Intel / video

Cooking with Codex — Charlie Guo & Gabriel Chua, OpenAI

Why it matters

Shows how to give coding agents checkable completion criteria and split work across subagents, hooks and automations, with real build examples.

Key takeaways · AI-distilled

In the opening demo, Codex turned a captured Slack conversation into a working macOS live translation app in four minutes and two seconds. The conversation supplied the idea, plugins supplied development practices, and a live German translation tested the result.
The presenters separate browser control for local testing from access to applications and sessions where a person is already signed in. Other demos use computer use to export dashboard data and the Chrome extension to build a feedback form.
For longer projects, such as a Python package with a Rust backend reconstructed from research papers, goals set verifiable completion criteria while subagents, visible threads, audits and progress dashboards divide the work and expose blockers.
Remote threads hand browser testing to a local machine, hooks add deterministic checks at set points in the workflow, and automations chain steps such as waiting for a server to finish provisioning or opening a separate worktree to fix a bug.