What Is Agent Mode in ChatGPT? The Short Answer
Agent mode in ChatGPT was a feature that let ChatGPT do multi-step tasks for you on its own virtual computer — browsing websites, filling in forms, running code, and handing back a finished document, spreadsheet, or presentation. OpenAI launched it as ChatGPT Agent in July 2025, and it combined the old Operator browsing tool and Deep Research into one mode. Here is the part most guides have not caught up with: OpenAI has retired ChatGPT Agent as a standalone feature.
Its help page now says agent mode is no longer available, and points users to ChatGPT Work for longer tasks and to workspace agents for teams. This guide explains what agent mode was, what it could do, why it changed, and what to use instead in 2027.
Table of Contents
Key Takeaways
- Agent mode in ChatGPT — launched as ChatGPT Agent in July 2025 — let ChatGPT carry out multi-step tasks on a virtual computer, instead of only answering in chat.
- It is no longer available as a standalone mode: OpenAI’s help page now says so, and points users to ChatGPT Work for longer tasks and workspace agents for teams.
- What replaced it: ChatGPT Work for research and finished deliverables, cloud browser for supported web workflows, and workspace agents on Business, Enterprise, and Edu plans.
- Agent mode was paid-only — free-plan users never had it — and independent reviews put Plus at around 40 agent runs a month.
- It was genuinely useful and genuinely limited: it got blocked by CAPTCHAs and login walls, made mistakes, and needed a person to check the result.
- If a guide still tells you to switch agent mode on in the tools menu, it is out of date — check OpenAI’s own help page for the current status.
What Is Agent Mode in ChatGPT?
Ordinary ChatGPT answers you. Agent mode did things for you. When you switched it on and gave it a task — “research three suppliers that meet these requirements and fill in their contact forms”, or “plan a five-day trip to Lisbon and put the itinerary in a spreadsheet” — ChatGPT would spin up a sandboxed virtual computer in the cloud, open a real browser, click through pages, run code, work with your files, and narrate each step as it went, so you could watch, interrupt, redirect, or stop it at any point.
OpenAI introduced ChatGPT Agent in July 2025 as the merger of two earlier features: Operator, which browsed the web for you, and Deep Research, which produced long researched reports. The agent combined them — research, browsing, code, and file creation — in a single mode you selected from the tools menu, or by typing /agent in the message box. At launch OpenAI was clear that the feature was in its early stages and could make mistakes, and that framing turned out to be accurate.
The difference from regular ChatGPT was the shift from conversation to action. A normal chat gives you a draft, a list, or an explanation, and you do the rest.
Agent mode gave you a result: a completed booking form, a filled spreadsheet, a researched shortlist, a slide deck. It used connected apps such as Gmail and Google Drive where you had approved access, and it could work with files you uploaded. That is also why it mattered for the best AI agent platforms conversation — it was the most accessible agent most people ever used, because it lived inside a chat app they already had. For the wider background on what chat AI can and cannot do, see our guide to what chat AI is.
Is ChatGPT Agent Mode Still Available in 2027?
No — not as the standalone experience described in the 2025 launch material. OpenAI’s own ChatGPT Agent help page now states that ChatGPT Agent is no longer available, and the launch announcement is marked as outdated. OpenAI directs users to ChatGPT Work for longer, multi-step tasks and finished deliverables, and to cloud browser for supported browser workflows.
This is worth stating plainly because most of the guides ranking for this question have not been updated: they still describe agent mode as a feature you can switch on today, with setup steps that no longer work. If you follow a guide that tells you to open the tools menu and pick Agent mode, and the option is not there, the guide is out of date — not your account. Check OpenAI’s help page for the current status before you trust any setup instructions.
The change fits a pattern. OpenAI also launched ChatGPT Atlas, a standalone browser with agent mode built in, in October 2025 — then announced in July 2026 that Atlas was being discontinued, and it stopped working in August 2026. The agentic browsing capability did not disappear; OpenAI folded it into the ChatGPT desktop app, the ChatGPT Chrome extension, and Codex instead. Agent mode, in other words, did not die — it was absorbed into other products under other names.
What Replaced ChatGPT Agent Mode?
| Replacement | What it is for | Who it is for |
|---|---|---|
| ChatGPT Work | Longer, multi-step work with a finished deliverable: research, a document, a spreadsheet, a presentation, or a report | Individual users on paid plans |
| Cloud browser | Supported tasks that require navigation across public web pages | Individual users on paid plans |
| Workspace agents | Agents a team builds, shares, and puts on a schedule, running against connected data such as Google Drive, SharePoint, or Slack | Business, Enterprise, and Edu plans |
| Codex | Longer coding tasks and finished deliverables for developers | Developers on paid plans |
| Regular ChatGPT + connected apps | Quick answers, drafting, and tasks that use your approved apps without a full agent run | Everyone, including free users |
The practical rule is simple. Use ordinary chat for quick questions and drafts. Use ChatGPT Work when the task needs several steps or a finished document, spreadsheet, presentation, or report. Use cloud browser when a supported task requires moving across public web pages. Use workspace agents when a team needs the same job done on a schedule. And use good ChatGPT prompts for all of them — the quality of what comes back still depends on the quality of what you ask for.
What Could Agent Mode Actually Do?
When it was live, agent mode’s real capabilities were narrower than the marketing and broader than a chatbot. It could research public information across many pages and compile it into a report; work through websites, clicking and typing like a person; fill in forms and submit them; run code to analyse a spreadsheet or a dataset; create editable documents, slides, and sheets; use connected apps such as Gmail; and narrate its steps so you could follow along and intervene. You could pause it, redirect it, or take over the browser yourself — for example, to type a password, which the agent never saw.
Its limits were just as real, and they are the reason many people found it frustrating. It was slow by design: most tasks took five to thirty minutes, and longer ones needed babysitting. It was routinely blocked by CAPTCHAs, Cloudflare checks, and login walls, which sank many real-world automations. It made mistakes — booking the wrong date, misreading a page, confidently summarising a source it had misunderstood — and the only defence was checking its work. And it could not run code in the browser, install extensions, or reach other applications on your machine; it lived entirely inside its own sandbox.
The honest verdict, then and now: agent mode was a genuine step toward AI that does work instead of describing it, but it was an early step. The tasks it handled best were research and document creation — jobs where a mistake is cheap and easy to spot. The tasks it handled worst were live transactions on unfamiliar websites, where a blocked login or a wrong click could waste half an hour.
How Much Did ChatGPT Agent Mode Cost?
Agent mode was paid-only. The free tier of ChatGPT never included it — not a single agent run. To use it at all you needed Plus, Pro, Business, or Enterprise, and the number of runs depended on the plan. Independent reviews put Plus at around 40 agent runs a month, with Pro at roughly ten times that, but those are reviewers’ figures, not a promise — check OpenAI’s own pricing page for the current limits of whatever replaces it, because plan limits change.
Whether it was worth the money depended entirely on your usage. If you ran a handful of research or document tasks a month, Plus covered it. If you wanted agents running all day, you were pushed toward Pro — which is why many reviewers concluded that agent mode was priced for enthusiasts and professionals, not casual users. Our guide to whether ChatGPT Plus is worth it works through that value question by user type, and our Gemini vs ChatGPT comparison covers the closest alternative if you are choosing between assistants.
Agent Mode vs Regular ChatGPT
| Regular ChatGPT | Agent mode (as it was) | |
|---|---|---|
| What it does | Answers, explains, drafts | Plans and carries out multi-step tasks |
| Tools | Chat, file upload, connected apps | Virtual browser, terminal, file handling, connected apps |
| Speed | Seconds | Minutes — often 5 to 30 per task |
| Output | A reply in the conversation | A finished deliverable: document, sheet, deck, report |
| Your role | Ask and read | Approve steps, check results, take over for logins |
| Availability | Free and paid plans | Paid plans only; now retired as a standalone mode |
What to Use Instead in 2027
The job you wanted agent mode for decides the replacement.
- Research and reports: use ChatGPT Work, or Deep Research inside ChatGPT, and ask for sources you can verify — then verify them.
- Repetitive web tasks on public pages: use cloud browser where supported, and expect to take over for any login yourself.
- Team workflows on a schedule: use workspace agents on a Business or Enterprise plan, where an admin can control which actions each agent may take in each connected app.
- Comparing assistants: our Gemini vs ChatGPT comparison and our AI agent platforms guide cover the alternatives that do similar work today.
- Simple automation without an agent: scheduled tasks inside ChatGPT, or a dedicated automation tool, will do many recurring jobs more reliably than a general-purpose agent — and with fewer surprises.
Is ChatGPT Agent Mode Safe?
The safety design was permission-based and human-in-the-loop, and the same principles apply to its replacements. The agent could only act inside its sandbox, could not see your passwords — it paused and asked you to take over the browser for logins, with no screenshots captured during that window — and it asked for confirmation before sensitive actions such as purchases. On Business and Enterprise plans, agent mode and workspace agents were off by default until a workspace owner switched them on.
The risk was never that the agent would go rogue; it was that it would act on a site you were already signed into, or upload something you did not mean it to, while you were not watching. The rule that applied then still applies now: never give an agent access to an account you would not hand to a new employee, check its finished work before you send or publish it, and keep sensitive actions — payments, password changes, anything irreversible — for yourself.
Frequently Asked Questions About Agent Mode in ChatGPT
What is agent mode in ChatGPT?
Agent mode — launched as ChatGPT Agent in July 2025 — was a mode in which ChatGPT could plan and carry out a multi-step task using tools, instead of only returning a conversational answer. It ran on its own virtual computer, browsed websites, ran code, and created finished documents, spreadsheets, and presentations.
Is ChatGPT agent mode still available in 2027?
No, not as the standalone feature described in the 2025 launch material. OpenAI’s help page now says ChatGPT Agent is no longer available, and points users to ChatGPT Work for longer tasks and workspace agents for teams.
What replaced ChatGPT agent mode?
ChatGPT Work for longer multi-step tasks and finished deliverables, cloud browser for supported web workflows, workspace agents for teams on Business, Enterprise, and Edu plans, and Codex for longer coding tasks.
What is ChatGPT Work?
ChatGPT Work is the agent inside ChatGPT for longer, multi-step work that ends in a finished deliverable — research, a document, a spreadsheet, a presentation, or a report. It is the direct successor to agent mode for individual users.
How much did ChatGPT agent mode cost?
It was paid-only: free-plan users never had it. Independent reviews put Plus at around 40 agent runs a month, with Pro offering roughly ten times that — check OpenAI’s pricing page for the current limits of its replacements.
What is the difference between ChatGPT agent mode and regular ChatGPT?
Regular ChatGPT answers and drafts; agent mode acted. It opened pages, clicked, typed, ran code, and delivered a finished result, narrating its steps so you could intervene — but it took minutes per task, needed supervision, and was blocked by CAPTCHAs and login walls.
The Bottom Line
Agent mode in ChatGPT was a real and useful idea — an AI that does the work instead of describing it — and its retirement says more about how fast this space moves than about whether the idea worked. The capability lives on under other names: ChatGPT Work for your long tasks, workspace agents for your team, and cloud browser for web workflows.
The lesson for 2027 is practical, not nostalgic: when you search for how to use an AI feature, check the date on the guide and the status on the maker’s own help page first, because in this field a year-old tutorial can describe a product that no longer exists. For the current alternatives, see our AI agent platforms guide and our Gemini vs ChatGPT comparison, and browse the Tech Insights archive here at Digital Discoverist, alongside our roundup of the best AI tools.
2 thoughts on “What Is Agent Mode in ChatGPT in 2027? What It Was, What Replaced It”