OpenAI is trying to move agentic AI from software teams into everyday office work. In a TechCrunch interview, Thibault Sottiaux described ChatGPT Work as an effort to package the capabilities behind Codex for a much broader group of users, including people working on documents, slides, reports and research.
Sottiaux said he leads OpenAI's core products, including API, agent infrastructure, enterprise, ChatGPT, ChatGPT Work, ChatGPT classic and Codex. His comments frame ChatGPT Work as both a product launch and a test of how ready the wider market is for AI systems that do more than answer questions.
Why ChatGPT Work matters to OpenAI
According to Sottiaux, the goal is to take what coding agents can do and make that power available to people who are not technical users. Codex began with a more forgiving audience: developers who could tolerate early product edges because they understood the technology and could evaluate its output.
ChatGPT Work is aimed at a different kind of user. Sottiaux described the challenge as making the system safe, pleasant to use and available on mobile and web. That matters because white-collar work often happens across devices, and a tool that only works in one narrow setting is less likely to become part of daily professional life.
OpenAI also made ChatGPT Work part of the Plus plan, which the interview identifies as $20 a month. Sottiaux's argument is straightforward: if users receive enough utility from the product, they will be more willing to pay for it.
The broader economic question is whether OpenAI can keep the direct relationship with users as AI becomes more central to work. Sottiaux did not reject the business logic, but he put the emphasis on utility. In his view, the subscription case rests on users feeling that the value they get is larger than the price they pay.
From writing help to autonomous tasks
Sottiaux positioned ChatGPT Work as part of a shift in what people expect from ChatGPT. The familiar version of the product can help with writing, advice and conversation. The newer ambition is for ChatGPT to complete complicated tasks autonomously while remaining simple and safe enough for nontechnical people.
That shift changes the design problem. A traditional software product often exposes features through buttons, menus and settings. Sottiaux described a different product philosophy: build highly capable models, then create the smallest useful surface that lets the model's ability come through.
ChatGPT Voice is one example he pointed to. He said voice has seen a lot of growth because speaking to the system can feel natural. The larger idea is that AI products should adapt to humans, rather than making humans learn complicated application logic first.
This helps explain why ChatGPT Work is designed around a sense of direct delegation. The product is not only about showing users more controls. It is about deciding when the model should take the lead and when the interface should ask the user to make choices.
Is the market ready for more automation?
The interview also raised a key tension: how much autonomy workers are ready to give an AI system. TechCrunch referenced Ethan Mollick, a Wharton professor who studies these tools, and compared ChatGPT Work with Claude Cowork, noting that one approach tries to feel more magical while the other puts A/B tests and more choices in front of the user.
Sottiaux's answer was that OpenAI sees signs of readiness. He said the company had announced 20 million users, and described the launch as simple but powerful. For OpenAI, that adoption is evidence that many people are willing to try a product that takes on more of the work itself.
Still, the product challenge is unusually broad. ChatGPT Work is not a single-purpose workflow tool. Sottiaux described it as partly a product of discovery: as models improve, OpenAI learns what they can do well, then builds product experiences around those strengths.
He gave GPT 5.6 as an example of a step forward for general work. The capabilities he named include processing a large amount of documents, generating quality slides, generating quality reports and doing deep research. Those are the kinds of tasks professionals already spend time on, which makes them natural targets for an AI work platform.
OpenAI's process, as Sottiaux described it, is iterative. The company releases capabilities, watches real use, gathers feedback from the community and continues improving the product. That approach treats user behavior as part of the product development cycle, not just as a result measured after launch.
The cost and safety questions
Cost remains one of the biggest questions around agentic AI. The interview noted the gap between what a Plus user pays and the amount of tokens that user may consume. For a product designed to complete larger tasks, that gap matters because more useful work can mean more computation.
Sottiaux said OpenAI is working every day on efficiency. He pointed to Luna and said the company announced major price cuts of 80% off, describing that as a permanent price correction. His broader claim is that frontier capabilities should become cheaper over time.
The practical promise is that the same dollar amount should include more utility as efficiency improves. Sottiaux said that if a user wakes up six months from now, they should be able to do the same work with less spend. He also left room for users who want to do more to pay more.
Safety is the other central concern, especially when agents connect to personal or work data. TechCrunch asked about users who feel stressed about giving the system access to e-mail or iMessages. Sottiaux answered by emphasizing model safety, alignment, OpenAI's safety stack and honest benchmarks.
For ChatGPT Work, that trust question may define adoption as much as capability. A system that can process documents, create reports and act across connected services has to be useful, but it also has to feel controlled enough for people to let it near sensitive work. OpenAI's bet is that the combination of broader access, simpler design and improving efficiency will make that tradeoff acceptable to a much wider audience.