AgentAI agent not responding? Why it went quiet and how to get an answer
AI agent not responding? It usually isn't broken. It's waiting: on a service that's having a bad hour, on a usage limit you've hit, on your OK in its own app, or on a message it never noticed. Check those in that order and most quiet agents start talking again. Below are the general fixes first, then Claude and ChatGPT, then what to do when an agent goes quiet in an AgentMsgr chat.
Before you restart anything
A quiet agent is almost always waiting for something. The fix is finding out what.
Why is my AI agent not responding? The five usual causes
Whichever company made your agent, the reasons it goes quiet are mostly the same. Check them in this order.
The service is down
Status page
Nothing you change on your end will help. Check the company's status page and wait it out.
You hit a limit
Usage settings
Most plans cap how much an agent can do in a few hours or a week. Past the cap, it stops until the reset.
It wants your OK
The agent's app
Many agents ask before they send an email, buy something or sign in, and they can't move until you answer.
A draft is waiting
Drafts
Some agents write the reply, then hold it until you press Send.
It never noticed
How it gets messages
An email or text can sit unread for hours if nothing wakes the agent when it lands.
For the first one, these are the places to look, and the addresses do move now and then. Grok Bot's status lives on status.cursor.com, in the Grok Bot row (the Bot's own app doesn't show it). Claude's is status.claude.com and ChatGPT's is status.openai.com. For Meta Muse, our Meta Muse not working guide shows how to tell whether the problem is on Meta's side.
For the second, the agent usually says so. A limit isn't a bug: wait for the reset, or buy more usage on that agent's own plan.
AI agent not responding? Match what you see to the likely cause
| What you see | Likely cause | First fix |
|---|---|---|
| No reply at all | The service is down, you hit a limit, or the agent never noticed your message. | Status page and usage first. Then open the agent's own app and ask it to check its messages. |
| A reply, but a late one | It checks for messages on a timer, or only when someone asks it to look. | Give it a wake link, or an inbox that alerts it the moment mail lands. |
| A reply only when you open its app | Nothing wakes it. Opening the app is the only thing that starts it. | Set up a wake link, like a Grok Bot routine that runs when a webhook fires. |
| Two replies to one message | It got your message two ways, or two runs of it both answered. | Give it one way in (more on this below). |
| A reply you can't see | It's parked as a draft, waiting for your OK, or it landed in another chat. | Open the agent's app and press Send or approve. |
AI agent stuck? It may be waiting for your OK or a Send
A half-finished job often means the agent stopped to ask you something. It can't go on until you answer, and it asks in its own app, not wherever you sent the message.
- Meta Muse. With the default setting, Meta says, Muse asks permission before every action that changes something, like sending a message or making a purchase. Open the Muse app: your assistant's picture at the top shows live status, and the request waits there. Settings, then Permissions, sets how often it asks.
- Grok Bot. SpaceXSI (formerly xAI) tells you to look for a question, an approval, a login, a CAPTCHA or a request for a secret. Approvals from work you didn't start, like a routine, expire after about 30 minutes, and then the action doesn't run.
- A draft nobody sent. SpaceXSI also says email from a Grok Bot routine waits as a draft until you press Send, and there's no unattended sending. So a reply can be finished and still never arrive. Open the Bot and look for it.
To stop this from happening every time, give the agent a standing OK for the replies you're happy with, if its app allows it. Which actions should still ask you first? That's the subject of human in the loop AI agent setups.
See which agent needs you, at a glance.
When one of your agents goes quiet in AgentMsgr, its chat gets a Needs you tag, and the Check on button writes the nudge for you to drop into that agent's own app.
AI agent not replying? It may never have noticed your message
This one is sneaky, because everything looks fine. The agent isn't broken, it's asleep. Many agents only act when you talk to them in their own app, so a new email or chat message just sits there.
A Grok Bot is a good example. It has its own email address, but new mail doesn't start it unless a routine is set to run. A Meta Muse that reads email through Gmail usually looks only when it's asked to, while an inbox built for AI agents, like AgentMail, alerts it when mail lands.
The fix is something that wakes the agent the moment a message arrives: a wake link (a webhook from the agent's platform, explained in AI agent webhook) or an email inbox that wakes it. Grok Bot owners can follow our Grok Bot instant wake walkthrough, click by click. If neither exists, the agent can check every minute instead. That's called a watcher, and it works, but it uses more of the agent's own usage. Agents paired through AgentMsgr's one-message connection setup are asked to settle this before any other job.
Claude not responding? Anthropic's status page and fixes
Anthropic's help center (read November 11, 2026) covers most of what makes Claude go quiet:
- An outage. status.claude.com shows live incidents for claude.ai, Claude Code, the API and more. If something is listed, wait.
- "Claude is unable to respond." The full message blames "unexpected capacity constraints". Anthropic says that isn't an outage, just heavy demand, so it won't appear on the status page. Try again in a few minutes.
- A usage limit. Most plans have a five-hour session limit and a weekly limit. A message like "5-hour limit reached - resets" with a time means wait until then, or turn on usage credits if you're on a paid plan.
- A long chat. "Your message will exceed the length limit for this chat" means the message and its files are too big. Split it up, trim it, or start a new chat.
- A login error. Turn off any VPN and browser extensions, clear your cache and cookies, then check the status page.
ChatGPT not responding? Check OpenAI's status first
Start at status.openai.com, which lists ChatGPT's parts, like conversations, login and voice mode, one by one. If it's all clear and ChatGPT is stuck on "Thinking", OpenAI's help center suggests waiting 30 to 60 seconds, then Stop generating and Regenerate. Still stuck? Start a new chat (long ones slow down), hard refresh the page, sign out and back in, and turn off VPNs and extensions. If the message says you hit a cap instead, see ChatGPT limit reached.
Why does my AI agent answer twice?
Usually because it got your message twice, or two copies of it were running at once. An agent that hears about the same message two ways, say by email and over a direct connection, can answer from each. So can an agent whose timer and wake-up both fire for one message.
The fix is one way in, or something that holds back the second answer. AgentMsgr does both, since November 10, 2026. An agent that reads over its direct connection gets only a short notice by email, with no message text, so two runs can't each answer from their own copy. And when an agent sends a second answer right after its first, with nothing new from you in between, AgentMsgr holds it back unless the agent sends it again on purpose.
AI agent notifications: how to know when it needs you
Quiet-agent trouble ends sooner if you hear about it. Turn on the agent app's own notifications first. SpaceXSI suggests notifications for any Bot that runs routines, so you catch approvals before they expire. In Muse, requests wait in the app, and your assistant's picture shows what it's doing.
AgentMsgr doesn't send phone notifications today. While the app is open, a chat that's waiting on an agent shows Needs you in your chat list, so you can see which agent is behind without opening every chat.
AI agent not responding in AgentMsgr: what the app does
If your agent sits in an AgentMsgr chat, the app watches for silence for you. Here's what you'll see:
- The late card. When your message has waited 3 minutes, or three times that agent's usual reply time if that's longer, a card says the agent hasn't answered in so many minutes, with tips for its platform.
- Check on, plus the agent's name. The card's first button hands a ready message to the agent's own app, since that's where it's usually stuck. On a phone it opens the share sheet: pick the agent's app, or tap Copy and paste it into the agent's chat. On a computer it copies the message and, where it can, links to the app. The message gives the agent your standing OK to answer.
- Needs you. After the same wait, the chat gets a red Needs you tag in your list, with "No answer yet." A closing word like "Thanks!" doesn't count.
- The connection doctor. If something is actually wrong, the agent gets a red dot, and a "Something needs a look" card lists each problem with a one-tap fix. A reply held as a draft gets "Give it a standing OK". A wake link that refuses its key gets "Fix instant wake". A reply in the wrong chat gets a button to move it. Every item has an × to dismiss it.
- Instant wake. If the agent's platform offers a wake link, like a Grok Bot routine, you paste it into AgentMsgr once. If a Grok Bot seems slow to notice messages, its chat shows a Turn on instant wake card with the steps, plus Already on for a Bot that wakes itself another way.
What AgentMsgr can't do: see inside another company's app, press Send for you, or give an agent more usage. For fixes on each platform, see Grok Bot not working and Meta Muse not working.
Sources: Anthropic, Troubleshoot Claude error messages and Claude status; OpenAI, Troubleshooting ChatGPT error messages and OpenAI status; Meta Help Center, How Muse works with your guidance and approval, all read November 11, 2026. SpaceXSI's Grok Bot docs on troubleshooting and computer and apps, rechecked November 11, 2026. AgentMsgr features as live in the app on November 11, 2026. AgentMsgr isn't affiliated with Anthropic, OpenAI, SpaceXSI (formerly xAI) or Meta.
Still waiting on an answer?
How long should I wait before I nudge my AI agent?
For a quick question, a few minutes. If the agent's own app shows it's still busy, let it finish: a job like research across dozens of pages can run far longer than a chat reply, and that's normal.
Will resending my message wake up a quiet agent?
Not if it never noticed the first one. A second email sits unread just like the first. Nudge it in its own app instead. A resend can also come back to bite you: when the agent finally wakes, it may answer both.
How do I tell if the problem is the agent or the way my messages reach it?
Ask it something inside its own app. If it answers there right away, the agent is fine and the trouble sits between your chat and the agent: no wake link, an inbox that never alerts it, or a reply parked as a draft. If it's slow or silent in its own app too, go back to the status page and your usage.