Every venture firm has an AI tool, usually ChatGPT, Gemini, Claude, or Perplexity. A few organizations have built something structurally different, and that gap is about to widen.
I am VP of AI Strategy & Enablement at VIC Tech, a life science venture studio. Over the past two years, we've gone from treating AI as a search engine to building something closer to operating systems. The work demanded it. When you're evaluating university technologies, building companies from scratch, and guiding them through regulatory and fundraising milestones, a chatbot in a browser window is not going to get you there.
This is the first in a series of posts about what we've learned building AI into how a venture studio operates. The version where things break, and you fix them, and the system gets better because someone went back and figured out what went wrong.
From search engine to operating system
Life science venture capital sits at the intersection of several domains that are individually difficult for AI and collectively even worse. Patent claims require legal precision. Preclinical data requires scientific judgment. Regulatory strategy requires understanding how agencies actually behave, not just what the guidance documents say. Financial modeling for early-stage biotech requires tolerating uncertainty that would make a fintech algorithm panic. A general-purpose AI will do a passable job on most of these and a dangerous job on some. In life science, an answer that is mostly right can still be dangerously wrong. The gap is where an investor spots the weakness you missed or a regulatory submission hits a wall.
Most organizations adopt AI by plugging it into existing workflows. Summarize this document. Draft this email, those memos. Score these leads. Build decks. The assumption is that the work stays the same, but now it's faster. I think this framing misses the larger opportunity.
At VIC Tech, we built a system of specialized AI capabilities rather than relying on one generalist tool. The core is designed to catch AI hallucinations, reduce over-agreement, and enforce source verification; in regulated domains, being confidently wrong is worse than being slow. On top of that core sit persistent memory systems and a knowledge base that preserves institutional context: reference documents, team deliverables, analyses that feed every subsequent workflow. Specialized agents handle specific parts of the venture process, from IP evaluation and due diligence to financial model review and regulatory pathway analysis, and they pass context to each other rather than working in isolation. Purpose-built tools support the team across front-office and back-office workflows, working in concert with the agents.
Humans make the decisions, always. The system doesn't decide; it makes the decision-maker faster, more informed, and harder to fool. A pitch deck that says "our AI makes the investment decisions" is a red flag. One that says "our AI makes our team's decisions more rigorous" is a real edge.
Speed matters. Some evaluations that used to take weeks now take days. But speed without quality is just faster mistakes. Our system runs work through multiple independent review layers before a human sees the output. If an initial analysis says a patent portfolio is strong, the review layer asks: strong compared to what? Based on which claims? What about the prior art in adjacent therapeutic areas? This is quality assurance for knowledge work, and it catches errors that a single pass, whether by a human or by AI, would miss.
When AI improves outcomes, not just speed
Our grant writers at VIC Tech have used AI tools we built to critique their own proposals, identifying weaknesses they had missed. They are editing their language and structuring their evidence accordingly. In a separate case, an IP landscape analysis built on our system gave a portfolio company the preparation it needed for a partnering discussion, a conversation that, by the team's account, went considerably better because of the analysis they brought in. The scientist writes; the system sharpens.
Our AI infrastructure covers the full venture lifecycle: from scouting university technologies through IP assessment, regulatory positioning, financial modeling, fundraising support, and back-office operations like legal and financial management. Every portfolio company gets access from Day 1. No need to build their own AI stack; the operating system is already running when they arrive.
This is at the core of our strategy. Build a solid foundation that can serve the entire VIC Tech ecosystem, where I build and maintain the core architecture, develop advanced tools, and educate and support our community to enable them to build individual tools that address their evolving needs and enrich the ecosystem.
What's next
In upcoming posts, we'll go deeper: how AI is changing our due diligence workflows, what it means to hand a portfolio company an AI operating system, and why we believe a single AI model reviewing its own work is never good enough.
One last thing. I wrote this post using the system I've been describing. The AI helped with research, structuring, and drafting. The thesis, the examples, the wins, and the final call on every sentence were mine. That's the model, and if it reads as if a person wrote it with the help of a very capable tool, that's because one did.
If you work in tech transfer, run an innovation office, or are building a life science company and want to see how AI can work for you subscribe to our newsletter for future posts.
-----
Andres Lorente, PhD, is the VP of AI Strategy & Enablement at VIC Tech, a life science venture studio that builds companies around university-born technologies. His expertise connects corporate strategy, business development, and AI infrastructure. He writes about where AI and venture building meet. He's also known for his collaborative style and for keeping teams engaged with dad jokes in two languages.