TechCrunch published an interview with Thibault Sottiaux, who leads product development at OpenAI. The main topic was the new ChatGPT Work platform, an attempt to bring agents from the programming environment into everyday office work. Reporter Tim Fernholz asked why the company is opening this technology to the general public. He was also interested in how Sottiaux envisions such tools being controlled and how concerns about the security of users’ own data should be addressed.
The Man Who Resets the Limits
Users of the Codex developer tool know Sottiaux as the person who increases their token limits whenever the product reaches another growth milestone. Formally, he is a member of OpenAI’s technical team, but in practice he oversees most of the services that customers interact with. He includes among these the developer interface, agent infrastructure, enterprise products, both the classic and new work-focused ChatGPT, as well as the Codex tool.
When asked whether he really reports to Greg Brockman, he jokingly replied that, ultimately, everyone at OpenAI reports to Brockman.
Why OpenAI Is Pushing ChatGPT Work
According to Sottiaux, the goal was to transform a tool originally created for a more technically proficient audience into something that anyone could use on a mobile device or in a browser. That is precisely why the company included the work version in the Plus plan for twenty dollars a month. The manager considers the value users receive at this price to be incomparably greater.
When Fernholz asked him about the economic motivation—OpenAI’s need to keep users in its own app—Sottiaux reframed it in terms of customer value. The more value the tool delivers, the more willing customers will be to pay for it and the more natural this twenty-dollar payment will seem to them.
He also sees the benefit of the ChatGPT Work platform in how the public thinks about artificial intelligence. Codex was developed for an audience willing to forgive many mistakes. Now, he says, the technology is mature enough for a much broader group of people to try it and discover that it is not merely a writing assistant or a source of life advice. Artificial intelligence can now perform complex tasks on a person’s behalf as well.
A Model That Should Have Nothing Standing in Its Way
The section about controls was particularly interesting. Sottiaux argues that designers should get out of the model’s way and allow it to express itself rather than placing buttons and menus in its path. He wants a minimal interface and a high degree of intuitiveness. He points to ChatGPT’s voice controls, which have grown rapidly since launch because speaking is more natural for people than learning how to operate an app.
Reporter Fernholz responded by citing a remark from Wharton’s Ethan Mollick. Mollick said that ChatGPT Work tries to feel like magic, while the competing Claude Cowork offers users options and lets them decide. Sottiaux replied that the world is clearly ready for such an approach, as demonstrated by the recent milestone of twenty million users.
He describes product development as a process of discovery. As model capabilities advance, the development team itself discovers new possibilities and then focuses on them. As an example, he mentioned GPT-5.6, which he says has made significant progress in everyday professional work, such as processing larger volumes of documents, preparing high-quality presentations and reports, and conducting more in-depth research.
Tokens, Pricing, and Email Access
Asked whether Plus subscribers should feel guilty about the number of tokens they consume and whether corporate chief financial officers should be concerned, Sottiaux pointed to an eighty-percent price reduction for the Luna model. He described it as a permanent adjustment and said that today’s level of capabilities will gradually become less expensive. Six months from now, users should be able to accomplish the same things at a lower cost.
The final question concerned fears about granting such an agent access to users’ own email, for which OpenAI has just released an integration. Sottiaux replied that users need to choose models that are safe and do what they ask of them. According to him, the company invests heavily in safety and regularly publishes the relevant measurements.
Source: techcrunch.com
