What happens when a browser agent gets stuck

Every browser agent gets stuck sometimes. A site shows a captcha. A login expires. A page changes and nothing on it matches what the agent expected. The useful question is not whether it happens but what the agent does next. A tool that keeps retrying, or quietly guesses, is worse than one that stops and says so.

In Beacon, a stuck agent asks the human.

When an agent cannot continue on its own, its lane pauses and its tab comes forward. The tab is brought to the front of Chrome so you can see exactly what the agent is looking at, and you can step in: solve the captcha, sign in again, or click the thing the agent could not find.

Once you have dealt with it, the agent can carry on. You do not have to hunt for the right tab, because the tab that needs you is the one that came forward.

Pausing is per lane. If you have several agents running, each in its own tab group, only the one that is stuck stops. The others keep going. That matters when you run agents in parallel: one agent waiting for you should not freeze the rest of the work.

It also keeps the interruption honest. You are asked about the one thing that needs a person, not about everything at once.

In agent mode, agents work in background tabs and the tab you are on is left alone. A tab only comes forward when an agent needs a human. That is the one exception, and it is the reason for it: bringing a tab forward is how an agent asks for you.

These are the situations where you are most likely to be asked:

• A login wall or an expired session in a profile.
• A captcha or another check meant for a person.
• A step that cannot be undone, such as sending or paying, flagged so the agent can ask first.
• A page that does not behave the way an automated click expects.

Some sites ignore automated clicks or typing. If a step keeps failing, reload the tab or finish that step yourself and let the agent continue. Sites that block automation are better handled through their own API or with a person in the loop; no browser tool can promise to work everywhere.

Look at the tab that came forward. Fix the one thing that blocked the agent, whether that is a sign-in, a check meant for a person or a choice only you can make, and then let the agent continue. If you would rather not continue at all, stop that agent from the Agent Board and nothing else is affected.

You do not have to wait for an agent to ask. The Agent Board in the side panel shows the running agents and lets you stop one. If an agent has been told to stop, press Reset on its row before you ask again.

Asking a human is the safest answer to being stuck, and it fits how people already work in a browser. The agent explains nothing in a long message you have to read; it shows you the page. You fix the one thing and move on.

To see how lanes keep agents apart in the first place, read how two agents share one Chrome without colliding. To get started, follow the setup guide.