OpenAI Launches ChatGPT Work, a Cloud Agent That Runs Your Calendar, Slack, and To-Do List

OpenAI has a new answer to the question of what ChatGPT is for. On Thursday, the company launched ChatGPT Work, an AI agent built into its chatbot that doesn't just answer questions but actually does the work: reading your Slack messages, checking your calendar, pulling commit histories from GitHub, and producing finished documents, spreadsheets, presentations, and websites without much hand-holding along the way.

The product runs on GPT-5.6, OpenAI's newest flagship model, and it's designed around a different premise than the chatbot most people know. Instead of a back-and-forth conversation, you give ChatGPT Work a goal. It breaks that goal into steps, gathers whatever context it needs from your connected apps and files, and stays on the task — sometimes for hours — until it's done.

That's a meaningful shift in ambition. OpenAI has spent two and a half years training the public to think of ChatGPT as a tool you talk to. ChatGPT Work asks people to think of it as a colleague you delegate to.

A persistent machine that follows you everywhere

The engineering choice underneath ChatGPT Work is a cloud-based virtual machine tied to the user rather than to any one device. Competing agents often need a local computer running and connected to keep working. OpenAI's version lives on its own servers, always on, accessible from phone, web, or desktop.

"What's really exciting about ChatGPT Work is that it's a virtual machine in the cloud that's always on for you, and this is available across all of our paid tiers," said Ty Geri, the OpenAI product manager who helped build it, in an interview with VentureBeat. "All Plus users are getting this. I think that's a very unique aspect of this."

Geri pointed to the mobile angle specifically as something he felt was "missing from the market." He described building and sharing a website entirely from a phone — including, notably, from the beach — as the kind of capability that wouldn't have been practical with earlier tools. Hosted sites are a recent addition to Codex, OpenAI's internal engineering agent, and ChatGPT Work extends that same feature to web and mobile.

The rollout starts with Pro, Enterprise, and Edu users, then expands to Plus and Business subscribers over the following days. Geri framed the Plus-tier availability — a $20-a-month plan, not a premium enterprise package — as central to the strategy. "It's accessible to all paid plans, including Plus users, which in my opinion is a really big feat, and really part of that OpenAI mission, which is about bringing all this power to as many people," he said.

How it actually connects to your tools

ChatGPT Work reaches into services like Gmail, Google Calendar, Slack, and GitHub through plugins built on the Model Context Protocol, or MCP, an increasingly common standard for linking AI models to outside software. Geri confirmed the plugin architecture is MCP-based and said support for connecting multiple Gmail accounts, a request he said comes up often, is on the roadmap.

The onboarding is built to get users acting immediately rather than just chatting. The system tailors its suggested tasks to a user's detected role — Geri, as a product manager, was offered prompts around evaluating systems, building research documents, and managing his schedule. A simple starting command like "catch me up on Slack" or "read today's calendar" can cascade into something more involved: Geri described a flow where the agent reviewed his calendar, flagged conflicts and meetings needing prep, then declined, accepted, or rescheduled events on his say-so.

Users can teach the agent their writing voice, organize its output by project, and pick a virtual pet to sit in the interface — a lighter touch amid the heavier automation. The product also introduces hosted, interactive websites as an output format, which Geri contrasted favorably with the constraints of a slide deck. "Now we suddenly have a collaborative interface that's actually more exciting and more accessible than a slide deck, which has all these formatting restrictions," he said.

What it looks like in practice

The clearest illustration of what ChatGPT Work is supposed to do came from Geri's own use of it while prepping the launch. He needed to schedule pre-release testing sessions, known internally as bug bashes, across many features and contributors.

"I just come to ChatGPT Work and say, 'Set up a bug bash for all the distinct features in ChatGPT Work. Add all the people that worked on that feature,' and it can check Slack, it can check GitHub, it can check Docs, and find a time that works for the four highest contributors to that feature," Geri said. "It went and scheduled 10 bug bashes, all coordinated across all those different people. That would have taken me 30 minutes at least."

He pushed back, though, on the idea that the tool is only good for scheduling and busywork. He described using it for more analytical jobs, like tracing the biggest drivers of user churn on specific features and generating fixes — work he said used to take months. "Things that we would have spent three months doing, we can now spend a week doing — and do much more, and make a much better product," he said. "Bugs that we would have found three or four weeks from now, we can now find within two days and fix for our users." He also said the agent has taken over much of the repetitive manual testing that used to eat up his time, letting him focus on deciding what to test rather than clicking through it himself.

The privacy question that comes with the access

Granting an AI agent read access to Slack messages, email, calendar invites, and code repositories is a different proposition than typing a prompt into a chat window. It means ChatGPT Work is actively reaching into systems that often contain sensitive internal communication, not just information a user chose to paste in.

Asked about this, Geri said privacy "is incredibly important, and the most important part of this is it's always in the user's control." He pointed to existing protections for enterprise accounts, including zero data retention, and noted that users can opt out of letting conversations train future models — an option he said many already use. That echoes language OpenAI used when it launched ChatGPT Enterprise in 2023, stating it does not train on business data or conversations.

Those assurances will matter more here than they did with a standard chatbot, simply because of how much more ChatGPT Work is designed to touch. Enterprise security teams evaluating the product will be looking closely at exactly what it can see and for how long.

OpenAI isn't the only one building this

ChatGPT Work drops into a market that already has two serious competitors chasing the same idea. Anthropic took Claude Cowork out of preview in April, bringing an AI agent to web and mobile for managing long-running tasks. Microsoft, working with Anthropic, made Copilot Cowork generally available worldwide in June. All three products now compete for the same customers: enterprise IT buyers and individual professionals who want an agent that finishes tasks rather than just discussing them.

The overlap in vision is notable. Each product offers a persistent cloud agent, plugin connections to workplace tools, and multi-step task execution across desktop, web, and mobile. What separates OpenAI is reach. ChatGPT has roughly 900 million weekly active users and 50 million paying subscribers, with more than 9 million paying business users and adoption at 92% of Fortune 500 companies, according to OpenAI's figures. Pricing ChatGPT Work into the $20 Plus tier rather than reserving it for premium customers is a bet that sheer distribution beats feature parity.

The IPO backdrop

None of this is happening in a vacuum. Last month OpenAI confidentially filed a draft S-1 with the SEC, setting up what could be one of the largest tech IPOs ever, with valuation estimates ranging from $730 billion to $852 billion. The company has said it's generating $2 billion in monthly revenue, with enterprise now accounting for more than 40% of that total and expected to reach parity with consumer revenue by the end of 2026. OpenAI remains deeply unprofitable, though, with internal projections pointing to roughly $14 billion in losses in 2026 alone and no expected profitability until around 2030.

Anthropic filed its own IPO paperwork on June 1 at a $965 billion valuation, meaning two of the most closely watched AI companies in the world may go public around the same time. Whether either can hold those valuations under public market scrutiny will hinge substantially on whether tools like ChatGPT Work and Claude Cowork can show enterprise customers real, measurable productivity gains, not just impressive demos.

ChatGPT Work also represents the endpoint of a longer product arc for OpenAI. It began with ChatGPT Enterprise in 2023, continued through the January 2025 launch of the Operator agent, and absorbed Operator's capabilities after its shutdown in August 2025. GPT-5.6 itself ships in three variants — Sol for heavier reasoning, Luna for speed, Terra for everyday balance — paired with the persistent cloud environment and growing MCP plugin library.

Geri was careful not to frame the product as a labor-market disruption, speaking instead from his own experience. "My job is not to schedule bug bashes and find out who contributed to a specific feature. That's a task I do in my job, but that's not my job," he said. "My job is to make an amazing product." He described the agent as "a partner" and "an extension of me, certainly not a replacement," while acknowledging the trade-off cuts both ways: "Everybody feels far more productive than before, but is also almost working harder than before, because you get to work on all the things you want to work on as opposed to the drudgery around it."

Whether that feeling scales to the hundreds of millions of people OpenAI wants using this is the real test. The company is about to find out in front of public market investors, with a valuation north of $800 billion riding on the answer.