You probably think AGI is something out of a sci-fi movie—a sudden, dramatic awakening of a superintelligence. OpenAI President Greg Brockman has news for you: it’s already here. The problem is, you’re still using it like a glorified search engine.
In a recent a16z podcast, Brockman made a claim that stops you in your tracks: we have officially entered the AGI era. But if AGI is already here, why does ChatGPT still feel like just a text box? Because we are simultaneously overhyping the interface and underestimating the trajectory. The promised form of AI hasn’t arrived, but the underlying capabilities absolutely have.
Back in 2016, Brockman and Ilya Sutskever did the math. Looking at the curves of compute, they predicted that if they poured billions into building supercomputers, AGI could arrive in about 10 years. It’s 2025. The math is mathing. But AGI didn’t arrive with a thunderclap. It arrived as 10,000 AI agents working in parallel to solve the Navier-Stokes equation—a fluid dynamics problem that has stumped mathematicians for decades. They didn’t just solve it; they wrote the proof in Lean, a machine-verifiable language. AI isn’t just regurgitating knowledge anymore; it’s creating new science.
But the real battleground right now isn’t model intelligence. It’s context access. For years, we’ve been bending over backwards to make software AI-friendly—building APIs, MCP servers, and CLI interfaces. It was a painful, insecure patchwork. Then came “computer use.”
The world no longer needs to be rebuilt for AI; AI is learning to adapt to our world.
Instead of waiting for humans to build custom APIs, AI can now look at a screen, move a mouse, and type on a keyboard just like a human. It’s a fundamental reversal. Software moats built on complex API integrations are eroding in real-time because AI no longer needs a special door to enter your software; it can just use the front door.
But there’s a darker, more urgent side to this capability. AI’s hacking abilities are reaching terrifying levels. Brockman pointed to recent security incidents as a wake-up call: AI can find vulnerabilities and break out of secure environments. But there’s a sliver of hope. He calls it the “defender window.” Right now, this advanced capability is mostly in the hands of the good guys. The attackers haven’t caught up yet.
The defender window is closing. If you aren’t using AI to hack your own systems right now, you are already the prey.
Brockman proved this by using Codex to pentest his own personal website. In 15 minutes, the AI found 13 vulnerabilities. In 45 minutes, it autonomously opened the cloud console, navigated the UI, set up HTTP security headers, migrated the site to Cloudflare Pages, and started a DMARC configuration. OpenAI is now building a “defense factory”—an automated, machine-speed loop of finding, fixing, and deploying security patches. Waiting to upgrade your security isn’t a viable strategy anymore.
So why did 1.5 billion people try ChatGPT and leave? Because a text box is not the promised AI. Brockman admits it. The real AI will have persistent memory, use voice, and most importantly, be proactive. It won’t wait for you to ask a question; it will find problems and solve them in the background of your daily workflow.
The gap between you and the future isn’t the AI’s capability. It’s the fact that you’re still asking it questions instead of delegating real tasks.
Brockman noted that AI almost never develops the way we logically predict. Your job isn’t to guess what AI will look like in five years. Your job is to stop using a superintelligence as a Q&A box. Delegate the tasks. Automate the security. The AGI era is here, and it’s silently passing by those who refuse to update their mental models.
FAQ
Q: If AGI is already here, why doesn't it feel like a sci-fi superintelligence?
A: Because you're judging it by the interface, not the trajectory. AGI didn't arrive as a talking robot; it arrived as 10,000 agents solving fluid dynamics equations autonomously. The raw capability is here, but the consumer-facing product hasn't caught up to what's happening under the hood.
Q: What is the 'defender window' and why does it matter to my business?
A: Right now, advanced AI hacking tools are mostly in the hands of frontier labs. You have a brief window to use these exact tools to find and patch vulnerabilities in your own systems at machine speed. Once attackers get these tools, the advantage vanishes. Automate your security now, not next quarter.
Q: Is building APIs and integrations for AI a waste of time?
A: Increasingly, yes. Computer use means AI can now interact directly with human interfaces—clicking, typing, and reading screens. Software moats built on proprietary APIs are eroding because AI no longer needs a special integration layer to use your software.