The Makers: Fiddler CEO Krishna Gade on Why Visibility Is Not Control
Every enterprise is putting AI agents into production. Very few can say what those agents did yesterday, and fewer still can stop one from doing it again tomorrow. A trace shows you what happened. It does not decide what is allowed to happen. Krishna Gade has been working on this problem since before it had a name. He led the News Feed ranking platform at Facebook and worked on the "Why am I seeing this?" feature, then founded Fiddler in 2018 to make enterprise AI explainable. Today, Fiddler is an AI control plane: it evaluates, monitors, and enforces policy on models and agents in production, so enterprises can see what their AI is doing, stop it from doing the wrong thing, and prove both to an auditor. It runs inside banks, healthcare payers, and the US Navy. In this live conversation with Andreas Sjostrom, we trace the full arc: 🌍 Big picture: why agents are distributed systems wearing a prompt, why plausible output is a failure mode rather than a feature, and why seeing an agent is no longer the same as controlling it 💡 Spark: the starting days in 2018, when nobody was asking for AI observability. What Krishna and co-founder Amit Paka saw inside Facebook, Pinterest, and Twitter that convinced them AI would become a first-class citizen in the enterprise software stack and would need to be explained, and how they decided to bet a company on it 🛠️ Make: the first build. Explainability for individual model predictions, the early pilots with banks and lenders, the customers who asked why they couldn't just use their existing monitoring stack, and the moments that forced Fiddler to become an observability platform and then a control plane 📈 Traction: how Fiddler is used today inside Nielsen, the US Navy, and Fortune 100 financial and healthcare institutions, what breaks first when agents move from pilot to production, why regulated industries came first, and how the company is scaling after a $30M Series C 🔭 Horizon: Krishna's view of where AI governance is heading. Whether a neutral control plane survives once the hyperscalers and model labs build governance into their own platforms, what conventional wisdom about agent safety will look wrong in three years, and what accountability means when software acts on its own Chapters: 00:00 Welcome & Introduction 02:10 Meet Krishna Gade & Fiddler's Mission 03:46 What Is the AI Control Plane? 05:07 Cost, Security & Quality for CIOs 07:43 The Makers: Fiddler's Origin Story 11:31 From Model Observability to the Control Plane 13:15 Visibility Is Not Control 16:22 Nielsen Case Study & Agent Categories 25:29 Live Demo: Guardrails in Action 38:43 Centaur Models: How They Work 40:03 Long-Running Agents & Cost Governance 44:04 Real-Time vs Forensic Governance 47:19 Control Plane Architecture & Independence 52:46 Competitive Differentiation & The Name 55:29 Looking Ahead & Closing Learn more about AI Control Plane: https://www.fiddler.ai/control-plane