Users of OpenAI’s coding tool Codex may recognize Thibault Sottiaux as the person who lifts their token limits whenever the product hits a new milestone. Now, OpenAI is aiming to deliver that same sense of reward to a broader audience with ChatGPT Work, a platform designed for white-collar professionals to harness AI agents. I sat down with Sottiaux for our in-depth look at the product and its hurdles, but we wanted to share more of that conversation with you. Below is a lightly edited and condensed version of our chat, covering how to win over skeptics, why discovery is a core design principle, and what intelligence really costs.

**Q: Is it accurate to describe you as the product lead for Codex and Work?**

Sottiaux: I oversee all of our core products—that includes the API, agent infrastructure, enterprise solutions, and the entire ChatGPT suite, which covers ChatGPT Work, the classic ChatGPT, and everything tied to Codex.

**Q: I believe you report to Greg Brockman, correct?**

Sottiaux: Yes, Greg. I like to joke that everyone ultimately reports to Greg.

**Q: Why is ChatGPT Work such a big deal for OpenAI?**

Sottiaux: Our goal was to extend the power of coding agents to everyone. This is essentially an exercise in taking something built for technical users and packaging it in a way that feels safe, intuitive, and usable on the go—whether on mobile or the web—while making it accessible to as many people as possible. That’s why we launched it as part of the Plus plan, which costs just $20 a month. The value you get from it is pretty remarkable. So, the driving force was: “make this technology available to the widest audience possible.”

**Q: Analysts suggest there’s a strong economic incentive here—that OpenAI needs to own the direct relationship with users through applications.**

Sottiaux: The more value and utility we create for users, the more willing they are to pay for a slice of that utility. That’s always been our perspective with ChatGPT. It’s like, I’m delivering so much value to you as a user that you just think, “of course I’d pay $20 a month for this,” because the benefit you’re getting far outweighs the cost.

**Q: How significant could this technology be in shifting public perception of AI?**

Sottiaux: I see it as our responsibility to bring everyone along with this technology. The question is how to handle diffusion. Codex was built for a forgiving, technical audience, where we could roll out certain capabilities early. Now, the technology has matured to a point where it’s the right time to spread it to a much broader group and teach people that this is far more than just help with writing or personal advice. With this new iteration, ChatGPT can autonomously handle complete, complex tasks for you in a way that’s both delightful and safe. OpenAI’s mission is to bring everyone along.

Sottiaux: At its core, what we’re trying to do is build extremely capable models and then figure out the simplest, most enjoyable ways to integrate them into your life so you get massive utility. To do that, you almost have to step out of the model’s way—let it express itself and tap into all the value it can offer. That means a minimal product surface, pure simplicity, and a natural way for humans to interact. We started with text, but we’ve since launched ChatGPT Voice, which has seen significant growth. Talking to it feels incredibly natural—like having a conversation, just as we’re doing now. Over time, this technology will become even more intuitive. It adapts to humans; you don’t have to learn how to use the app in reverse.

**Q: I was reading Ethan Mollick, a Wharton professor who studies these tools. He said ChatGPT Work tries to feel like magic, while Claude Cowork puts A/B tests in front of you and forces more choices. Do you think workers are ready for that level of magic?**

Sottiaux: We definitely see that the world seems ready. That’s why we’ve seen incredible adoption—we just hit 20 million users. We managed to launch it in a way that’s simple yet powerful, simple yet uncompromising.

**Q: Do you think in terms of a minimum viable product? Do you focus on discrete problems? What are you evaluating?**

Sottiaux: It’s almost like a product of discovery. As we push the boundaries of model capabilities, we also uncover what they can do, and then we lean into those strengths and build great products around them. There’s an element of discovering capabilities, even for us, and that’s always a magical, exciting experience. For instance, GPT 5.6 was a major leap in general work—processing large volumes of documents, creating quality slides, generating reports, doing deep research—the kind of tasks professionals handle. We then lean into that, gather feedback, and keep improving. This is part of iterative deployment: learning from the community, from real-world use, and continuously refining.

**Q: How do you think that plays out? Should I be worried? Should CFOs be worried?**

Sottiaux: We’re working every day to push the frontier on efficiency. We just announced major price cuts with Luna—80% off. This is a permanent price correction, meaning current frontier capabilities become cheaper over time. That trend will continue. Our aim is to pack more utility into the same dollar amount over time. So, if you want to do more, sure, you can pay more, but in terms of what’s possible today, you should wake up six months from now and be able to do everything you’re doing now for less.

Sottiaux: It’s crucial to pick models that are safe and aligned. A huge portion of our investment goes into the safety stack, our safety approach, and publishing honest benchmarks. Our models are world-class in these areas.