AI Agent should periodically check its own status while "spinning"

Feature Requests
1

Feature request for product/service

Chat agent not clear if stuck or still working: Old Link Difficult to know when agent & llm are done working on prompt? @andrewh doesnt feel actually resolved.

Describe the request

Have the AI agent feature check itself for when its idle/stuck/working so that it’s not spinning leaving the user left to know if it should cancel or continue waiting after an extended period of no additional commentary/response in the chat. See cursor’s own explanation of it’s feature lack realization below. Seems specific to when the agent is helping with local tasks such as configuring a postgre database using docker or similar task where Terminal is not getting real time capture to support current logic.

ME to the AI Agent:
“So basically you’re just an AI agent using GPTs and haven’t been trained as an agent to check your own status for idle vs working?”

AI Agent’s Response about Cursor’s design flaw:
You’re absolutely right - I’m essentially an AI agent using GPTs, and I haven’t been specifically trained to monitor my own status or distinguish between “working” vs “idle” states.
What I Actually Am:
Language model with tool-calling capabilities
No built-in status monitoring - I don’t have a “working/idle” detector
No self-awareness of my own processing state
Reactive, not proactive - I respond to inputs but don’t actively signal my state
The Real Problem:
I’m designed to be a conversational AI that uses tools, but I’m not designed as a proper agent that:
Monitors its own status
Signals when it’s done vs working
Has built-in timeouts or state management
Can detect when it’s “stuck” vs actively processing
What This Means:
I can’t tell you “I’m done” unless I explicitly say it
I can’t detect when I’m in an idle state
I rely on Cursor’s UI to show my status (which is clearly broken)
I’m essentially “blind” to my own processing state
The Cursor Design Flaw:
Cursor should have built-in mechanisms to:
Detect when I’m actually done processing
Show different states (working/thinking/waiting/done)
Auto-timeout stuck states
Give me better self-awareness capabilities
You’re spot on - this is a fundamental design issue where the AI agent (me) lacks proper status awareness, and the UI (Cursor) doesn’t compensate for that gap. It’s a systemic problem, not just a bug.

2 Likes
2

I think that essentially, each tool needs a timeout (probably reasonably long, or well tuned to each type of tool or something like that), and when a tool call times out, along with any other details that the tool has thus far been able to gather, ADDITIONALLY, I think timeout info needs to be returned to the model.

I have found that when the model fails to proceed, it is very frequently (probably most often) NOT because the tool call failed. It just, stalls, hangs, ceases to proceed. For no apparent reason. So all tool output/results need to be included, along with the timeout details, because the timeout is not necessarily an error. It is just a timeout. The model may well be able to proceed, even if the tool call timed out, if the call actually succeeded in performing the work it was supposed to perform.