OpenAI said on July 31, 2026, that ChatGPT Work and Codex together have surpassed 10 million weekly active users, according to a report from InfotechLead and the company’s own channels. The milestone, reached just weeks after ChatGPT Work’s July 9 launch built on GPT-5.6, marks a sharp pivot from asking an AI to answer questions to handing it multi-step workplace tasks.
Beyond the Chat Box: What ChatGPT Work Actually Does
ChatGPT Work is not another writing button. It is designed to operate as a digital chief of staff—an agent that can pull context from apps and files a user connects, then produce tangible deliverables. The tool can retrieve information from emails, shared documents, calendars, project trackers, and CRMs, and convert it into presentations, spreadsheets, websites, project summaries, or business briefings. It can work for hours on complex jobs, pausing to ask for direction or approval before taking sensitive actions, according to OpenAI’s release notes.
Scheduled Tasks add a layer of automation: workflows can run once, repeat on a schedule, trigger from an event, or watch for changes in connected sources. A manager who once spent half a day pulling together a weekly status report—reconciling Outlook messages, Teams discussions, Jira tickets, and a spreadsheet—can now let the agent produce a draft, complete with source citations.
Codex, the coding agent included in the combined 10 million figure, handles software development tasks such as refactoring, debugging, and writing boilerplate code. Together, the two products turn the ChatGPT desktop app on Windows from a prompt-driven companion into a multi-tasking agent.
10 Million Users: What the Number Actually Means
OpenAI hasn’t broken down the figure by product, user type, or activity level. The 10 million total is a combined metric for ChatGPT Work and Codex. It likely includes both free trials and paid subscribers across Plus, Pro, Team, and Enterprise plans. The company counts a weekly active user as someone who interacts with either tool at least once in a seven-day period, but the threshold for “active” isn’t public.
So the number is best read as a measure of early interest, not proof that millions of companies have embedded autonomous agents into their operations. Still, it is a meaningful signal. The AI market is moving quickly beyond developer tools and experimental pilots toward tools that finish entire workflows, not just answer prompts.
For Windows Users: Three Audiences, Three Impact Zones
Everyday Windows Users
If you already use the ChatGPT desktop app, this upgrade turns it into something much more capable. Instead of generating a single email, you could ask the agent to “Review my emails, calendar, and shared files from the last week. Summarize decisions, open tasks, and highlight anything that needs my attention by Monday. Do not send anything without my approval.” The agent works across connected apps—Outlook, Google Drive, OneDrive, whatever you explicitly link. The prompt library that OpenAI published alongside the release includes templates for organizing a workday, cleaning spreadsheets, preparing meeting briefings, and catching up after time away.
The practical difference is that you’re no longer just generating text; you’re delegating the preparation work that surrounds your actual job. But this also means the agent needs access to sensitive data—your inbox, your calendar, your files. That’s a trust leap.
Power Users and Developers
Codex folds into the same desktop app, making it possible to ask the agent to review pull requests, write documentation, debug code snippets, or connect to GitHub repositories. Combined with ChatGPT Work, a developer could tell the agent “Check our Jira backlog for the highest-priority bugs, pull the related code files from GitHub, and draft a fix plan.” This crosses from coding assistant to development orchestrator.
IT Administrators
The governance burden lands here. ChatGPT Work requires users to grant explicit OAuth-like permissions to third-party services—Outlook, Google Workspace, Salesforce, Jira, and dozens of others. Unlike Microsoft 365 Copilot, which operates within a tenant’s existing permissions, audit logs, and compliance boundaries, ChatGPT Work runs as an external agent with delegated access. IT teams have no native centralized control over what data leaves the organization, how long it’s retained, or which actions the agent takes.
For organizations standardized on Windows and Microsoft 365, Copilot remains the more natural fit. It lives inside Outlook, Teams, Word, Excel, PowerPoint, and SharePoint, and respects all tenant-level controls, data-loss prevention policies, and user permissions. ChatGPT Work, by contrast, begins with the question, “What outcome do you want, and which sources should the agent use?” That offers flexibility but introduces risk. The sensible initial posture for IT is to block the app until they’ve evaluated connector scopes, data retention policies, and the ability to require human approval for high-stakes actions.
How We Got Here: The Road from Chatbot to Agent
The timeline is compressed but clear. In late 2022, ChatGPT launched as a conversational model. By 2023, GPT-4 added deeper reasoning, and plugins allowed limited connections to third-party services. Microsoft Copilot arrived in 2024, embedding AI directly into Office and Windows. Google Gemini followed, targeting Workspace. In 2025, Anthropic delivered Claude Enterprise with long-context reasoning. Then, in July 2026, OpenAI released ChatGPT Work, powered by GPT-5.6, turning the assistant into an agent that could handle multi-app, multi-step jobs.
The 10-million-user milestone reflects a broader shift that McKinsey quantified: generative AI could automate activities representing 60 to 70 percent of employees’ current work activities, particularly information processing, communication, and knowledge work. The competition is no longer about which model answers questions best. It’s about which platform can safely finish the most valuable preparation before a human must decide.
What to Do Now: Practical First Steps
If you’re curious or considering deploying ChatGPT Work, start small and stay safe.
For individuals:
- Update the ChatGPT desktop app on your Windows machine to the latest version. Check if your subscription tier includes Work (Plus, Pro, Team, and Enterprise plans should have access; consult OpenAI’s help articles for specifics).
- Choose a read-only task with a clear audit trail. One recommended first run: “Review this week’s project emails, messages, meeting notes, and task tracker. Summarize completed work, progress, decisions, blockers, and next priorities. Cite sources.” Then verify every statement.
- Do not connect accounts that contain sensitive financial, legal, or HR data until you understand the permissions model. Never grant the agent authority to send external messages or modify records without your explicit, separate review.
- Lean on the five prompt templates OpenAI published: organize your day, prepare weekly updates, clean spreadsheets, catch up after time away, and prepare meeting briefings. All are designed to be approval-gated and source-citing.
For IT teams:
- Audit what data connectors your employees are using, if any. Block the app temporarily if you lack visibility.
- Compare ChatGPT Work’s governance model against Microsoft 365 Copilot’s tenant-level controls. If your data lives primarily in Microsoft 365, Copilot may already address the approved use cases with less risk.
- If you choose to permit ChatGPT Work, start with a pilot group. Require that agents never send emails, write to CRMs, or modify production data without human sign-off. Demand that outputs cite sources, and maintain logs of agent actions for compliance review.
What Comes Next
The 10-million-user surge is an early acquisition metric. The test will be retention and trust. Can ChatGPT Work keep users coming back week after week? Will organizations embed it into repeatable, controlled workflows? Or will governance fears stall adoption? The pressure is on OpenAI to deliver not just capability but accountability.
Expect Microsoft to deepen Copilot’s agent capabilities inside Windows and Office—potentially integrating similar multi-step automation while leaning on enterprise trust. Google and Anthropic will continue to push their own agents. For Windows users, the desktop is becoming a battleground for AI agents. The winner won’t just answer better; it will finish more work, safely.