Browse · People
Shawn Lewis
Weights & Biases cofounder and former CTO, now leading AI software-platform efforts at CoreWeave.
Shawn Lewis is Chief Superintelligence Officer at CoreWeave, leading efforts to build and scale software platforms for AI researchers and engineers. Previously a founder and CTO of Weights & Biases, he helped establish its tools for experiment tracking, evaluation, and AI observability. CoreWeave acquired the AI developer platform; its 2026 speaker biography identifies Lewis's current role.[3]
Tools for the AI research loop
Lewis describes the combination of Weights & Biases and CoreWeave as a way to connect training and inference with tracking, evaluation, observability, feedback, and hardware. In his 2026 account, he states: “A unified platform built for the entire loop was why we brought our companies together.”[4]
His public direction also extends beyond tools for machine-learning researchers. Lewis describes engineers shipping agents, domain experts reviewing outputs, and internal teams building and maintaining agents as increasingly common participants in AI development. He presents tooling for these groups as a growing responsibility. As one example, he describes ARIA within Weights & Biases as helping analyze experiments, generate hypotheses, launch runs, and evaluate results to automate the research loop.[4]
Lewis has also published an account of his own programming-agent development. In his January 16, 2025 article, “Creating a state-of-the-art AI programming agent with OpenAI’s o1”, he reports: “In fact, I ran 977 evals in the course of achieving this solution.”[2]
Earlier history: building Weights & Biases
Before this broader platform effort, Lewis and his fellow founders began with experiment tracking. Lewis recalls the team building its first tracker in a San Francisco karate studio, with OpenAI, Toyota Research, and Uber adopting it early. Cofounder Lukas Biewald explains the practical problem in his 2023 interview: researchers logged training results to locally hosted TensorBoard, then emailed screenshots because colleagues could not open the underlying results.[4][5]
Biewald credits Lewis with calling their early approach “demo-driven development”: the founders built demos convincing enough to earn longer meetings. Biewald recalls working closely with OpenAI and Toyota as early design partners. He also describes a week spent cleaning up the interface after Hamel Husain criticized its usability, and believes that cleanup helped the product begin growing steadily. These were collective founding-team efforts.[5]
Before Weights & Biases, Lewis held engineering roles at Google and was founder and CTO of Beep.[3]
Timeline
- research
Publishes his account of building an AI programming agent with OpenAI’s o1
Lewis has also published an account of his own programming-agent development. In his January 16, 2025 article, “Creating a state-of-the-art AI programming agent with OpenAI’s o1”, he reports: “In fact, I ran 977 evals in the course of achieving this solution.”[2]
Sources
- 1
- 2Creating a state-of-the-art AI programming agent with OpenAI’s o1Accessed 2026-09-20
- 3Fully Connected 2026: SpeakersAccessed 2026-09-20
- 4Celebrating the One Billion Runs Milestone on Weights & BiasesAccessed 2026-09-20
- 5How Weights & Biases found product-market fitAccessed 2026-09-20
Model source and teaching reviews passed. Human editorial review pending. This profile describes the sources as of 2026-09-20.