Roman Engeler

Roman Engler: Atler, Evals for AI Agents & AGI Timelines

Atler co-founder Roman Engler walks through his robotics/RL background, why the team did YC and stayed in London, and how Atler helps teams find and fix AI agent failures at scale, citing roughly 80–87% failure-spotting accuracy from a research collaboration with Cohere. The conversation then widens to AGI timelines — possibly two to three years if the current pre-training/post-training paradigm holds — the limits of transformers and end-of-task rewards, and AI safety risks ranging from misuse to chatbot-related harm.

35:22