AI in health and life sciences is at an inflection point. The hype is loud, the demos are impressive, and the press releases promise everything. But the reality in the lab, where scientists actually have to work with these systems, is messier, more interesting, and far less understood. I’ve been there. I’ve built these systems. I know how they actually work.
I’ve spent my career at the intersection of design, product, science, and engineering: building clinical genomics products, founding life science companies, and now designing human-AI collaboration systems for drug discovery and clinical trials. What I keep seeing is a gap between how we talk about AI collaboration and how it actually works when domain experts try to use it.
This Substack is where I’ll write about that gap.
What’s real, what’s hype, and what we’re getting wrong. Where AI in science is genuinely transforming workflows, and where it’s creating new problems we haven’t named yet. Lessons from building these systems every day. Not theory, but practice.
I also want to explore what this means for the future of work. As AI reshapes how science gets done, what happens to the roles?
What does it mean to be a bench scientist when AI can design your experiments? A computational biologist when AI runs the pipelines? A researcher when AI generates hypotheses?
And beyond science: what does it mean to be a designer, an engineer, or a PM when AI agents are part of the team? When the lines between who writes the code, who shapes the product, and who makes the decisions start to blur?
The boundaries are shifting. I’m interested in what emerges. And I want to take a scientific approach to it, while applying this technology, I see so many benefits, but also aforementioned gaps. We need to be intentional.
How do domain experts maintain agency when working with multiple AI agents?
This is the question I keep coming back to. As AI systems become more capable, and more agentic, how do scientists stay in control? What interaction patterns support genuine collaboration versus just delegation? When should humans lead and when should they let AI run?
I don’t have all the answers. But I’m building and testing these systems daily at AWS Applied AI, and I’ll share what I’m learning (after the product releases, of course)
Field reports: What’s actually happening at conferences like NeurIPS, in research labs, and inside companies building AI for life sciences
System design: How to build human-AI collaboration that actually works for domain experts
The messy reality: Where AI fails, where it surprises, and what the hype cycle gets wrong
Future of work: How roles in science and tech are evolving as AI becomes a collaborator, not just a tool
Research and tools: New papers, models, and platforms worth paying attention to
Currently: Building and studying human-AI collaboration systems for drug discovery and clinical trials at Amazon/AWS Applied AI. Exploring how bench scientists, computational biologists, and AI agents work together across the entire drug discovery pipeline.
Previously:
Roche/Genentech: Built clinical genomics (cancer diagnostics) products including NAVIFY (10K+ clinicians, CE-IVD certified) and contributed to Roche 454 & Newbler, Roche Nanopore SBX
Founder: Proprius Labs, yuFlu, UXBio
Academia: Head of Bioinformatics, 700+ research citations
Education:
PhD in Human-Computer Interaction + Bioinformatics, focused on how biology experts leverage computational tools without losing their expertise
BS in Computer Science + Philosophy, machine learning, computer vision, and the philosophical implications of AI
I just got back from NeurIPS 2025, where AI in biology was standing room only. Literally. The first post will be a field report on what I saw, who’s building what, and why this field just hit an inflection point.
Subscribe to get it when it drops.
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.