Rohan Saraf

How do we design systems where humans and AI can collaborate effectively — while preserving trust, understanding, control, and agency?

I've worked for 15 years on enterprise systems where the cost of getting it wrong is measured in patient outcomes, financial decisions, and clinical trust. That context shapes how I think about AI.

I use product creation as a research method. Instead of writing theories about AI interaction, I build systems and observe what happens. The laboratory is not separate from the work — it is the work.

Two sides of the same problem: AI that is legible to humans, and software that is legible to AI agents. The gap between them is where the interesting design questions live.

From Human Factors to Agent Factors — the same discipline, extended to a new kind of actor in the loop.

On the work

I started at the Human Factors Institute, where I learned that failures attributed to user error are almost always failures of system design. That insight has organised everything since.

At The Minimalist I built a practice from scratch. At GE HealthCare I work where the complexity is not interface complexity but epistemic complexity — how do you encode clinical knowledge in a way that is trustworthy, updateable, and legible to the people who act on it?

The AI laboratory started from a different kind of question. Not "how do I use these tools?" but "what actually happens when you build systems with these tools and live with them?"