Summary
Nate B Jones travels to OpenAI’s offices to interview Andrew and Ash, two team members working on Codex and the company’s broader agent products, about what it actually looks like to work alongside AI agents day to day. The conversation covers surprising use cases that have emerged since GPT Astra’s public release, including viral 3D Blender simulations and a family member turning a home renovation blueprint PDF into a fully walkable 3D model.
The interview digs into how internal product development at OpenAI has shifted toward faster, more iterative cycles resembling consumer software practices, and how knowledge workers without engineering backgrounds are now building their own internal tools by describing problems directly to models instead of writing specs for engineers. Andrew and Ash also discuss where they see the biggest predictable improvements coming over the next six months to a year, particularly around models’ ability to infer true intent and pull in accurate context and sources.
For viewers trying to understand where agentic AI tools like Codex are headed and what skills will matter most as these systems mature, the discussion offers a rare first-person view from inside OpenAI of both the current friction points — like losing track of what agents are doing across multiple windows — and the long-term trends the team is betting on.
📺 Source: AI News & Strategy Daily | Nate B Jones · Published September 22, 2026
🏷️ Format: Interview
