Sep 14, 2026 · 51 min

OpenAI frames computer-using agents as the next AGI threshold

Greg Brockman on Why OpenAI Says We’re Entering the AGI Era

The discussion links AGI’s definition to sustained computer-based work while confronting the infrastructure, security, and governance needed to deploy it safely.

3 key takeaways

1AGI increasingly means systems that can sustain coherent work through computer interfaces, not merely produce impressive outputs.
2Advanced agents could accelerate science and cybersecurity, but restricted access and coordination may determine whether defenders keep an advantage.
3OpenAI’s next phase pairs broader deployment with heavier investment in infrastructure, safety, security, and alignment standards.

Don't miss

Brockman describes using 10,000 agents on the Navier–Stokes problem, then connects that capability to machine-speed cybersecurity.

The brief

Greg Brockman argues that AGI is arriving as a spectrum: the decisive shift is toward systems that can sustain coherent work through computer interfaces.

The conversation pairs optimism about scientific discovery and automated cybersecurity with a warning that frontier capabilities may spread faster than infrastructure and safeguards.

Brockman describes a defender’s window in which advanced AI can find, triage, and fix vulnerabilities before comparable tools become broadly available to attackers.

A striking example is the use of 10,000 agents on the Navier–Stokes problem, suggesting how machine-speed collaboration could change scientific research.

The episode closes on deployment discipline: restricted access, frontier-lab coordination, data-center investment, and stronger safety, security, and alignment standards.

Featuring