Tom talks to OpenAI's Katia Gil Guzman about why “vibe coding” is the wrong model for AI development and what real coding agents mean for trust, safety and the future of software engineering.
In this bonus episode Tom Allen sits down with Katia Gil Guzman, a founding member of OpenAI’s Developer Experience team, to explore how AI coding is evolving beyond the hype. Katia explains why OpenAI actively discourages “vibe coding” for anything beyond small experiments, arguing that enterprise-ready AI tools should behave more like software engineers - producing pull requests, following project rules and aligning to established guardrails.
She also discusses how companies can improve the trustworthiness of AI-generated code; why you probably don't need multi-agent orchestration; and what the next 12-24 months hold for the AI space.