Who is Greg Brockman?

Greg Brockman is the co-founder and President of OpenAI, which he co-founded in December 2015 with Sam Altman, Ilya Sutskever, and others. Before OpenAI he was an early employee at Stripe and became its first CTO in 2013, helping scale the company from about 4 to 250 people. Born in 1987 in Thompson, North Dakota, he enrolled at Harvard, transferred to MIT, and dropped out in 2010 to join Stripe.

What can founders learn from Greg Brockman?

Brockman's playbook for founders building with AI is to treat human attention as the scarce resource and ration it, prototype by default because building is now cheap, delegate the code to agentic tools while keeping the judgment and review for yourself, staff lean because AI substitutes for headcount, and treat operating models as a learnable discipline you build by using them daily rather than a tool you simply purchase.

What is Greg Brockman's role at OpenAI now?

He is co-founder and President. He previously served as OpenAI's CTO before becoming President. Fortune profiled him in November 2025 as the builder steering OpenAI's data-center infrastructure, including the multiyear compute partnership with AMD, and TechCrunch reported in May 2026 that he took charge of the company's product strategy, consolidating products like ChatGPT and Codex into a unified platform.

What did Greg Brockman mean that human attention is the new bottleneck?

He argues that as models get cheaper and more capable, the limit on getting value from AI stops being compute or model quality and becomes how much human attention you can spend directing the systems. Demand for intelligence is effectively unlimited, so supply never catches up. For founders, that means designing workflows that spend scarce human hours only on the decisions and review that models cannot do.

Greg Brockman: The Founder Playbook

The playbook of Greg Brockman, Co-founder and President of OpenAI. Part of the founder playbooks.

Greg Brockman is the co-founder and President of OpenAI, and few operators in AI ship with his intensity. Founders study greg brockman because he has done the two hardest technical jobs in a startup back to back: he was the first CTO of Stripe (helping scale it from about 4 to 250 people between 2013 and 2015) before co-founding OpenAI in December 2015 alongside Sam Altman and Ilya Sutskever, a company that at first ran out of his living room. He is one of the tech founders worth modeling because he pairs deep technical range with a plain way of describing what a company actually does, and he is still hands-on: Fortune called him OpenAI's builder-in-chief for its data-center buildout, and in 2026 he reportedly took charge of product strategy. You can follow his thinking on his X profile.

Design for human attention, not model capability

Brockman argues that the scarce resource has moved. As models get cheaper and more capable, the constraint stops being compute or model quality and becomes how much human attention you can spend directing them. He describes human attention as the most important bottleneck in the use of AI systems, and he does not expect the supply side to catch up. In OpenAI's Greg Brockman: Why Human Attention Is the New Bottleneck, Greg says "No matter how fast we try to ramp compute, I guarantee you we're not going to be able to keep up with demand and that has been true ever since." His one-line description of the business makes the same point from the supply side: "We buy, rent, build compute, and we resell it at a margin. That's it."

So-what: if attention is the bottleneck, your job as a founder is to protect it. Point the team's scarce hours at the few decisions only humans can make (what to build, what "good" looks like, what to ship) and hand the mechanical work to the models. Andrej Karpathy makes the same case from the other direction: understanding, not typing, is what limits how fast you can direct agents.

Prototype by default

When building costs almost nothing, the cheapest way to settle an argument is to build the thing. Brockman points out that the cost of a first version has collapsed. In OpenAI's Greg Brockman: Why Human Attention Is the New Bottleneck, Greg says "The cost of building a prototype is cheap now. It's so cheap, and if you want to build a dashboard... you just do it now." The meeting to decide whether an internal tool is worth building now costs more than the tool.

So-what: change your default from "should we build this" to "build it and look at it." For internal dashboards, throwaway scripts, and one-off tools, a working prototype is a faster and more honest answer than a spec debate. Our guide on how to build software with AI agents walks through making this the team's reflex instead of a special occasion.

Delegate the code, keep the judgment

Brockman has watched agentic tools move from assistant to primary author inside his own field. In OpenAI's Greg Brockman: Why Human Attention Is the New Bottleneck, Greg says "The tools right now have become incredibly useful. We went from these agentic coding tools writing 20% of your code to writing 80% of your code." He gives a concrete example: a systems engineer handed a design document to a model, and overnight the model implemented the spec, optimized the code, and kept iterating until it reached an optimized result.

So-what: if the model writes most of the code, your leverage is no longer in typing it. It moves to writing the design doc, framing the problem, and reviewing the output well enough to trust it. Garry Tan frames this as directing AI tools rather than relying on them; the founders who win treat the model as an implementer they specify for and audit, not an oracle. For where this leaves engineering roles, see our take on the future of software engineering.

Staff lean and stay small longer

Brockman expects the shape of teams to change. He sees a future of work with smaller teams and more solopreneurs, because capable AI tools substitute for headcount instead of merely assisting it. That is a hiring thesis as much as a technology one: the first response to more work should be better tooling, not more people.

So-what: before you open a role, ask whether the work is attention-bound or hands-bound. If a model plus one focused person can absorb it, resist the headcount. Staying small keeps decision speed high and attention concentrated, which is the resource Brockman says is now scarcest. Founders who staff this way get more done per employee and defer the coordination tax that comes with scale.

Treat operating AI as a discipline, not a purchase

Buying access to a frontier model is easy; using it well is a skill you have to build. In OpenAI's Greg Brockman: Why Human Attention Is the New Bottleneck, Greg says "These models have such power, and really understanding how to operate them well takes thought." He also insists the learning has to be hands-on: "It's very different to hear AI described versus to use it. But, the beautiful thing about AI is it's so intuitive."

So-what: make "how we operate models" an explicit, written part of how your startup works, not a tool you bolted on. Document your prompting patterns, your review steps, and where a human sign-off is mandatory, then improve them by using the models daily rather than reading about them. That operating layer is exactly what the AI Operating System for Startups course is built to give you.

What to copy this week

Audit where your team's attention goes for one week, then move the mechanical work to a model so humans spend hours only on decisions and review.
Kill your next "should we build this dashboard" debate by just building the prototype and looking at it.
Take one real spec, hand it to an agentic coding tool overnight, and practice reviewing the output instead of writing the code yourself.
Before your next hire, ask whether better tooling and one focused person can absorb the work; default to staying small.
Write down your team's model-operating patterns (prompts, review steps, human sign-off points) and improve them by using the models every day.

AI Operating System for Startups