Summary
In this a16z podcast episode, OpenAI co-founder Greg Brockman declares that we have entered the AGI era, pointing specifically to Astra — OpenAI’s GPT-6 — running coherently for 24 hours on complex tasks as the benchmark that convinced him the label is warranted. The conversation covers Brockman’s original timeline predictions from 2016–2017, the current compute supply crunch, and what “pacing the frontier” means for OpenAI going forward.
A major theme is the maturation of AI safety: Brockman describes the shift from surface-level content filtering and RLHF guardrails to deep alignment work that he now views as a potential bottleneck to capability progress rather than a downstream concern. He also addresses the challenge of distributing powerful AI broadly when compute demand consistently outpaces supply — calling it an underestimated problem central to OpenAI’s mission.
Brockman demos computer use concretely, describing how an Astra agent autonomously fixed DMARC email security vulnerabilities on his personal website in 45 minutes — navigating Cloudflare’s control panel, configuring headers, initiating a 48-hour demarking process, and scheduling an automated follow-up check. The episode also covers the GPT-6 naming decision (waiting for a genuine discontinuous capability step), OpenAI’s infrastructure investment priorities, and Brockman’s early roles at Stripe and OpenAI. Essential viewing for anyone tracking the AGI milestone debate and OpenAI’s strategic direction.
📺 Source: a16z · Published September 14, 2026
🏷️ Format: Interview
