Don't stop here
Hand-picked guides our readers explore right after this one.
Master ChatGPT with advanced prompting techniques, mega-prompts, and proven frameworks
Read the guideExpert guide to Claude prompts with XML tags, artifacts, and complex reasoning
Read the guideAI prompts for product strategy, user research, roadmapping, and stakeholder communication
Read the guideChatGPT agent mode breaks in three recognizable ways: it is not available at all, it is available but hangs or fails mid-task, or it stops working because you hit the monthly task quota. Agent mode needs a paid plan (Plus, Pro, or Team) and it runs an automated browser, so the usual culprits are plan/eligibility, a stale Custom Browsing connection, browser extensions that block its iframe-based browser control (uBlock Origin, Privacy Badger, dark-mode injectors), and the task cap (about 40 tasks per month on Plus, 400 on Pro). This guide covers enabling it, getting past the quota, and fixing the extensions and hangs.
Agent mode does not appear in the tools menu
It is greyed out or says it is unavailable on your plan
The agent starts a task then hangs, stalls, or silently fails
It gets stuck on cookie banners, logins, or CAPTCHAs and stops
It says you have reached your task limit for the cycle
Agent tasks are capped per billing cycle, roughly 40 per month on Plus and 400 on Pro. Once you hit it, agent mode stays disabled until the cycle resets. There is no fix except waiting for the reset or upgrading.
Agent mode requires a paid plan (Plus, Pro, or Team) and can roll out by region. On a free plan, or an ineligible region, it will not appear.
The agent connects to external services through Custom Browsing. If that connection is stale, the agent cannot reach sites and tasks fail. Resetting it re-establishes the connection.
The agent drives an iframe-based browser panel. uBlock Origin can block that panel, Privacy Badger can intercept its WebSocket connections, and dark-mode injectors can corrupt the pages it renders, all of which make it hang or fail.
Cookie consent walls, logins, and CAPTCHAs can stall the agent mid-task, since it cannot always clear them on its own.
When to try: First, if agent mode is missing or disabled
Go to Settings, and confirm you are on Plus, Pro, or Team. Then check whether you have hit the monthly agent-task limit (about 40 on Plus, 400 on Pro). If you are at the cap, the only options are to wait for the cycle to reset or upgrade, there is no other fix for a quota block.
When to try: If the agent fails as soon as it tries to browse
In Settings, reset Custom Browsing so the agent can reconnect to external services. A stale connection is a common reason tasks fail immediately, and re-establishing it fixes many 'agent won't run' cases.
When to try: If the agent hangs or the panel does not render
Turn off uBlock Origin (it blocks the agent's browser control panel), Privacy Badger (it intercepts the agent's WebSocket connections), and any dark-mode injector (it can corrupt rendered pages). Or run agent mode in a clean browser profile with no extensions.
When to try: When the agent hangs partway through
If the agent stalls mid-task, refresh the page, simplify your instruction into clearer steps, and restart the agent from scratch. Narrow, well-specified tasks succeed far more often than broad, open-ended ones.
When to try: If it stalls on a consent wall, login, or CAPTCHA
The agent struggles with logins, CAPTCHAs, and aggressive cookie walls. Where possible, point it at tasks that do not require passing those, or complete the login step yourself first if the workflow allows.
When to try: To confirm whether it is environment-specific
If agent mode still will not run, test in a different Chromium browser or a fresh profile to rule out local interference. This isolates whether the problem is your environment or your account.
Track your monthly agent-task count so you do not hit the cap unexpectedly
Run agent mode in a clean browser profile without ad blockers or privacy extensions
Give the agent narrow, well-specified tasks rather than broad open-ended ones
Avoid workflows that depend on CAPTCHAs or logins the agent cannot clear
If agent mode still shows as unavailable after confirming a paid plan, resetting Custom Browsing, and testing without extensions, file a ticket via help.openai.com with your plan name and billing-cycle start date. OpenAI typically resolves confirmed account bugs within one to three business days. A quota block, though, is not a bug and support cannot lift it.
Agent mode needs a paid plan (Plus, Pro, or Team) and rolls out by region, so it will not appear on free or ineligible accounts. Confirm your plan in Settings. If you have a paid plan and it is still missing, reset Custom Browsing and test in a clean browser profile.
You likely hit the monthly agent-task limit (about 40 on Plus, 400 on Pro). Once you reach it, agent mode is disabled until your billing cycle resets. There is no workaround except waiting for the reset or upgrading to a higher tier.
uBlock Origin blocks the agent's iframe-based browser panel, Privacy Badger intercepts its WebSocket connections, and dark-mode injectors can corrupt the pages it renders. Disable those, or run agent mode in a clean browser profile with no extensions.
The agent cannot always clear CAPTCHAs, logins, and aggressive cookie walls. Refresh and simplify the task, complete the login yourself first if the workflow allows, or choose a task that does not depend on passing those blockers.
That usually means a stale Custom Browsing connection. Reset Custom Browsing in Settings so the agent can reconnect to external services, then retry with a simplified task.