OpenAI launched ChatGPT Work on July 9, 2026, positioning ChatGPT more explicitly as an agent that can complete work. The update is not just another chat mode. It is a way for an agent to move across apps, files, browser work, desktop work, and scheduled tasks, then break a goal into steps and keep going.
OpenAI says ChatGPT Work is built with Codex technology and can create sheets, slides, docs, and Sites while staying with complex projects for hours. That shifts ChatGPT from a real-time Q&A surface toward a delegated unit of work. OpenAI also says more than 5 million people use Codex every week, and more than 1 million now use it outside software development, which shows the coding-agent pattern spreading into broader knowledge work.
The release is tied to GPT-5.6. OpenAI describes the model as suited for multi-step reasoning and creating materials that follow templates and reference files. The examples include month-end budget variance analysis, campaign briefs, sales meeting preparation, and turning customer research into marketing assets for different markets. These are cross-source, cross-format tasks that still need human direction and review.
Scheduled Tasks and connected tools are central to the update. ChatGPT Work can check Slack, Teams, email, dashboards, or websites on a schedule, then refresh docs, slides, meeting agendas, or product ideas. That moves the agent from one-off requests toward recurring monitoring and update loops.
OpenAI is also merging the Codex app into the new ChatGPT desktop app, adding browser, Computer Use, local file, and app-control capabilities. That matters because a lot of enterprise work does not live only behind cloud APIs. It sits across desktop files, browser tools, CRMs, calendars, and team communications.
Governance is a major part of the announcement. OpenAI says Enterprise and Edu admins can manage access, company context, connected tools, browser use, network access, and sensitive actions. The Compliance API also gives visibility into conversations and actions. Once an agent can access tools and take action, permissions, auditability, and approval points become as important as model quality.
The signal is clear: AI agents are moving from personal productivity features into the enterprise workflow layer. The next competitive question is not only which model answers best. It is whether an agent can safely connect data, tools, desktop work, browser work, and scheduled execution while keeping humans in control.
