Skip to main content
There are multiple failure modes for agents, and it’s not always clear why an agent failed to complete the task.

​
Can’t click on the correct element

This usually occurs due to an issue in agent context. The parsed DOM likely does not contain the desired element.

Pause and view clickable elements

To debug, you can pause the agent and view the Clickable Elements on the page during a live session. Edit prompts based on the parsed representation the agent is viewing. If the element you are looking for is not included, you can add a custom element.

​
Lack of proper tool access

Sometimes you’ve simply forgotten to add the proper tools to your agent. If you go to your agent’s Agent editor, you’ll be able to both create and add tools to your agent.

​
Long-horizon tasks

For longer horizon tasks, single-prompt agents might not make sense. We have a detailed blog post on how we built multi-agents here. We’re always happy to help you build complex multi-agents. Contact us in Slack to make some time to go over best pratices and tips.
⌘I