In a TechCrunch interview, OpenAI‘s head of product, Thibault Sottiaux, discusses ChatGPT Work, the company’s platform for giving white-collar workers access to AI agents. Sottiaux leads core products including the API, agent infrastructure, enterprise, all of ChatGPT, and Codex, and he reports to Greg Brockman. He frames ChatGPT Work as an exercise in taking coding-agent capabilities and packaging them so they are safe, delightful, usable on mobile and web, and available to the broadest possible population. That is why it launched as part of the $20-per-month Plus plan.

Sottiaux describes the economic motivation in terms of value: the more utility OpenAI generates, the more willing users are to pay for part of it. He also presents ChatGPT Work as a diffusion moment. Codex was built for a forgiving technical audience, but OpenAI now sees the technology as mature enough to bring to everyone else, with ChatGPT able to complete complex tasks autonomously. The mission, he says, is to bring everyone along.

On product design, Sottiaux says the key is to avoid explaining the model’s way. The namered approach: a minimal product surface, natural interaction, and models that adapt to humans rather than the other way under. He notes that ChatGPT Voice has seen a lot of growth; discovery is a design philosophy. OpenAI pushes frontier of capability, discovers what models can do, and then, then lean into those strengths. Examples include GPT 5.6 steps up in document processing, slides generation, making reports, and deep research. This is iterative deployment, using feedback and real use to improve.

When asked whether workers are ready for a ‘magical’ agent experience, he points to adoption and says ‘the world seems to be ready.’ It says ChatGPT has over 20 million users. On token costs and the gap between paid plan and heavy usage, Sottiaux says OpenAI works daily on efficiency and has announced major price cuts with Luna, including an 80% permanent correction price cut. The goal is to include more utility in the same dollar amount over time, so users should be able to do the same work with less spend six months. Nominated about agents accessing email and iMessages, he emphasizes safe, aligned models, the safety stack approach, published honest benchmarks, and his models being world-class in this area.