b12 Labs: AI planning and automation for chemical synthesis

b12 Labs is building a software-to-lab bridge for pharma and biotech teams that need to make difficult molecules, not just generate another plausible route on a screen. Its product is a set of AI tools for synthesis planning, condition optimization and automation-ready execution.

What it does

The current b12 homepage ↗ breaks the workflow into three products. Selenium proposes ranked retrosynthesis strategies from a target molecule. Palladium designs high-throughput condition screens across catalysts, ligands, bases and solvents. Cobalt translates an approved method into protocols for lab automation platforms such as Chemspeed and Opentrons. The site says each tool can be used independently, and that scientists can describe intent in natural language without writing the protocol code themselves.

The buyer is a chemistry or process-development group with real experiments, expensive equipment and a planning-to-execution bottleneck. b12 says teams can start free at its app, but the checked pages do not publish a full production pricing model. Its YC launch material ↗ describes paid pilots and a broader goal of connecting agents to robotic execution; those are company-reported program claims, not an independent lab result.

Why I’d look closer

The founders’ backgrounds are unusually close to the product. b12 identifies Andres Bran as a chemistry-and-AI researcher who built agents for robotic molecular work, and Zlatko Jončev as a medicinal chemist with Roche and lab-robotics experience. The public team page ↗ makes the division clear: synthesis expertise, agentic systems and research support in one small team.

The advantage is workflow continuity. A chemist can begin with a target, move through route selection and plate design, then get an instrument-ready protocol without manually translating between disconnected tools. The tradeoff is that chemical plausibility is not the same as a validated experiment. Teams still need their own review of reagents, constraints, provenance, instrument behavior and results before a robot runs anything.

What I’d ask

Which reactions and instruments are supported in our lab today? Can every suggested route show its sources, confidence and assumptions? What approval step exists between a natural-language request and a robot command? How are failed, partial or unsafe runs stopped and recorded? Does the data model preserve our internal procedures and experiment history?

My editorial take

Shortlist b12 if your chemistry team loses time translating between route design, optimization and automation. Start with the planning layer on owned chemistry data, then expand toward execution only when the lab can review and reproduce the generated protocols. Its pitch is strongest as an integration and human-operability product, not as permission to remove scientific judgment.

Quick facts

Sources checked

Cohort context

b12 Labs is listed in Summer 2025. In our 2026-09-18 directory snapshot, 11 of 166 listed companies in that cohort have YC’s primary industry label Healthcare (6.6%). This is a current-directory comparison, not an original intake count or a performance ranking. Nine-cohort dataset.

Public website snapshot

Observed 2026-09-19 16:18 UTC in raw homepage HTML. This records visible metadata and advertised links, not agent execution or product quality.

Public observations · Collection method. Missing links here do not establish that a capability or file is absent elsewhere.