Greg Brockman is OpenAI's co-founder and president, and a former Stripe CTO. His engineering work includes reinforcement-learning tools and infrastructure for making AI models available to developers.[1][2][3][7]
Start with The inside story of ChatGPT’s astonishing potential →
X · gdb 
GitHub · gdb Career
Infrastructure before OpenAI
Greg Brockman's engineering background predates the modern language-model industry. At Stripe, he was chief technology officer and helped build the company as it grew from four to 250 employees, according to his own account. His public projects include Rubysh, a library for running subprocesses from Ruby, and Einhorn, a process manager. Earlier work included systems for the Harvard Computer Society and services at MIT. That record helps explain his role at OpenAI as an engineer building the machinery around research as well as a company leader.[2]
Founding OpenAI
OpenAI's December 2015 announcement named Brockman as its CTO and Ilya Sutskever as research director. Brockman and Sutskever were also credited as authors of that announcement. The organization began as a nonprofit research company promising to publish research, share code, and pursue broad public benefit. Its founding team included Andrej Karpathy, John Schulman, Wojciech Zaremba, and other researchers and engineers. Brockman's current title is president and co-founder. The founding document records the original organization and commitments; it should not be read as a description of every aspect of OpenAI today.[5][1]
Leadership and governance
Brockman's leadership also became part of OpenAI's governance dispute in November 2023. The company's subsequent review records his removal from the board alongside Sam Altman's dismissal as CEO, followed by both men’s return to the company. In March 2024, OpenAI's board endorsed their continued leadership after a review by WilmerHale. The company announced changes to its governance guidelines, conflict-of-interest policy, reporting channels, and board committees. This account comes from OpenAI's published summary of that review, rather than an independent account of the dispute.[8]
Contributions
Reliable software services
Einhorn addresses a practical problem in operating software servers. It runs multiple copies of a long-lived process and can share a listening socket among them. Its upgrade mechanism starts replacement workers before asking old ones to shut down gracefully, allowing software to be updated without abruptly stopping the service. Workers can explicitly signal when they are ready. The project is designed to work across programming languages and frameworks. Brockman links it among his engineering projects, providing a concrete example of his earlier work on deployment and operational reliability.[4][2]
Reinforcement-learning infrastructure
Brockman is the first-listed author of the 2016 OpenAI Gym paper. Gym supplied reinforcement-learning researchers with benchmark environments that exposed a common interface, plus a way to compare results. A shared interface reduced the need to rewrite experiments for every task. His project list also includes OpenAI Five, the Dota 2 system. OpenAI's 2018 account describes five neural networks learning through large-scale self-play, addressing long time horizons, incomplete information, and coordination. These were team projects that joined research methods with substantial training infrastructure.[3][6][2]
Making models available to developers
Brockman lists the OpenAI API among his projects. Its June 2020 launch offered developers a general-purpose text interface backed by models from the GPT-3 family. Developers could specify tasks through prompts and examples rather than build a separate model for every application. OpenAI launched it in private beta, describing both the cost of running large models and the need to learn about misuse through deployment. The API connected research models to software products, while giving OpenAI a way to update the underlying models centrally.[7][2]
Tools, writing & talks
Writing, research & talks
Start here
Brockman demonstrates GPT-4 using external tools and discusses how user feedback shaped ChatGPT’s development.
Watch this 2023 talk for a nontechnical introduction to Brockman's thinking about AI that uses tools and learns from user feedback. The demonstrations document an early product vision, so treat them as history rather than a current tutorial.
Speaker · TED 2023[11]Greg Brockman describes a coalition effort to patch php2shell, a 14-year-old PHP remote code execution flaw found with OpenAI models. He covers the leaked disclosure timeline, a Node.js RCE, why manual handling will not scale, and defender access programs.
Catalog workOriginal sourceOpenAI's president explains the aggressive compute buildout as a simple resell-at-margin business and argues human attention is the new bottleneck as AI systems take on more work.
Catalog workOriginal source- OpenAI Gymarxiv.org
A toolkit giving reinforcement-learning researchers a common interface for running tasks and comparing results.
Co-authored paper[3] - Introducing OpenAIopenai.com
The 2015 founding announcement sets out OpenAI's original nonprofit research mission and introduces its first team.
Co-authored announcement · The original research organization[5]