ChatGPT agent no longer exists as a separate mode. OpenAI’s help page for ChatGPT agent now says it “is no longer available” and points people to ChatGPT Work for longer, multi-step tasks and finished deliverables. So if you’re searching for ChatGPT agent mode in 2026, Work is the thing to learn. This guide covers what changed, what Work can do, which plans get it, and how to run your first task step by step.
ChatGPT agent in 2026: the short answer
| Question | Answer (checked 20 Nov 2026) |
|---|---|
| Is ChatGPT agent mode still available? | No. OpenAI’s help article says ChatGPT agent is no longer available |
| What replaced it? | ChatGPT Work, an agent inside ChatGPT for long, multi-step tasks |
| When did Work launch? | Announced 9 July 2026, rolling out on web and mobile from that day |
| Which plans get Work on web and mobile? | Pro, Enterprise and Edu first, then Plus and Business |
| Is there a free way to try it? | OpenAI says Chat, Work and Codex are in the new desktop app on every plan, including Free |
| What can it make? | Docs, slides, sheets and web apps, using your files and connected apps |
| Does it act without asking? | You set what it can access and when it must stop for your approval |
What was ChatGPT agent?
OpenAI introduced ChatGPT agent on 17 July 2025. It joined two earlier tools into one mode: Operator, which could click and type on websites in a remote browser, and deep research, which read many sources and wrote a report. You switched on agent mode from the tools menu, gave it a task such as “find three hotels under $150 near the venue and compare them”, and watched it browse, fill forms and build a file, asking you before anything sensitive.
It was useful but slow, and it lived in a small corner of ChatGPT. In 2026 OpenAI rebuilt the idea around a bigger product. Searches for “chatgpt agent” are still climbing (about 8,100 a month in the US in October 2026), but many of the top-ranking guides still describe the 2025 version, which is why people end up confused when the button isn’t there.
What is ChatGPT Work?
OpenAI describes ChatGPT Work as “an agent in ChatGPT that helps you take on more ambitious tasks.” In practice it is agent mode grown up: instead of one task in one browser window, Work can stay with a project for hours, pull material from your apps, and hand back finished files. Its main abilities, per OpenAI’s launch announcement:
- Finished deliverables: it creates sheets, slides, docs and web apps, not only chat answers.
- Your apps as sources: plugins connect apps such as Slack, Teams, Google Drive, SharePoint, email, calendars and CRMs. Type
@plus the app name in your prompt to use one. - Scheduled tasks: a task can run once, on a schedule, when something happens, or to watch for changes over time.
- Computer Use on desktop: in the desktop app it has a built-in browser and can click and type across your apps and browser in the background.
- Sites (beta): it can build an interactive site or small web app you share by URL.
- Check-ins and approvals: you decide what it can access, when it should check in, and when it needs your approval before acting. You can follow progress, answer its questions and change direction mid-task.
Work is built on Codex technology. OpenAI is also merging the Codex app into the new ChatGPT desktop app, so Chat, Work and Codex sit side by side; Codex stays the coding agent for developers.
ChatGPT agent vs ChatGPT Work
| Feature | ChatGPT agent (2025) | ChatGPT Work (2026) |
|---|---|---|
| Status | Retired | Current |
| How you start it | Agent mode in the tools menu | Switch to Work in the app |
| Task length | Minutes, one task at a time | Can stay with a project for hours |
| Output | Chat answer plus files | Docs, slides, sheets, web apps, shareable Sites |
| Your data | Connectors for some apps | Plugins for Slack, Teams, Drive, SharePoint, email, calendar, CRMs |
| Recurring jobs | Limited | Scheduled and event-based tasks |
| Control of your computer | Its own virtual computer (browser and terminal) | Built-in browser and Computer Use in the desktop app |
| Usage | Monthly agent message limits | Draws on your plan’s included usage, like Codex |
Is ChatGPT agent free? Plans and limits
On web and mobile, OpenAI rolled Work out to Pro, Enterprise and Edu first, then to Plus and Business “over the next few days” after the 9 July 2026 launch. The announcement does not list Free or Go for web and mobile.
The desktop app is different: OpenAI says Chat, Work and Codex are “available to users on every plan, including Free”, and the app is available to download on Windows and Mac. That is the realistic way to try an agent in ChatGPT without paying, but expect tight limits on a free account.
OpenAI doesn’t publish a fixed number of Work tasks per month. Its wording is that “usage varies with the amount of work required, and more complex tasks may use more of your plan’s included usage.” In other words, a two-hour research-and-slides job costs more than a quick checklist. If you are deciding between plans, our ChatGPT Go vs Plus comparison covers what each tier includes.
How to use ChatGPT agent mode now (step by step with Work)
These steps follow OpenAI’s Get started with Work guide.
1. Open Work
In ChatGPT on the web, mobile or the new desktop app, switch from Chat to Work. On desktop, choose Cloud in the composer if you want the task to keep running when your computer is off; Enterprise workspaces keep their Local/Cloud selector.
2. Pick a first task with a clear finish line
Good first tasks have an obvious outcome, the source material ready, and an output you can check. OpenAI’s own examples are analysing a month-end budget variance and preparing for a sales meeting. “Make me rich” is a bad task; “turn these three CSV exports into a one-page summary with a chart” is a good one.
3. Connect the apps it needs
Open Customize → Plugins to connect apps, then mention them in the prompt with @, for example @Google Drive. Only connect what the task needs.
4. Write the prompt like a brief
Cover five things: the outcome, the sources, the constraints, what “good” looks like, and when to stop for review. Example:
Using the Q3 sales sheet in @Google Drive, build a 6-slide deck for Monday’s team meeting: top 5 products by revenue, month-on-month change, one chart per slide, plain language. Don’t email or share anything. Stop and show me the outline before you design the slides.
Our guide to writing ChatGPT prompts that work goes deeper on this structure.
5. Watch, answer and approve
Work shows its progress as it goes. Answer its questions, redirect it if it drifts, and approve important actions when it pauses. If it gets something wrong, say what to fix rather than starting over.
6. Check the output before you use it
Open the doc, sheet or slides and check every number against the source. Agents are fast, not infallible.
5 useful ChatGPT agent tasks to try
- Meeting prep: “Read last month’s emails with Client X and my calendar invite, and write a one-page brief with open questions.”
- Budget check: “Compare this month’s spend sheet to the budget, flag lines more than 10% over, and draft a short explanation.”
- Research to slides: “Find the five most-cited 2026 reports on remote work, summarise each in two lines with links, and turn it into a 5-slide deck.”
- Weekly monitor: a scheduled task that checks a competitor’s pricing page every Monday and tells you what changed.
- Quick web app: “Build a simple Site where my team can pick lunch orders, and give me the share link.”
Safety: what to keep in mind
- Set boundaries in the prompt. OpenAI suggests lines like “stop before sending anything”. Use them for email, posting, purchases and deletions.
- Give the least access that works. Connect one folder, not your whole drive, when that is enough.
- Treat web pages as untrusted. Any agent that browses can meet pages that try to trick it. Read what it did before you approve an action.
- Know where it runs. If a desktop task continues in the cloud while your computer is offline, OpenAI notes the cloud container cannot reach your local files or tools.
- For teams: Enterprise and Edu admins can restrict sensitive actions in connected systems, set spend limits per group, and review Work actions through the Compliance API.
Pros and cons of ChatGPT Work
Pros
- Delivers finished files instead of chat text
- Works across your own apps through plugins
- Scheduled and event-based tasks save repeat work
- Asks for approval on important actions
- Desktop app brings it to every plan, including Free
Cons
- Not on web or mobile for Free and Go users
- No fixed task limit published, so heavy jobs can eat your plan’s usage
- Long tasks still need checking; it can misread data or sources
- The old agent mode guides and tutorials online no longer match the app
ChatGPT Work alternatives
If you want to compare assistants before paying, our Claude vs ChatGPT and Gemini vs ChatGPT comparisons cover each tool’s agent and research features. For browsing-heavy tasks, see our ranking of the best free AI browsers, and if images are part of the job, our ChatGPT image generator guide shows what ChatGPT can create on the free plan.
FAQ
What is ChatGPT agent?
ChatGPT agent was a mode launched in July 2025 that let ChatGPT browse websites, run code and build files on its own virtual computer to finish multi-step tasks. OpenAI has since retired it and moved that work into ChatGPT Work.
What happened to agent mode in ChatGPT?
OpenAI’s help article now says ChatGPT agent is no longer available and directs users to ChatGPT Work for longer, multi-step tasks and finished deliverables. Work was announced on 9 July 2026.
Is ChatGPT agent free?
On web and mobile, ChatGPT Work is for Plus, Pro, Business, Enterprise and Edu. OpenAI says the new desktop app includes Chat, Work and Codex on every plan, including Free, with usage limits that depend on how much work each task needs.
How do I use ChatGPT agent mode?
Switch to Work in ChatGPT, connect any apps you need under Customize → Plugins, write a prompt with the outcome, sources, limits and a review point, then follow its progress and approve important actions.
Is ChatGPT Work the same as Codex?
Not quite. Work is built on Codex technology and shares its core abilities, but it is designed for everyday work like docs, sheets and slides. Codex remains the coding agent for developers.
Can ChatGPT Work send emails or buy things for me?
It can act in connected apps, but you decide what it can access and when it must stop for approval. Tell it plainly, for example “stop before sending anything”, and review each action before you approve it.
Bottom line
ChatGPT agent mode is gone, and ChatGPT Work is the version to learn in 2026. Start with one small task that has a clear finish line, keep access tight, and make it stop before anything is sent or shared. Want help turning AI tools into real results for your website? Get a free SEO audit from SSA Agency.
Sources: OpenAI, ChatGPT is now a partner for your most ambitious work (9 July 2026); Get started with ChatGPT Work; OpenAI Help Center: ChatGPT agent. Checked 20 November 2026.