OpenAI on Thursday launched ChatGPT Work: an AI agent embedded inside its flagship chatbot that can execute complex, multi-step tasks across workplace tools like email, Slack, and calendars. It is powered by OpenAI's latest flagship model, GPT-5.6, and the point is not just generating text. The product can gather context from connected apps, files, and workflows, then produce finished documents, spreadsheets, presentations, reports, and websites, staying with complex projects for hours.

The architecture is the real differentiator. ChatGPT Work is powered by a persistent, cloud-based virtual machine that runs on OpenAI's servers and stays available across devices. In OpenAI product manager Ty Geri's words, it is “a virtual machine in the cloud that's always on for you” and it is “available across all of our paid tiers,” including Plus. OpenAI is also pushing mobile as a core surface area: Geri described creating a website on a phone and sharing it with collaborators as “missing from the market,” and specifically called out that sites are “new in general to Codex,” then now launching in web and mobile as well.

Under the hood, the integration story matters as much as the headline promise. ChatGPT Work relies on MCP-based plugins to connect to external services like Gmail, Google Calendar, Slack, and GitHub. When asked whether the plugin architecture is based on the Model Context Protocol standard, Geri confirmed: “These are all based on MCP.” He also said connecting multiple Gmail accounts, a frequent user request, is “definitely on the roadmap.” Translation for executives: this is less a standalone chatbot feature and more an agent platform wired into the systems teams actually use, where permissioning, auditability, and data boundaries become the entire business.

OpenAI is positioning this as a shift in what ChatGPT is. The launch is described as the clearest attempt yet to reposition ChatGPT from a question-and-answer tool into an autonomous work platform. That is not just product strategy. It is competitive strategy. Agents that can act in your tools are where the value compounds, because work does not end at the answer; it ends when the calendar invite goes out, the report is drafted, and the code or docs change is ready for review.

The rollout also telegraphs where OpenAI wants early adoption and where it expects enterprise scrutiny. ChatGPT Work will roll out beginning with Pro, Enterprise, and Edu users, and expand to Plus and Business users over the next few days. Geri emphasized that Plus access is central, not a footnote. “It's accessible to all paid plans, including Plus users,” he said, framing it as a big feat and “part of that OpenAI mission… about bringing all this power to as many people.” For decision-makers, that means this is not just a high-end pilot for a few risk-tolerant teams. It is a broad distribution play that will likely accelerate user expectations and internal demand inside organizations that already bought into the ChatGPT ecosystem.

What does “agentic” look like in practice? Geri gave examples that sound like productivity theater until you map them to how teams actually work. Ahead of the product's launch, he needed to organize pre-release testing sessions, internally called “bug bashes,” across dozens of features and team members. He said he could tell ChatGPT Work: set up a bug bash for all the distinct features, add the people that worked on each feature, and then have the system check Slack, GitHub, and Docs to find time that works for the four highest contributors. He said it scheduled 10 bug bashes, coordinated across those different people, in a way that “would have taken me 30 minutes at least.”

He also pushed back on the idea that this is only rote admin. He described analytically complex tasks like identifying the biggest causes of user churn for specific product features and generating product solutions that previously took months, now taking “a week doing - and do much more.” He added that bugs that would be found three or four weeks from now can be identified within two days, and that the testing loop can move from clicking the same thing repeatedly to defining what to test, then having ChatGPT Work or Codex go test it, deliver a bug report, and hand back something teams can fix.

If there is one place boards and security teams will focus immediately, it is privacy. Because ChatGPT Work pulls sensitive information from workplace tools like Slack, Google Drive, and email, Geri said privacy “is incredibly important,” and the key point is user control. He pointed to OpenAI's existing enterprise security infrastructure, saying enterprise accounts have ZDR, and users can always opt out of letting their conversations help improve future models, which “many users do.” This aligns with OpenAI's assurances when it first launched ChatGPT Enterprise in August 2023, where it wrote it does “not train on your business data or conversations.”

This matters even more because OpenAI is making this push at a moment of potentially huge capital-market significance. Last month, OpenAI confidentially submitted a draft S-1 registration statement to the SEC, starting what could become one of the largest technology IPOs in history. Reported valuations were clustering between $730 billion and $852 billion, with annualized revenue that has blown past $25 billion. In that context, ChatGPT Work is not just a feature. It is part of how OpenAI reframes ChatGPT into a platform category, where usage spreads across workflows, and where enterprise readiness becomes a competitive moat.

For executives watching from other AI labs, software companies, and enterprise vendors, the strategic stake is clear: the product is designed to take stated outcomes, break them into smaller steps, and complete them independently for hours, then connect those actions to real workplace systems via MCP plugins. That is exactly the kind of capability that can rewrite internal work processes fast, and also exactly the kind of capability that forces new decisions about permissions, monitoring, and what “in control” really means in production.