The bottleneck in modern science isn’t always the experiment; often, it’s the sheer volume of reading required to design one. Andrew White, an AI-for-science researcher, has raised a seed round to fix this with a “flock” of autonomous agents.
Co-founded with CEO Sam, the new startup is pivoting from academic theory to venture-backed execution. While the company declined to disclose the specific capital raised or current valuation, the operational footprint tells its own story. The team has already swelled to nearly 30 scientists and engineers—a headcount that implies a significant burn rate and a substantial seed check.
“We raised our seed round! It’s been an insane six months,”
White noted, confirming the shift from proof-of-concept to product development.
The “Flock” Architecture
White’s thesis stems from his time at the University of Chicago and the University of Rochester, where he ran a lab attempting to bridge molecular simulations with messy experimental data. He found that brilliant researchers were losing weeks to drudgery: manual literature reviews, cross-checking conflicting papers, and slow trial-and-error planning.
Instead of relying on a single, monolithic Large Language Model (LLM) which can be prone to hallucinations, the company is building a multi-agent system. This “flock” comprises specialized agents—nicknamed Crow, Finch, and Owl—that divide the cognitive labor.
The workflow is designed to mimic a human lab team:
- One agent scans the literature and data.
- Another proposes hypotheses or maps out experiments.
- A third acts as a critic, cross-checking claims against the scientific record to flag inconsistencies.
This verification layer is the startup’s primary technical moat. In computational chemistry, a hallucinating model isn’t just annoying; it’s a waste of expensive lab resources. By forcing agents to challenge one another, the team aims to produce results that are scientifically defensible, not just plausible-sounding.
Early Validation
The company is currently targeting pharmaceutical researchers and materials scientists—sectors where shortening the discovery timeline directly correlates to billions in value. They have already produced a notable case study: the platform identified an existing drug as a potential treatment for a common cause of blindness. The system completed the identification and reasoning process in three weeks, a task that typically consumes years of manual research.
The Market Reality
The funding comes at a noisy time for the sector. “AI for Science” has become a crowded vertical, with both incumbents and new entrants racing to apply foundation models to biology and chemistry. However, most existing solutions still require heavy human hand-holding.
White’s team is betting that an agent-based approach—where the AI does the reasoning and the verification—can outpace tools that simply summarize text. With a massive team for a seed-stage company, the pressure is now on to prove that their digital flock can consistently replicate the success of their early blindness study across new, more complex domains.
