Raise the Ceiling of Ambition

Greg Brockman is a co-founder and the president of OpenAI, the company behind ChatGPT, which he says now has 1.2 billion weekly active users. Before OpenAI he was the CTO of Stripe, where he remembers emailing the data team a question and waiting a week for the answer. He calls himself a good programmer who no longer writes code, preferring to direct an AI and give it feedback. In this conversation he argues that a wave of new businesses is coming over the next 1 to 2 years and that most people are still asking too little of their AI. These signals are from Greg Brockman's interview with Marina Mogilko on Silicon Valley Girl.

1. The Three-Use-Case Threshold. ChatGPT users who find 3 different things it can do for them become power users who stick around and want more. Getting people to those 3 use cases is hard, Brockman says, because "here's this AI that can do anything, but what's one thing it can do?" About 300 million people a week already use it for health questions, and he thinks most usage barely scratches the surface. OpenAI is working on an AI that knows its own range and suggests the next step, like turning a request into a full presentation or spreadsheet. People want simplicity, he adds, and buttons, sliders and model pickers are "not the AI we were promised."
2. The Drafted Reply. Brockman describes an AI that spots an email you've ignored for a couple of days, researches the answer, drafts the reply and asks "Do you approve?" He says this happened to him right before the interview. His point is that the burden of figuring out what the system can do should move from the user to the AI. Voice mode now connects to the same tools as chat, so you can ask it to look at your calendar and free up time for deep thinking.
3. Directing the Code. Brockman says the models now write software better than basically anyone he knows. "I don't write software anymore. I direct an AI for it to write software." He suggests asking for a small game to feel how fast it delivers, then asking for something harder. A friend used OpenAI's Astra model to produce CAD plans for a guest house, took them to a contractor, and the house is now being built.
4. The Answer Is Always No. When Marina Mogilko showed him an app she built to prepare her podcast interviews, she asked if she was ambitious enough with AI. "I think the answer to being ambitious enough will always be no because there's always more we can do," he replied. Her team still spends at least 40 hours on each episode after recording, so he pushed her to have the app handle post-production too, leaving her only to approve the questions and show up. His habit is to keep asking for something slightly more ambitious than the AI has done for him so far.
5. A Programmer on Payroll. An OpenAI employee met an antique watch dealer who said AI was useless for a watch store. The employee used voice mode to build the shop a website, then had the owner tell the AI the story behind each watch, which became a visual catalog. Brockman says small businesses are easier to start than ever because founders carry a free, PhD-level business consultant in their pocket. The question he wants every owner to ask: if you had the world's best programmer on your payroll, what would you put them to work on?
6. Differentiation After Code. Writing software is becoming available to everyone, which makes differentiation matter more. Brockman locates the durable value in understanding the problem and the users. An education product has to satisfy teachers, parents, students and administrators, each needing the right information and guardrails. Whether you write it "in C++ or Swift," he says, "who cares about that? Like that was never what the barrier should have been."
7. Enterprise Sales for Consumers. Brockman expects "this massive renaissance of entrepreneurship" to kick in over the next 1 to 2 years, and the line between consumer and enterprise businesses to blur. Enterprise sales is fundamentally about relationships, and it never worked for consumers because there are too many of them. If AI lets a company keep meaningful relationships with people at huge scale, the enterprise sales motion may start to make sense for consumer businesses. What stays fixed is delivering value and keeping a two-way relationship with the people who receive it.
8. Scaling With Smarter Models. Startups that add prompts and handcrafted workarounds to patch model limits have a hard time lasting. Brockman tells builders to pick work that gets better as the models get smarter. Each level of capability suggests a new interface: chat when GPT-3.5 and GPT-4 were good enough to read, coding agents at the end of 2025 when AI went from 20% to 80% of software production, and now cloud-based personal agents like OpenAI's Dots. A company already embedded with a hospital's patients, doctors, admins and insurers sees its product get better and more defensible with every model upgrade.
9. The Fractal Economy. Brockman says one of his biggest surprises at OpenAI is how far from optimal every sector is. Zoom into healthcare, then into individual companies, and you find low-hanging fruit everywhere. Experts specialize ever more narrowly, to the point where a biology PhD can center on one reaction, so few people connect fields. He expects new hybrid businesses, the way edtech combined education and technology, in pairings no one has tried yet.
10. The Morning Briefing Ladder. Brockman defends the much-mocked morning briefing agent because the same label hides very different depths. Level one repeats your notifications. Level two lists unanswered Slacks in priority order, and level three researches a market decision across every presentation and tells you what matters, which is where AI is today. The next level calls your coworkers and an outside partner, puts everything in order, and asks you to approve the purchase order. He makes the same point about coding: AI has written software for 5 years, and today's version is a different thing.
11. The Outage Test. OpenAI tracks token usage against commits and PRs, but Brockman's favorite signal is what happens during an internal model outage. People who had nothing like this 6 months earlier say they can't do their work without it for 2 hours. He asks his agent first about anything at OpenAI, including data warehouse queries on growth or retention curves that a Stripe data team used to answer in a week. Its finance team closes the books faster, and one comms staffer used Codex to run a journalist event's seating chart and dietary needs alone. At the far end, OpenAI pointed about 10,000 agents at the Navier-Stokes millennium problem for several days.
12. The Five-Times Rule. Brockman's advice for a blank page is to start small: type something, hit enter and see what happens. Pick one pain point, like an email you've been dreading, and do it by hand first, copy-pasting before you connect Gmail. After you've done a task 5 times, it's time to automate, a maxim he takes from software engineering. Doing the work by hand first keeps owners close to the details, because accountability for whether the customer signs up and pays stays with them.

Watch the full video at https://www.youtube.com/watch?v=qy8Gr27yLMk.