Debugging AI Agents at Scale: Roman Engeler on Traces, Guardrails & Atla

Host Susan Kish talks with Roman Engeler, ETH Zurich alumnus and co-founder of Atla, about how modern AI agents make decisions and how to keep them safe, reliable, and fast. Roman explains traces – the step-by-step records of an agent’s reasoning – s and shows how Atla helps teams debug, evaluate and improve agent performance at scale. He shares the move from ETH research to building a company in London, compares startup cultures in Zurich, Palo Alto and London, and reflects on how ETH’s rigor translates into product execution. The conversation covers the promise and risks of growing autonomy, why guardrails matter for real deployments, and ways alumni can give back, including the Founder’s Pledge. Clear, candid and practical, this episode is for anyone shaping the next wave of intelligent systems.

Chapters:

(00:00:48) Living in London: Startup Scene Comparison

(00:03:36) From Research to Startups: Roman's Journey

(00:05:33) What Are AI Agents and Traces?

(00:07:30) Introducing Atla: Debugging AI Agents

(00:09:24) Case Studies: Real-World Applications

(00:11:08) Pricing Model and Customer Success

(00:13:43) The ETH Difference: Rigor and Freedom

(00:16:05) Founder's Pledge: Giving Back to ETH

(00:18:21) The Future of AI: Progress and Predictions

(00:21:42) Advice for Future AI Students

(00:23:42) Giving Back: Scholarships and Support

(00:25:19) Personal Questions: Growing Up and Early Dreams

(00:26:03) Current Reading: Biographies and AI Safety

(00:28:34) Favorite Zurich Spot and Closing