
If AI Isn’t Failing You, You’re Not Pushing It Enough
Jerry Tworek founded Core Automation on one equation: research plus agents. Six months in, his team uses agents to do as much of its own research as possible, and that research goes into building better agents. Working that closely with agents every day means he mostly sees where they fail. In this episode, Josh and Markiesha sit down with Jerry to unpack why today’s agents still disappoint him: why they struggle with high-level research and system design, why he believes the fix is agents that learn from experience, and why AI adoption is as much a people problem as a technical one. Jerry also talks through why every company needs someone responsible for AI adoption, how to keep continually learning models under control, and why he hopes AI lets people work a lot less without sliding into a WALL-E future. What You’ll Learn: Why today’s agents still struggle with high-level research reasoning and software system design Why Jerry thinks agents need to learn from experience instead of being fixed one failure at a time Why forward deployed engineering is as much about people as technology Why every company needs one person responsible for AI adoption What changes about AI safety when models keep learning Why founders’ ambitions haven’t kept up with what AI can already do Learn more: www.nebius.com/podcast









