OpenAI Podcast Episode 6 - The future of coding with AI
298 snips
Sep 15, 2025 
Greg Brockman, co-founder and president of OpenAI, teams up with Codex engineering lead Thibault Sottiaux to discuss the evolution of AI coding. They cover how GPT-5 Codex agents can tackle complex coding tasks and the significance of building 'harnesses' for coding automation. The duo explores breakthroughs in code review and the potential transformation of software development. They also ponder the future of collaborative AI in coding, addressing concerns like security and whether learning to code remains relevant in this AI-driven landscape.
AI Snips
Chapters
Transcript
Episode notes
Why Coding Became A Core Focus
- Coding became a focused domain because early GPT-3 completions showed clear viability and outsized impact.
- OpenAI invested special training, data, and metrics to push coding capabilities further.
Harnesses Matter As Much As Models
- The harness (tools + loop) is as important as the model for usable coding agents.
- Integrations, execution, and tool access let text "come to life" beyond simple completions.
Optimize For Latency Or Change The Harness
- Prioritize latency as a product feature when integrating models into real-time workflows like autocomplete.
- Choose the smartest model that meets your latency budget or adapt the harness to accommodate slower, higher-quality models.