# Run a virtual lab meeting of AI scientist agents

Stanford's Virtual Lab agents designed 92 nanobodies; the team reports in Nature that two bound newer COVID variants better in lab tests.

Organisation: Stanford University (Zou group) and collaborators
Status: Research (published in Nature)
Place: United States
Categories: Science & discovery, Health & care
Built with: ChatGPT / OpenAI models
Try it with: Claude, ChatGPT / OpenAI models, Gemini, Grok
Time to try: 30 to 60 minutes

## Why it matters

Fast-mutating viruses need fast responses. A team of AI agents, with a human giving high-level feedback, designed binders that held up in the lab.

## How it works

An LLM 'principal investigator' agent runs research meetings with LLM scientist agents, and a human researcher gives high-level feedback. The team built a design pipeline using ESM, AlphaFold-Multimer and Rosetta to design the nanobodies. The two that bound better did so against the JN.1 or KP.3 variants while still binding the ancestral spike well (Nature). The code is open source and uses OpenAI models by default.

## Who it's for

Research groups, biotech teams and anyone running complex technical projects.

## The prompt

```
Run a virtual lab meeting on this project: [project goal].

Agents:
- Principal Investigator: sets the agenda, asks questions and makes the final decisions.
- [Specialist 1, e.g. immunologist], [Specialist 2, e.g. computational biologist], [Specialist 3, e.g. machine-learning engineer].
- Scientific Critic: points out errors, weak evidence and missing controls.

Hold three rounds of discussion. Each agent speaks once per round and must respond to the critic. Then the Principal Investigator writes:
1. The decision and the reasoning behind it.
2. The tools or methods chosen and why.
3. A step-by-step plan with owners and success criteria.
4. Open questions for me, the human researcher.
```

## Caveats

Lab-validated binding only; not a therapeutic.

## Sources

Both sources are the Stanford team's own work: its peer-reviewed Nature paper and its open-source code. No independent evaluation was found.

1. Nature: PI agent plus scientist agents; ESM, AlphaFold-Multimer, Rosetta; 92 nanobodies; two improved binders. The Virtual Lab of AI agents designs new SARS-CoV-2 nanobodies (https://www.nature.com/articles/s41586-025-09442-9)
2. GitHub (Zou group): Open-source code; nanobody design notebook; uses OpenAI models. zou-group/virtual-lab (https://github.com/zou-group/virtual-lab)

Fact-checked: 16 Oct 2026

Page: https://escapevelocitysociety.com/picks/stanford-virtual-lab-nanobodies

See Selected Projects: https://escapevelocitysociety.com/picks.md
