Last Updated: 11 July 2026
OpenAI released ChatGPT Work, a tool built for projects that need more than a normal chat response. Give it a goal, connect the files and services it needs, and it can research, plan, create documents, prepare slides, build websites, and keep working through a longer project while you review its progress.
The July 9 launch also brings Codex, browser tools, scheduled tasks, and local file access into a wider ChatGPT experience. Atlas is heading toward closure on August 9, and its browser features are moving into the new desktop app. One product is becoming the central place for research, creation, and daily office work.
ChatGPT Work is a new agent inside ChatGPT that can handle longer projects involving many connected requests. You can give it a goal, connect useful apps and files, and follow its progress as it gathers information, creates materials, and completes several linked steps with limited supervision across a single working session.
The agent can create written reports. It can also build slide decks or simple websites when needed. It connects with tools you already use, such as email or calendars. It can also work with team chats and cloud storage. This lets you gather information, build your work, and make changes in one place without switching between many apps.
GPT 5.6 provides the reasoning behind ChatGPT Work, while Codex supplies tools that were first built for software tasks. That combination is meant to bring the power of coding agents to people who do not write code, including workers in sales, finance, marketing, operations, research, and education during their daily work.
ChatGPT Work can connect with tools you already use at work. This includes Gmail, Outlook, Slack, Microsoft Teams, Google Drive, SharePoint, calendars, and customer systems. It uses plugins to access them. You can let it pick the right one based on your request. You can also guide it by typing the at symbol and the name of the app.
Scheduled Tasks let the agent run on its own. You can set it to run once or at a specific time. It can also react when something changes. A team might use this to update a meeting agenda. It can check a dashboard every morning. It can review customer feedback. It can also update a presentation when new emails arrive.
Examples include:
The desktop app includes a browser for online research and Computer Use for actions across local apps. It can click, type, move files, open cloud documents, and continue a workflow while you review the work. The user can interrupt, redirect, or approve important actions at any time before the agent moves ahead.
Access controls remain important because the agent may reach private files and business systems. Users and administrators decide which tools it can open and which actions require approval.
A finance team could ask ChatGPT Work to gather quarterly revenue data from internal systems, building a clean financial model, flagging unusual trends, and preparing a board ready report with clear charts. A sales team could turn detailed call transcripts into a structured account strategy, identifying upsell opportunities, drafting personalized outreach emails, and creating a polished presentation aligned with each client’s goals.
Launch examples covered lead reviews, product launch checks, airline research, conference planning, forecasting, and sales preparation. Some tests reduced work that once took days or weeks to a matter of hours. These results came from selected early users, so wider performance will depend on the task, data, setup, and review process.
For most people, the useful part may be continuity. The agent can carry details from research into a brief, then use the brief to make campaign materials and adapt them for different markets. That reduces repeated explanations and keeps related tasks inside one project, where the user can review every stage.
ChatGPT Work enters a crowded market for AI agents built for office tasks. It as a direct answer to Anthropic's Claude Cowork and Microsoft's Copilot Cowork. All three aim to help people complete complex projects with fewer manual steps, especially users who want agent tools without learning how to code.
The product also brings Codex into a wider business setting. Coding agents can plan, use tools, inspect files, and correct their own work, which can make them more capable than a normal chat assistant. ChatGPT Work packages those abilities inside a familiar interface for documents, presentations, websites, research, and business systems.
Price will influence whether teams use these agents every day. Complex tasks may consume more of a plan's included usage, and business administrators can set limits for groups or individuals. Smaller GPT 5.6 models may also lower costs for some jobs, giving organizations more options when speed, quality, and spending must be balanced.
Accurate output, access to workplace data, cost, and trust now define the contest. Agents will be judged by how safely they handle ordinary files and daily tasks.
The biggest concern is access to private business data. The agent may connect with email, cloud files, calendars, team chats, and customer systems. A weak permission setup could expose sensitive information to the wrong task. Companies will need clear access rules, regular reviews, and limits for every connected service.
"Before letting something like ChatGPT Work anywhere near client files, we'd want tiered permissions from the start. We actually prefer read-only by default, write access only after slow testing. Every connected system needs an audit trail that can be checked by humans, not just a default checkbox."
Tim Jones Founder and CEO at TKX Media
Accuracy is another risk. This tool can complete several steps without constant supervision, which means one early mistake could affect a full report, spreadsheet, or presentation. Incorrect data may look polished and ready to use. Users still need to check sources, numbers, conclusions, and final files before making business decisions.
There is also concern about employee dependence on automated work. Teams may begin using the agent for research, writing, planning, and routine decisions without building the same skills themselves. ChatGPT Work can save time, but poor oversight may create weaker judgment, repeated errors, and confusion about who is responsible when something goes wrong.
ChatGPT Work began rolling out on July 9, 2026, for Pro, Enterprise, and Edu users on web and mobile. Plus and Business access was scheduled to follow over the next few days. The updated desktop app launched globally for Windows and Mac, with Chat, Work, and Codex available across every plan, including Free.
The launch also marks the end of the separate Atlas browser. Atlas is scheduled for deprecation on August 9, less than one year after its introduction. Its browser tools are moving into the desktop app and Work mode, where online research can sit beside local files, connected apps, and longer projects.
Codex is also merging into the new desktop application, while the older ChatGPT desktop app will be renamed ChatGPT Classic. This consolidation gives users fewer separate products to manage. It also places browser actions, coding work, scheduled tasks, documents, and connected services inside one interface, which may make the product easier to understand.
ChatGPT Work brings several existing ideas into one practical system. It can use connected apps, browser tools, local files, Codex, and scheduled tasks to move a project from an initial request to finished materials. The user still needs to provide clear goals, choose sensible permissions, review outputs, and approve actions that carry risk.
Its early value will be easiest to judge through ordinary work. A useful agent must handle messy files, changing instructions, private data, and repeated tasks with reliable results. The Atlas closure also points to a unified product direction, with browser work, coding tools, documents, and automation gathered inside ChatGPT.
ChatGPT Work is an AI agent that can complete projects involving multiple files, apps, and tasks. A user provides the goal and chooses which services the agent can access. It can then research the topic, organize information, create materials, and revise them when needed. Its output may include documents, presentations, spreadsheets, reports, websites, and simple apps. Users remain involved by reviewing progress, answering questions, correcting the direction, and approving actions that could affect private data or connected systems.
ChatGPT Work started rolling out on web and mobile to Pro, Enterprise, and Edu users on July 9, 2026. Plus and Business users were scheduled to receive access during the following days. The updated desktop app offers Chat, Work, and Codex on every plan, including Free. A separate fixed price was not listed at launch. Usage depends on the size and difficulty of each task, and complex projects may use more of the allowance included with your plan across different kinds of work.
Yes. ChatGPT Atlas is scheduled for deprecation on August 9, 2026, less than one year after its launch. Its useful browser features are being moved into the updated ChatGPT desktop app and Work mode. The new setup combines online research, local files, connected services, scheduled tasks, and Codex in one place. People using Atlas should receive transition information before the shutdown. Future browser tools will likely remain inside ChatGPT as part of the larger app for regular work and research.
ChatGPT Work includes controls for users and business administrators. Every AI agent should begin with limited access. Organizations can manage which people use it, which apps it can connect to, what information it can reach, and which actions need approval. Enterprise and Edu administrators also receive oversight tools and spending controls. Teams should test ordinary tasks, review outputs, and expand permissions gradually. Sensitive financial, legal, customer, or personal data deserves a clear internal policy, named owners, and regular careful checks.
ChatGPT Work can prepare a business report by collecting information from approved files, email accounts, calendars, team chats, cloud storage, and customer systems. It can organize the findings, write the report, create charts, and prepare a related presentation or spreadsheet. The agent can also update the material when new information becomes available. Users should confirm that the data is accurate and review every conclusion. Reports involving financial, legal, customer, or private information need especially careful human checks.
