a16z Podcast: Greg Brockman on Why the AGI Era Is Already Here

0:00So, power, compute, and safety are colliding in real time. And Greg Brockman from OpenAI has made a pretty bold claim. We are currently in the AGI era. He says Astra has reached a level where it's reasonable to call it AGI. That's huge. It really is. And what's fascinating is that Brockman says the breakthrough isn't just about raw intelligence. It's endurance. Astra has apparently run coherently for 24 hours on difficult tasks, which signals a genuine step change in capability. Right. But there's a catch. He warns that access will be constrained, because there won't be enough compute to serve everyone who wants these models. That scarcity makes discipline more important than ever. Exactly. Brockman keeps returning to this idea of blocking and tackling. He's saying the real challenge now is to keep raising standards for safety, security,

1:03and alignment. As he puts it, we have to pace the frontier, because those protections must become the main bottleneck as models get stronger. That's interesting, because he connects today's moment to OpenAI's earlier safety work, right? Things like reinforcement learning from human preferences, debate, and iterative amplification. Things were foundational ideas already in motion years ago. Yes. Any argues the company now needs coordination across frontier labs, because no single company can manage the implications alone. Building on that point, cybersecurity comes up as one of the clearest near-term risks. Brockman describes the hugging face incident as a watershed moment. What does he say about that specifically? He says it showed why defenders need to use frontier tools before attackers do. OpenAI has already used its models internally to find vulnerabilities and build what he calls defense factories for machine speed remediation.

2:08That makes sense. But one of Brockman's biggest themes is actually computer use. He calls it the headline of recent progress, because it lets models interact with the world like people do, through screens, keyboards, and mice. Instead of clunky APIs, the machine is here to help and empower us. And that shift could reshape work itself. He imagines a future where repetitive menu clicking and spreadsheet labor disappear, freeing people for more human work, like relationships, judgment, and accountability. Though he admits the transition won't be painless, he sees it as a wonderful renaissance for entrepreneurship. What about OpenAI's internal focus? I heard they had to make some tough calls. They did. Brockman says the company had to cut side projects, including the painful cancellation of Sora to concentrate on deployment, productization, and one unified stack.

3:09He borrows from a management classic, saying, the score takes care of itself. To explain why execution matters more than slogans. So the bottom line is that the AGI era isn't some distant future. It's here now. Precisely. The task in Brockman's words is to build capability, distribute benefits, and make sure the systems that change the world are safe enough to trust.