OpenAI links ChatGPT Work’s 20 million users to agent adoption
OpenAI’s ChatGPT Work has reached 20 million users, product leader Thibault Sottiaux said in an interview, as the company expands agent-style capabilities beyond software engineering. Sottiaux leads OpenAI’s core products, including its API, agent infrastructure, enterprise offerings, ChatGPT and Codex.
ChatGPT Work is OpenAI’s attempt to package the power of coding agents for white-collar workers in a form designed for mobile and web use. The service launched within the $20-per-month Plus plan, with the stated aim of making the technology available to a broader population rather than only technical users.
From Codex to broader workplace tasks
Sottiaux described Codex as an early environment for a more forgiving technical audience. OpenAI now considers the technology mature enough to diffuse more widely, presenting it as more than assistance with writing or personal advice. The company’s objective is for ChatGPT to undertake complex tasks autonomously in a way that users find safe and straightforward.
That direction aligns with OpenAI’s personal agent strategy as OpenAI pushes ChatGPT toward a broader personal-agent role rather than a single-purpose conversational interface. Sottiaux said the product philosophy is to build highly capable models, then expose them through the simplest possible experience instead of requiring users to learn a complicated application workflow.
Voice is one example of this approach. OpenAI has launched ChatGPT Voice, which Sottiaux said has experienced significant growth, and he argued that interaction will become more natural over time as the technology adapts to people rather than asking people to adapt to software.
Capability discovery shapes the product
OpenAI does not present product development as a fixed list of predefined use cases. Sottiaux called it a process of discovery: as models advance, the company learns what they can do well, builds products around those strengths, collects feedback from real-world use and continues improving the systems through iterative deployment.
He cited GPT-5.6 as a step forward for general work. Its highlighted capabilities include processing large volumes of documents, producing slides and reports, and carrying out deep research—tasks associated with professional knowledge work.
Cost and access remain operational questions
For businesses weighing broader deployment, model economics and data access remain material considerations. Sottiaux said OpenAI is working to improve efficiency and announced a permanent 80% price cut for Luna. His stated goal is that the same level of capability should cost less over time, while customers that want greater usage can pay for more utility.
On concerns about connecting email and iMessages, Sottiaux pointed to model safety and alignment, investment in a safety stack, and the publication of benchmarks. The practical business implication is to evaluate agent workflows alongside their promised productivity: define which services can be connected, set spending and usage controls, and retain appropriate review for consequential work.