Greg Brockman is the president and co-founder of OpenAI and the former founding engineer and chief technology officer of Stripe. His career has centered on translating ambitious technical ideas into working systems, from online payments to frontier AI research, coding agents, and large-scale computing infrastructure.
From self-taught programmer to OpenAI founder
Brockman initially expected to become a mathematician. After exhausting his school’s mathematics curriculum, he took classes at the University of North Dakota while still in high school. Programming captured his attention when he wanted to put a chemistry project online, taught himself PHP through W3Schools, and realized software could make an idea immediately useful to other people.
He attended Harvard, transferred to MIT, and left to join Stripe. As its founding engineer and later CTO, he helped grow the company from four employees to roughly 250. Stripe reinforced his preference for direct customer feedback, practical execution, and questioning supposedly immovable organizational constraints.
Research-engineering partnership: Brockman considers strong engineering essential to scientific progress: promising model ideas require optimized kernels, distributed training, dependable infrastructure, and production deployment. His case for machine-learning engineers emphasizes technical humility: researchers may distrust clean abstractions when hidden defects appear only as subtle performance degradation.
AI-assisted software engineering: He expects coding agents to thrive in repositories organized around small modules, clear interfaces, explicit documentation, and fast tests. Their most consequential applications include maintaining existing systems, upgrading dependencies, and modernizing legacy code—not merely generating new demonstrations.
Long-horizon autonomous agents: Brockman envisions cloud-based coding agents continuing work while users are offline and coordinating across multiple tasks. Longer-running workflows introduce concrete challenges involving persistent state, nondeterminism, stateful tools, and virtual-machine checkpointing.
AI infrastructure and hardware co-design: He sees accelerator architecture, memory, networking, and serving software as interconnected design decisions. Long-running reasoning and low-latency interactive applications create different infrastructure demands, while specialized hardware introduces difficult tradeoffs in flexibility and resource allocation.
At the AI Engineer World’s Fair, Brockman argued that increasingly capable models still need engineers with domain expertise to build dependable applications, particularly in fields such as healthcare and education.
