Chris White is the president and chief technology officer of Prefect, the workflow orchestration company he joined as its first employee and founding engineer. He builds infrastructure that helps teams coordinate data, machine-learning, and AI systems when execution is unpredictable and failure is unavoidable.
White earned a doctorate in mathematics at the University of Texas at Austin, researching non-convex optimization, before becoming a data science manager at Capital One. There, he helped develop a Python platform connecting trained models with business analysts. His involvement with Dask and distributed workflows led him to Prefect founder Jeremiah Lowin; White joined the company in 2018, ahead of its open-source launch in 2019.
At Prefect, White challenged orchestration tools built around rigid, scheduled batch jobs, arguing that real data workflows involve changing parameters, distributed execution, and unpredictable conditions. His work on Prefect 3.0 advanced event-driven workflow orchestration and introduced transactions for coordinating work more reliably. In August 2025, he became Prefect’s president, expanding his responsibilities beyond technical leadership.
- Design around failure modes. White treats expensive, unreliable, nondeterministic model APIs as distributed-systems dependencies. His approach to production AI observability prioritizes monitoring, recoverability, runaway-process detection, and clear diagnostics before agent workflows reach production.
- Make AI usable by other software. White advocates typed machine-to-machine interfaces, structured outputs, and function calling because predictable contracts make model behavior easier to validate, test, and integrate into backend systems. He resists making chat the default interface for every AI capability.
- Experiment publicly; ship selectively. Prefect’s open-source Marvin project created room to explore AI-native developer tooling without forcing speculative features into the company’s core product. Within Prefect itself, White has favored focused improvements such as error summaries and faster troubleshooting, with explicit customer consent before information reaches external model providers.
