OpenAI wants ChatGPT to stop being a chat window and start being an employee. Its new product, ChatGPT Work, connects the chatbot directly to your email, Slack, calendar, Google Drive and other business apps, then lets it complete entire multistep projects on its own, not just answer questions. It costs twenty dollars a month, the same as OpenAI's cheapest plan.
The idea is simple. Coders have had this kind of tool for a while through Codex, OpenAI's coding assistant. ChatGPT Work takes that same technology and points it at accountants, doctors, sales teams, and anyone else whose job runs through a computer rather than a codebase.
The catch is trust. To get real use out of the tool, you generally cannot hand it limited access. One OpenAI engineer running the desktop app gave it entry to his inbox, Slack, phone and design tools, and admitted the AI could accidentally pull private information into a document it wasn't supposed to touch. Our own testing found the same pattern: asking for simple read-only access to a cloud drive frequently failed, and the app eventually demanded full permissions to work at all. For any business handling client files, contracts or financial records, that all-or-nothing setup is worth thinking through before connecting anything.
The economics are just as telling. Inside OpenAI, the tool has become nearly universal among staff. Outside the company, uptake among paying customers is a fraction of that. Part of the reason is cost that OpenAI itself is absorbing: heavy use of the agent can run up computing bills several times larger than the subscription price, which means the current twenty dollar fee is likely a temporary discount rather than the real cost of the service. Businesses that build workflows around it now should expect pricing to shift once usage climbs.
OpenAI isn't alone in this race. Anthropic has its own version, called Claude Cowork, and specialist companies like Harvey for law firms and Clay for sales teams are building narrower tools that plug into whichever AI model performs best. That competition matters more than it looks. If OpenAI can't make its agent useful enough to become the default choice for ordinary office work, cheaper or better-suited rivals will take that market instead, and OpenAI's massive spending on computing power will have paid for market share it can't hold onto.
The bigger question for any non-technical business is not whether this technology works, since it clearly can automate real busywork like building reports or organizing schedules. The question is whether the convenience is worth the access you have to give up, and whether you can afford to wait until pricing and permissions settle down before betting a workflow on it. Right now, early adopters are the ones testing that trade-off, and their experience will decide how this plays out for everyone else.