OpenAI launches ChatGPT Work to run tasks across Slack, Gmail, calendars

OpenAI on Thursday launched ChatGPT Work, an AI agent embedded inside its flagship chatbot and powered by GPT-5.6. The whole point is to move beyond “here’s an answer” and into “here’s the finished work,” with the agent able to break down a stated outcome into smaller steps, stay with complex projects for hours, and produce deliverables across connected tools like email, calendars, code repositories, and messaging apps.

This is not a subtle product tweak. OpenAI is effectively trying to turn ChatGPT into an always-on workplace operator, and OpenAI product manager Ty Geri framed it that way in a demonstration and conversation with VentureBeat on Friday, saying the mission is to democratize the kind of agentic capabilities OpenAI’s internal tool Codex has already shown. Geri also highlighted a structural differentiator: ChatGPT Work runs on a persistent cloud-based virtual machine on OpenAI’s servers, “always available” regardless of which device a user is on, and it is included across paid tiers, with all Plus users getting it.

Why now? The timing has real financial gravity for OpenAI, even if today’s news sounds like workplace software. Last month, OpenAI confidentially submitted a draft S-1 registration statement to the SEC, initiating what could become one of the largest technology IPOs in history, with reported valuations clustering between $730 billion and $852 billion and annualized revenue that has blown past $25 billion. When an IPO story is in motion, product positioning becomes more than product positioning. It becomes a narrative about where the money will come from after the chatbot era.

ChatGPT Work’s architecture is designed to make that narrative believable. Geri described the agent as a “virtual machine in the cloud that’s always on,” and emphasized that this availability covers paid tiers, including Plus. He also called out mobile-first capability as a gap in the market, pointing to the ability to create a website from a phone and share it with collaborators. He said Sites launched in Codex about a week and a half ago, but now OpenAI is launching them not just in web, but also in web and mobile through ChatGPT Work, with the punchy example that you can create a site “on your phone at the beach” and share it with friends.

Under the hood, ChatGPT Work relies on MCP-based plugins to connect to external services like Gmail, Google Calendar, Slack, and GitHub. And yes, the plugin architecture is based on the Model Context Protocol standard, Geri confirmed when asked. He also noted that connecting multiple Gmail accounts, a frequent user request, is “definitely on the roadmap.” The execution layer is what matters here: instead of asking ChatGPT to draft a plan and then manually doing the work, the system is designed to act. It offers a personalized onboarding flow that surfaces suggested use cases depending on the user’s role, and it can start with practical prompts like “catch me up on Slack or Teams or read today’s calendar.”

Geri’s demo leaned into what the product calls out as day-to-day usefulness: the agent can review calendars, detect scheduling conflicts, flag meetings that need preparation, and then, on the user’s instruction, decline, accept, or reschedule events directly. It also lets users customize the agent by teaching it their writing style, organizing outputs into projects, and even, in a lighter touch, choosing a virtual pet that accompanies them in the interface. Meanwhile, the interface adds a hosted website feature so users can build and share interactive sites through ChatGPT Work, aiming to make collaborative outputs more accessible than a static slide deck because slide decks come with formatting restrictions.

If you want to understand the “agentic productivity” claim beyond marketing, Geri’s own workflow is the clearest proof point in the source. In the run-up to launch, he organized pre-release testing sessions internally called “bug bashes” across dozens of features and team members. He said he asked ChatGPT Work to set up a bug bash for all the distinct features, add the people who worked on each feature, and then coordinate time selection based on input from Slack, GitHub, and Docs. He said it scheduled 10 bug bashes across all those different contributors in coordination that would have taken him at least 30 minutes. That may sound like time savings until you remember the hidden cost: coordinating humans across tools is usually the part that breaks teams.

He also resisted the idea that ChatGPT Work is just admin work. He described using it for analytically complex tasks like identifying the biggest causes of user churn for specific product features and generating product solutions, work he said previously would have taken months. He argued the same pattern holds for QA: bugs that would have been found three or four weeks from now can be found within two days, and instead of spending most of his time clicking through repetitive test steps, he can define what to test, have ChatGPT Work or Codex run it, and then receive bug reports to help the team fix issues.

Of course, when you combine “always on” autonomy with access to Slack and email, executives immediately think about security, compliance, and data governance. When pressed on privacy concerns, Geri said privacy “is incredibly important,” and emphasized the key principle is that it is always in the user’s control. He pointed to OpenAI’s enterprise security infrastructure, saying enterprise accounts have ZDR, and users can opt out of letting their conversations help improve future models, which many users do. The comment ties back to assurances OpenAI made when it first launched ChatGPT Enterprise in August 2023, including that OpenAI wrote it does not train on business data or conversations. For workplace deployments, especially in regulated industries, those statements will matter as much as the feature list.

Bottom line: ChatGPT Work is OpenAI’s clearest attempt yet to reposition ChatGPT as a workplace platform, not just a chatbot. It rolls out starting with Pro, Enterprise, and Edu users, then expands to Plus and Business users over the next few days, with Plus included from day one in line with Geri’s “big feat” point about making the power accessible across paid plans. For founders, investors, and anyone building or buying enterprise AI workflows, the strategic question is simple: if an agent can keep running across hours, coordinate across Slack, Gmail, calendars, GitHub, and Docs, and package outputs into real deliverables, what happens to the workflows and budgets that used to sit between “request” and “execution”?