If Agent Mode in ChatGPT Atlas has stopped responding, gotten stuck mid-task, or won’t activate at all, you’re not imagining it. Multiple users on OpenAI’s own developer forum flagged this exact problem in the past week, and reinstalling the app hasn’t fixed it for most of them.
We dug into what’s actually going on, tested the usual fixes, and sorted out what’s worth trying from what just wastes your time.
What Agent Mode Is Supposed to Do

Quick refresher before we troubleshoot. Agent Mode is the feature inside Atlas that lets ChatGPT actually act in your browser; it opens tabs, clicks through pages, fills forms, and completes multi-step tasks instead of just reading and answering.
Plus, Pro, and Business subscribers can use it, and you can turn it on either from the tools menu in the chat composer or from Settings > Agent mode.
It runs under some built-in guardrails: it can’t execute code, install extensions, touch your file system, or use saved passwords and autofill. It also skips your regular browsing history when it visits pages.
That matters for troubleshooting, because some “fixes” people suggest like clearing your browsing history won’t touch anything Agent Mode actually depends on.
Agent Mode vs. the Regular Ask ChatGPT Sidebar

A lot of people report “Atlas is broken” when they’ve simply landed in the wrong mode. Here’s the actual split:
| Ask ChatGPT Sidebar | Agent Mode | |
| Reads the current page | ✅ | ✅ |
| Answers questions with citations | ✅ | ✅ |
| Clicks buttons / fills forms | ❌ | ✅ |
| Opens new tabs on its own | ❌ | ✅ |
| Available on free tier | ✅ | ❌ (Plus/Pro/Business only) |
| Shows a visible cursor acting on-screen | ❌ | ✅ |
| Can be paused/interrupted mid-task | N/A | ✅ |
If you’re on the free tier, Agent Mode won’t appear. OpenAI doesn’t include it there. Rule that out before you assume something’s broken.
Known Guardrails (So You Don’t Mistake a Feature for a Bug)
Some of the most common “Agent Mode isn’t working” reports actually describe the feature doing exactly what OpenAI built it to do. By default, Agent Mode:
- Can’t run code in the browser or install extensions
- Can’t access your file system, other apps, or saved passwords/autofill data
- Won’t auto-download files
- Pauses and asks for your approval before logins, checkouts, or anything sensitive
- Skips your normal browsing history
If your agent stalls right before a login screen or a payment page, it’s very likely waiting on your confirmation, not frozen. Check for a confirmation prompt before you assume it crashed.
For Business/Enterprise users specifically: workspace admins can disable Agent Mode organization-wide from their admin settings. If it’s missing only for you and not your coworkers, ask your IT admin first before you troubleshoot further.
The Current Problem
Here’s what makes this round of complaints different. It’s not the usual “agent got confused mid-task” issue that’s dogged Atlas since launch. Users report Agent Mode failing outright; it generates no plan, takes no action, and just stops dead.
One user on the OpenAI community forum uninstalled and reinstalled the whole app and still hit the same wall. Another said they’d been stuck for three days without a single response from OpenAI.
As of writing, OpenAI hasn’t acknowledged or fixed the issue, which points to a backend problem rather than something wrong with individual machines.
That’s an important distinction: a local fix won’t help you if the real problem sits on OpenAI’s servers. Worth knowing before you burn an hour clearing caches.
Fixes to Try (In Order)
| Step | What to Do | Why It Might Help |
| 1 | Check your subscription tier | Free tier doesn’t include Agent Mode at all |
| 2 | Confirm you’re on macOS | Atlas only runs on Mac right now; Windows support hasn’t arrived yet |
| 3 | Update Atlas to the latest version | OpenAI regularly ships point releases that squash bugs |
| 4 | Toggle logged-in vs. logged-out mode | Session data sometimes conflicts with the task and fails silently |
| 5 | Restart Atlas completely (not just the tab) | Clears stuck background processes running the agent session |
| 6 | Try a different, simpler task | Shows you whether Agent Mode is fully down or just struggling with complex tasks |
| 7 | Check for workspace admin restrictions (Business/Enterprise) | Admins can flip a toggle that disables Agent Mode org-wide |
If you’ve worked through all seven and it’s still dead, you’ve likely run into the same backend issue other users are reporting nothing you can fix on your end.
What to Do If Nothing Works
- Post your case in the OpenAI Developer Community bug thread more reports in one place tends to push OpenAI to act faster
- Report the issue directly from inside Atlas via Help → Report an Issue
- Fall back on the regular Ask ChatGPT sidebar for anything urgent users haven’t reported problems there
FAQ
Is Agent Mode down for everyone, or just some accounts?
Reports don’t agree; some users say it’s completely dead, others say it works intermittently. That pattern usually points to a partial backend rollout rather than a total outage.
Will reinstalling Atlas fix it?
Current reports say no. At least one user reinstalled the whole app and saw no change, which means the problem doesn’t live on your machine.
Does this affect the regular ChatGPT sidebar too?
No the reports target Agent Mode specifically. The standard “Ask ChatGPT” sidebar for reading and answering questions still works fine.
Is Agent Mode available on Windows?
Not yet. Atlas only runs on macOS right now, so this issue only hits Mac users for the time being.
Why did my agent stop right before finishing a task?
Look for a confirmation prompt first. Agent Mode pauses on logins, checkouts, and other sensitive steps by design that’s not a crash, it’s waiting on you.
Can a workspace admin turn Agent Mode off for me?
Yes, on Business and Enterprise accounts. If coworkers on your plan have it and you don’t, ask your admin before you troubleshoot anything else.
Having other ChatGPT issues? Check out our guide on ChatGPT History Not Showing Fix.