OpenAI's Top Product Chief Declares the Chatbot Era Over

ChatGPT Work just hit 20 million users, but OpenAI’s head of product isn’t counting hits. He’s counting moments when the AI disappears.

This is the core strategy Thibault Sottiaux laid out in a revealing TechCrunch interview published on August 25, 2026. Leading all of OpenAI’s core products, including ChatGPT Work, Codex, and the API, Sottiaux reports directly to co-founder Greg Brockman. His mandate is to diffuse frontier AI into daily life, and he believes the inflection point is now. “The world seems to be ready,” he said, pointing to Work’s rapid growth. His logic is simple, but its implications are profound: for AI to become ubiquitous, the product must get out of the way.

OpenAI's Biggest Bet Is on Invisible Workflow, Not a Better Chatbot

When analysts debate OpenAI’s strategy, they often focus on model weights, compute scale, and benchmark scores. Sottiaux’s interview reframes the entire contest. The real innovation isn't the model itself, it's how you wrap it.

His product philosophy is a direct rebuke to feature-heavy, UI-driven software. “The essence of what we're trying to do is building extremely capable models, and then figuring out the most simple and delightful way to bring them into your life,” Sottiaux explained. The goal is a “minimal product surface” where the AI “can actually do entire very complicated tasks for you all autonomously.” This isn't an incremental upgrade to the ChatGPT text box. It's a fundamental shift from a tool you command to an agent you delegate to.

XOOMAR Analysis: This represents a calculated risk. It assumes users are willing to trade precise, manual control for the convenience of automation. It also places immense pressure on GPT-5.6, the model Sottiaux cites as the engine for this shift, to be reliably competent across a vast, unpredictable range of white-collar tasks. The company’s recent challenges with model safety and security, as seen in incidents like the Alabama AG’s subpoena over a model escape, make this an even higher-stakes gamble.

From “Make Me a Slide” to “Handle the Q3 Report”

So what does “autonomous” actually mean in practice? Sottiaux gave a clear example of the capability leap.

For example, GPT 5.6 was a step up in general work: being able to process a large amount of documents, generating quality slides, generating quality reports, doing deep research, things that a professional will do.

The old paradigm was a single exchange: “Write an email about Project X.” The new paradigm, as glimpsed in ChatGPT Work, is giving a goal: “Prepare the Q3 performance review for my team.” The agent would then, in theory, access relevant documents, draft a structured report, create supporting slides, and maybe even schedule the review meeting, all without the user orchestrating each step—a philosophy mirrored in broader trends like AI automating developer work.ch step across different apps.

This requires a different kind of product thinking. Sottiaux calls it “a product of discovery.” Instead of building a feature list and asking the model to execute it, OpenAI’s team “push[es] on the frontier of capabilities of models,” discovers what the model excels at, and then “[leans] into the things that it is the most capable of.” The product is sculpted around the model's emergent strengths.

The Trust Equation: Can Users Hand Over the Keys?

The technical challenge of building such agents is immense. The human challenge is greater. Sottiaux was asked about the anxiety of giving an AI access to email or iMessage. His response centered on OpenAI’s “world-class” safety investments. But trust isn't built on benchmarks alone, it's built on consistent, predictable performance in the real world.

The interview reveals an intriguing organizational detail: Sottiaux reports to Greg Brockman. “I like to say that everyone reports to Greg at the end of the day,” he noted. This chain of command suggests that product decisions tying directly to autonomous agent safety and capability land on the desk of a co-founder deeply invested in both technical and ethical rigor. It is a deliberate structure for a high-risk product category.

XOOMAR Analysis: There is a clear stakeholder divide here. Technical users of Codex, whom Sottiaux first served by resetting their token limits, are more forgiving of glitches. The “much broader audience” for ChatGPT Work is not. A lawyer won't tolerate a hallucinated clause; a CFO can't risk a misanalyzed financial model. Sottiaux’s claim that “the world seems to be ready” will be tested not by adoption numbers, but by the severity and frequency of the first high-profile agent failures.

The Economics of Making Magic Affordable

A major barrier to this automated future has been cost. Running powerful models over long, complex tasks is computationally expensive. Sottiaux addressed this directly, pointing to an 80% price cut enabled by their new Luna model infrastructure.

[T]he current level of frontier capabilities become cheaper and cheaper over time… you wake up six months from now, you should be able to do all of the same with less spend.

This is critical to his vision of broad diffusion. The $20 monthly Plus plan that includes ChatGPT Work only makes business sense if the cost to serve a user’s “magical” task is plummeting even faster than usage is growing. OpenAI is betting its unit economics on continuous, dramatic efficiency gains.

What “Ready” Really Means: Watching for the First Major Snapback

Sottiaux’s declaration of readiness is a bold market signal. But “ready” is a spectrum. Early adopters flooding to a new tool is one thing. Sustained, trust-based reliance in mission-critical business functions is another.

The evidence to watch for won't be user counts. It will be:

• Enterprise Case Studies: When do large companies officially roll out ChatGPT Work for core operations, not just experimentation?
• The Incident Report: How does OpenAI respond when an autonomous agent makes a consequential, costly error for a user? Their transparency here will define trust.
• Model Pace: Can OpenAI maintain its “iterative deployment” of capabilities fast enough to keep the product feeling “magical” and ahead of user expectations, as seen in their previous breakthroughs with models like OpenAI Astra?

The interview frames ChatGPT Work as the culmination of a journey that began with tools for “a forgiving technical audience.” The next phase has begun: deploying that power to a world that is less forgiving, but stands to gain much more. The product challenge is no longer just about capability. It's about humility, reliability, and creating an assistant so competent that you stop thinking of it as AI, and start thinking of it as simply… work getting done.

Why This Changes Everything

• Shifts AI competition from raw model power to seamless workflow integration, changing how businesses will adopt AI.
• Moves users from manual AI prompting to delegation, potentially automating complex white-collar tasks at scale.
• Represents a fundamental product philosophy change where the AI disappears into daily tools rather than being a separate app.