Sean Grove

Sean Grove is the founder and chief executive of Linzumi, a company building shared infrastructure for teams working with AI coding agents. Previously an OpenAI alignment researcher and co-founder of OneGraph, he focuses on making human intentions explicit enough for software and AI systems to execute, evaluate, and preserve.

In 2018, Grove co-founded OneGraph with Daniel Woelfel, connecting services such as GitHub, Stripe, and Salesforce through a unified GraphQL interface. Netlify acquired the company in 2021, and Grove became a principal architect working on its developer platform and API ecosystem.

At OpenAI, he worked on alignment reasoning, post-training, and model-specification infrastructure. His contributions included collective alignment work incorporating public feedback into model-behavior policies. He founded Linzumi in 2025; the company joined Y Combinator’s Spring 2026 batch.

• Specifications as the new source code. Grove argues that durable specifications—not discarded prompts or generated code—should capture a team’s goals, constraints, and success criteria. One specification can generate implementations, tests, documentation, and tutorials while preserving the reasoning behind them. His case for specification-driven development treats clear communication as the increasingly scarce engineering skill.

• Executable alignment policies. The OpenAI Model Spec illustrates how readable, versioned behavioral requirements can anchor human judgment, generate targeted evaluations, and inform training. Grove uses sycophantic model behavior to distinguish an identifiable failure against stated policy from an unexplained change in product behavior. His approach incorporates deliberative alignment, which trains models to reason about explicit safety specifications.

• Shared context for coding agents. Linzumi gives teams a common record of decisions, approvals, changing priorities, and intended outcomes while multiple agents execute concurrently. Grove treats human judgment and verification as operational bottlenecks that faster code generation alone cannot resolve.

• Express, shape, prove, scale. His framework for abundant AI computation starts with expressing intent, shaping tasks for additional computation, proving that results meet requirements, and then scaling execution. He envisions tools that expose ambiguous instructions and conflicting requirements before agents act.