Modern biology’s dirty secret isn’t the science. It’s the workflow. A researcher finishes sequencing, manually exports to a spreadsheet, debugs a Python script that breaks between conda environments, re-creates the experimental setup from scratch, and stitches together three databases that weren’t designed to talk to each other. Breakthroughs happen in spite of this infrastructure, not because of it.
On February 3, 2026, Kexin Huang and Yuanhao Qu — both Stanford AI and biology PhDs — launched Phylo and its first product, Biomni Lab: an integrated workspace where agentic AI handles the entire research stack from question to result. Alongside the launch came a $13.5 million seed round co-led by Andreessen Horowitz and Menlo Ventures’ Anthology Fund, with Anthropic as a direct participant.
This is not another AI chatbot for scientists. Biomni Lab is a systematic bet that biology is ready for the same AI-native productivity leap that GitHub Copilot delivered to software engineering — and the early data from Ginkgo Bioworks suggests the bet is already paying off.
The Fragmentation Problem Has a Name Now: Pre-IBE Biology
Biology as a discipline has never lacked data, tools, or intelligence. What it has lacked is integration. Scientific literature lives in PDFs behind paywalls. Data lives in spreadsheets. Analysis runs through R and Python scripts patched together by researchers acting as their own DevOps teams. Results get compared manually, often irreproducibly.
Huang’s framing in the launch announcement is sharp: this is exactly what software development looked like before AI-native tooling arrived. The parallel holds. Biomni Lab introduces what Phylo calls an Integrated Biology Environment (IBE) — a single workspace where an AI agent orchestrates over 300 biological databases, domain-specific tools, molecular AI models, and even external lab services end-to-end.
The IBE concept is the actual innovation here, not any individual model. Biomni Lab’s agent decomposes a biological question into tasks, retrieves relevant literature, writes and runs code, dispatches jobs to lab automation systems, and returns versioned, auditable results. Every step is logged. Every run is reproducible. The agent outperforms comparable systems by more than 20% on standard benchmarks — though the benchmark methodology is Phylo’s own, and independent validation is still in progress.
Ginkgo Bioworks Already Has the Numbers
Theory only gets you so far. The most credible signal in Phylo’s launch announcement is the Ginkgo Bioworks case study — and it’s specific enough to take seriously.
Ginkgo’s data science team used Biomni Lab to run more than 10 complex cell-painting and transcriptomic analyses. Workflows that typically take weeks were reduced to hours. The results were validated by Ginkgo scientists and described as publication-quality. Ayla Ergun, Senior Director of Data Science at Ginkgo Bioworks, said the platform was accessible to scientists across disciplines — not just bioinformaticians — and that Ginkgo’s Datapoints division is considering adopting Biomni as a standard workflow tool.
Weeks-to-hours compression on validated, publication-grade biology is not a demo. That’s a production signal. It’s also exactly the kind of closed-loop throughput acceleration that separates real agentic biology infrastructure from marketing copy.
The open-source predecessor — also called Biomni, released in June 2025 — had already reached 7,000+ labs, biopharma companies, and healthcare organizations before Phylo raised a dollar. a16z’s Jorge Conde cited sustained user adoption as one of the primary investment signals: rare traction in life sciences, where adoption typically lags product launch by 18 months or more.
The Advisory Bench Is Doing Work Here
Phylo’s scientific credibility isn’t just founder credentials. The company’s advisory and co-founder roster reads like a who’s who of the two fields being merged.
Scientific co-founders include Stanford professors Jure Leskovec (graph neural networks, social network analysis) and Le Cong (CRISPR engineering). Founding advisors include Nobel laureate Carolyn Bertozzi, CRISPR pioneer Feng Zhang, and computational biology leader Fabian Theis. The founding team’s prior research outputs span Biomni, POPPER, CRISPR-GPT, TDC, DeepPurpose, ClinicalBERT, and TxGNN — a track record that signals genuine cross-disciplinary depth, not just AI applied to biology as a market opportunity.
That matters because agentic biology at production scale requires more than a capable LLM. It requires biological domain knowledge deep enough to know when the agent is wrong, experimental design expertise to catch protocol errors before they propagate, and safety frameworks sophisticated enough to satisfy both biosecurity requirements and eventual FDA electronic record standards. Phylo claims its action logs already meet those standards, though enterprise validation will be the real test.
This Is Figma for Biology, Not a Smarter PubMed
The framing that a16z uses in their investment memo is deliberately provocative: designers have Figma, analysts have Excel, engineers have GitHub — what do biologists have? Until Biomni Lab, the honest answer was: a browser, a terminal, and institutional memory.
Biomni Lab’s ambition is to become the tab that’s always open on every scientist’s laptop. Not a tool that answers questions, but a persistent workspace where the agent maintains context across experiments, remembers what was tried, connects results to literature, and proposes what to try next. Scientists define the question. The agent handles the mechanics. Scientists review, steer, and decide.
This is a different architectural claim than “AI for drug discovery.” It’s a claim about operating infrastructure — the same category shift Amy Webb’s Convergence Outlook 2026 identified when it named Living Intelligence and Programmable Biology as core economic forces, not research trends.
The open-source Biomni community will continue through Project Biomni. The commercial Biomni Lab is where Phylo will build the enterprise product. That two-track strategy — open research stack plus commercial platform — mirrors what worked for companies like Hugging Face and, earlier, Red Hat. It keeps the scientific credibility intact while building the business.
The Real Test Is the Feedback Loop, Not the Launch
Biomni Lab is live in research preview. The $13.5M from a16z, Menlo/Anthropic’s Anthology Fund, Zetta, Conviction, and SV Angel funds scale-up of production systems, continued open-source maintenance, and agentic AI research for biological applications — specifically next-generation open-weight agents and real-world evaluation frameworks.
What Phylo is building toward is clear: an AI operating system for biology that spans the full research lifecycle. Not a faster literature search. Not a smarter sequence aligner. An end-to-end environment where biology can be conducted at the speed and reliability of software engineering.
The question isn’t whether the vision is compelling. It clearly is. The question is whether the feedback loops hold at scale — when agents are running experiments across hundreds of labs simultaneously, generating hypotheses that cost real money to validate, and operating in regulatory environments where errors compound. Biological data is messier than code. Biology doesn’t have unit tests.
Ginkgo is the first real-world pressure test. The results are promising. The full verdict is still being written — one experiment at a time.
References
- Huang, K. et al. (2026). Phylo Launch Announcement. Phylo Blog. https://phylo.bio/blog/company-fundraising-announcement
- PRNewswire. (2026). Phylo Introduces Biomni Lab. PRNewswire. https://www.prnewswire.com/news-releases/phylo-introduces-biomni-lab-an-integrated-environment-for-ai-native-biology-302677036.html
- Conde, J. (2026). Investing in Phylo. Andreessen Horowitz. https://a16z.com/announcement/why-we-invested-in-phylo/
- GenomeWeb. (2026). Phylo Raises $13.5M in Seed Funding. GenomeWeb. https://www.genomeweb.com/business-news/phylo-raises-135m-seed-funding-launch-ai-enabled-biomedical-lab-environment
- Wang, W. et al. (2025). Agentic Lab: An Agentic-physical AI system for cell and organoid experimentation. bioRxiv. https://doi.org/10.1101/2025.11.11.686354
Related: Amy Webb’s Convergence Outlook 2026: Agentic Biology Is No Longer a Forecast · What Is a Biocomputer in 2026? · Programmable Biology: When Cells Become Living Software
Feature image: AI-generated using Grok.