the incentives, and the guardrails, however in any other case we go away it open to incentivize brokers to collaborate," Zou stated.
Optimization at scale means engineering the atmosphere reasonably than fine-tuning particular person fashions. Whereas single brokers can enhance by way of reinforcement studying or supervised fine-tuning within the agent college, the success of a large multi-agent system depends on adjusting the parameters governing their collaboration.
"On the multi-agent [side], we're not truly fine-tuning and altering the person fashions anymore, however we're optimizing the atmosphere," Zou defined. "The atmosphere itself is the thing that we optimize to enhance the brokers."
