AI coding agents are transforming the software development process, shifting the focus from code generation to aspects like planning, review, deployment, and coordination. As these agents become more capable, a new class of systems is emerging that functions in controlled environments, manages extended timeframes, and integrates with tools such as IDEs, version control systems, and issue trackers.
OpenAI leads the advancement of AI research and products. In 2025, it launched Codex, an agentic coding system that operates securely within sandbox environments, collaborating across the modern software development landscape.
Thibault Sottiaux, Codex engineering lead, and Ed Bayes, Codex product designer, join Kevin Ball in this episode to talk about Codex’s architecture, the evolution of models and harnesses, the future of multi-agent systems, Codex’s open-source CLI, model specialization, considerations of latency and performance, and more.
Kevin Ball, also known as KBall, is the vice president of engineering at Mento and an independent coach for engineers and engineering leaders. He co-founded two companies and served as CTO, started the San Diego JavaScript meetup, and leads the AI inaction discussion at Latent Space.
Please see the transcript of this episode by clicking here.