OpenAI and Codex with Thibault Sottiaux and Ed Bayes

The podcast explores the co-evolution of AI models and infrastructure in software development, focusing on OpenAI's Codex, an agentic coding system. It features Thibault Sottiaux, the Codex Engineering Lead, and Ed Bayes, the Codex Product Designer. A key discussion point is the balance between model capabilities and the surrounding harness, emphasizing the importance of sandboxing for safety. The speakers detail how Codex is used internally at OpenAI, including code review, Slack integration, and empowering non-technical staff. They also touch on the shift in software development bottlenecks from code generation to code review and planning, and the potential for multi-agent collaboration.

Outlines

Part 1: Introduction, Safety, and Sandboxing

Part 2: Internal Use and Workflow Integration

Part 3: Model Specialization and Architecture

Part 4: Agent Evolution and Complex Task Management

Part 5: Accessibility and Future Outlook

Sign in to continue reading, translating and more.

Continue

Software Engineering Daily

Part 1: Introduction, Safety, and Sandboxing

AI Coding Agents: Reshaping Software Development with Constrained Environments

Codex's Co-evolution of Models and Harnesses for Safe AI Coding

Balancing Safety and User Experience in Codex's Sandboxing Model

Part 2: Internal Use and Workflow Integration

Internal Use of Codex: Code Review, UX Copy, and Cross-Disciplinary Empowerment

Adapting Software Development Lifecycle with Codex: Bottleneck Shifts and Team Evolution

Part 3: Model Specialization and Architecture

Multi-Model Future: Collaboration Between Agents for Simplified User Experience

GPT-5 Codex: Model Specialization and Harness Integration for Enhanced Performance

Open Source Codex CLI: System Prompts, Optimal Tools, and Model Performance

Locality and Agent Architecture: Model Inference, Tool Calls, and Control Loops

Part 4: Agent Evolution and Complex Task Management

Product Evolution: Control, Steering, and Supervision of Complex Agents

Beyond Code Generation: Agents for Planning, Ideation, and User Feedback Understanding

Transient Documentation and General-Purpose Interfaces for Creative Workflows

Codex SDK, Hooks, and Multi-Agent Networks for Complex Work

Managing Agent Complexity: Tracking Actions, Permissions, and Artifacts

Part 5: Accessibility and Future Outlook

Non-Technical Users: General Agents and On-Ramps for Learning to Code

Mindset and Curiosity: Embracing Change and Rewarding Creativity

OpenAI and Codex with Thibault Sottiaux and Ed Bayes

Software Engineering Daily

Part 1: Introduction, Safety, and Sandboxing

00:00AI Coding Agents: Reshaping Software Development with Constrained Environments

AI Coding Agents: Reshaping Software Development with Constrained Environments

01:47Codex's Co-evolution of Models and Harnesses for Safe AI Coding

Codex's Co-evolution of Models and Harnesses for Safe AI Coding

06:20Balancing Safety and User Experience in Codex's Sandboxing Model

Balancing Safety and User Experience in Codex's Sandboxing Model

Part 2: Internal Use and Workflow Integration

11:55Internal Use of Codex: Code Review, UX Copy, and Cross-Disciplinary Empowerment

Internal Use of Codex: Code Review, UX Copy, and Cross-Disciplinary Empowerment

14:16Adapting Software Development Lifecycle with Codex: Bottleneck Shifts and Team Evolution

Adapting Software Development Lifecycle with Codex: Bottleneck Shifts and Team Evolution

Part 3: Model Specialization and Architecture

19:31Multi-Model Future: Collaboration Between Agents for Simplified User Experience

Multi-Model Future: Collaboration Between Agents for Simplified User Experience

21:43GPT-5 Codex: Model Specialization and Harness Integration for Enhanced Performance

GPT-5 Codex: Model Specialization and Harness Integration for Enhanced Performance

25:31Open Source Codex CLI: System Prompts, Optimal Tools, and Model Performance

Open Source Codex CLI: System Prompts, Optimal Tools, and Model Performance

29:47Locality and Agent Architecture: Model Inference, Tool Calls, and Control Loops

Locality and Agent Architecture: Model Inference, Tool Calls, and Control Loops

Part 4: Agent Evolution and Complex Task Management

32:57Product Evolution: Control, Steering, and Supervision of Complex Agents

Product Evolution: Control, Steering, and Supervision of Complex Agents

35:41Beyond Code Generation: Agents for Planning, Ideation, and User Feedback Understanding

Beyond Code Generation: Agents for Planning, Ideation, and User Feedback Understanding

38:16Transient Documentation and General-Purpose Interfaces for Creative Workflows

Transient Documentation and General-Purpose Interfaces for Creative Workflows

42:08Codex SDK, Hooks, and Multi-Agent Networks for Complex Work

Codex SDK, Hooks, and Multi-Agent Networks for Complex Work

44:25Managing Agent Complexity: Tracking Actions, Permissions, and Artifacts

Managing Agent Complexity: Tracking Actions, Permissions, and Artifacts

Part 5: Accessibility and Future Outlook

46:18Non-Technical Users: General Agents and On-Ramps for Learning to Code

Non-Technical Users: General Agents and On-Ramps for Learning to Code

49:22Mindset and Curiosity: Embracing Change and Rewarding Creativity

Mindset and Curiosity: Embracing Change and Rewarding Creativity