OpenAI Codex's New Coding Workflow

AI Engineer · youtube · 2026-07-10

This OpenAI presentation review covers the evolution of AI coding workflows: from Dev Day 2024, where humans had to babysit models running code, to a year later where models can complete full demos of "testing their own work," highlighting Codex's rapid progress.

The talk also details Codex's tech stack and product design: it's built on the same responses API, open-source harness, and the AGENTS.md file format distributed to developers. It shared new performance metrics—frontier-level intelligence is priced at about $1 per million input tokens and $6 per million output tokens, with generation speeds reaching 750 tokens per second, fast enough to produce a real pull request in roughly 10 seconds. Finally, it notes that the future bottleneck for AI engineering will shift from "model capability" to "human attention," demonstrating a workflow where agents autonomously resolve open-source issues.

Original post →

More from coding & agent

coding & agent channel →