I frequently get connection failed errors when the thinking agent (e.g. claude 3.7, claude 4) doing multiple tasks and continues work after tool calls limit. I have stable 200 mbp/s optic fiber connection, connected PC with ethernet cable to the router and no other services in my house have issues with internet connection. Regardless, it would be great if agents could save their work plans before implementation so they can use it as a reference when the connection fails and the next request will require reading multiple files with hundreds lines of code. It is rather common in my use case to use most of tool calls on analyzing the project files before doing the actual work and this can take a couple minutes. Having a feature that allows agents quickly resume the planned work in case of failure would help tremendiously. Sure, I can copy-paste agent’s outline and write my description of changes it made so far but it is time consuming and can be automated.
5 Likes
You’re pretty good, I often encounter it not doing anything, connecting incorrectly, and then deducting the number of requests (paid)
1 Like
This would be a great feature. It would also help in all those unfortunate “(thinks for a minute, making a whole long plan) I will now write code you requested (ends response without writing anything, losing all the plan)” cases.