Artificial intelligence has already rewired the front end of the drug industry, accelerating the search for new molecules. Katalyze AI is betting the next frontier lies much deeper inside the pharmaceutical company: on the plant floor and in the quality lab, where scientists and manufacturing engineers still burn countless hours stitching together data from disconnected systems before anyone can make a decision.
The San Francisco startup this month announced $10.5 million in seed funding to bring what it calls an "agentic operating system" to pharmaceutical manufacturing and life-sciences operations. The round was led by Bonfire Ventures, with participation from Inovia Capital, Ripple Ventures, Alumni Ventures, and angel investors including Gokul Rajaram and Farzad Soleimani.
What Katalyze is building
Katalyze's pitch is that the biggest untapped opportunity in life sciences is not another chatbot bolted onto existing software, but the connective tissue underneath it. The platform lets scientists, engineers, and analysts assemble teams of specialized AI agents that take on real engineering, scientific, and manufacturing work, drawing on verified operational data rather than a general-purpose model's best guess.
To do that, Katalyze integrates the systems that run a modern biomanufacturing site: manufacturing execution systems (MES), laboratory information management systems (LIMS), electronic lab notebooks (ELN), process historians, and enterprise platforms such as SAP. The company knits those sources into what it describes as a single operational record that both people and agents can query. Crucially, every agent output stays linked back to its underlying data source, giving drugmakers the traceability required for regulated environments and Good Manufacturing Practice (GxP) compliance.
Under the hood, Katalyze says the intelligence rests on an operations-specific ontology and knowledge graph that give agents context for each molecule and manufacturing process, so they can investigate production issues, support quality teams, and analyze manufacturing data while keeping results auditable.
The numbers behind the bet
Katalyze is not arriving as a science project. The company says its platform is already in use at five of the world's 20 largest pharmaceutical companies, including Sanofi, an unusually deep enterprise footprint for a startup announcing its seed round. The roughly 40-person team plans to use the new capital to expand its engineering, scientific, and go-to-market groups, grow its catalog of agents, and scale those deployments.
The company also leans on a community of more than 100 tenured scientists and bioprocess engineers, drawn from firms such as Pfizer, Eli Lilly, and Sanofi, to crowdsource the "skills" its agents learn, an attempt to encode hard-won domain expertise rather than hope a large language model infers it.
The most eye-catching figure is a single early deployment: Katalyze says an analysis that would have taken roughly a year and cost between $4 million and $6 million was completed in about 45 minutes on its platform. The company did not disclose further details about the project, so the claim is best read as a directional signal of ambition rather than an audited benchmark.
Why founders think the timing is right
The backdrop is an industry squeezed from several sides. Patent expirations are eroding revenue, drug shortages continue to strain supply chains, and the cost of bringing a new therapy to market remains punishing. Yet much of the day-to-day work of research and manufacturing still depends on information scattered across lab systems, quality records, and factory software.
"The pressure to get medicine to patients faster, and at lower cost, has never been higher, but the bar for accuracy in our industry is absolute," said Reza Farahani, co-founder and CEO of Katalyze AI. "We built an agentic operating system where every answer is grounded in an immutable record, so teams can deploy agents that are right every time and cut lab and manufacturing cycles from quarters to weeks, without ever exposing sensitive data."
Katalyze was founded by Farahani alongside Shreyas Becker, Hannes Bretschneider, and Matt Cruz, a team the company says spans AI, enterprise software, and life sciences.
Its lead investor frames the appeal in terms of infrastructure over interface. "Most AI in this category is a thin copilot bolted onto legacy tools," said Brett Queener, General Partner at Bonfire Ventures. "Katalyze went the other way and built real infrastructure."
That distinction is echoed by an early customer. "What really separated Katalyze was that it was built for an enterprise like Sanofi from day one," said Sabya Dasgupta, Global Head (VP) of R&D Data Platforms and Products at Sanofi, noting that the ontology layer, data ingestion, security, governance, and deployment were already in place. "Everything we needed to scale this across R&D, not just run a pilot in one corner of the organization."
The analysis: agentic AI meets a compliance-first industry
Katalyze is a clean test case for a broader thesis sweeping enterprise software in 2026: that the value of AI shifts from assistants that suggest to agents that act. In most sectors, the risk of an agent getting something wrong is an annoyance. In regulated biomanufacturing, an unexplained deviation can halt a production batch, trigger a regulatory finding, or, at the extreme, reach a patient. That raises the bar considerably, which is why Katalyze's emphasis on traceability, immutable records, and GxP alignment is not marketing garnish but the core of the product.
The strategy also inverts the usual startup playbook. Rather than selling a lightweight tool that spreads bottom-up, Katalyze is planting itself in the data layer of some of the most conservative buyers on earth. That is hard to win but even harder to displace once embedded, which helps explain why investors are comfortable with a $10.5 million seed for a company already live inside top-20 pharma names.
The open questions are the ones every enterprise-AI infrastructure company faces: whether agents can reliably clear the accuracy bar the industry demands, whether integrations into aging MES and LIMS stacks scale without heavy services work, and whether "right every time" survives contact with the messy realities of a factory floor.
What to watch
Watch for named, quantified customer outcomes that go beyond the single 45-minute anecdote, ideally validated by the pharma companies themselves. Watch how quickly Katalyze converts pilots into enterprise-wide rollouts, the exact transition Dasgupta flagged. And watch the competitive response: with drugmakers under margin pressure and every major software vendor racing to ship agents, the manufacturing-and-quality layer of life sciences is about to get crowded. Katalyze's wager is that being the trusted operational record, not just another agent, is what will keep it standing when it does.
"We built an agentic operating system where every answer is grounded in an immutable record, so teams can deploy agents that are right every time and cut lab and manufacturing cycles from quarters to weeks, without ever exposing sensitive data."— Reza Farahani, Co-Founder and CEO, Katalyze AI