When you need to scale up your infrastructure, you should enable as much automated options for maintenance as possible. One of the available options for devops agents are included under Organization Settings -> Agent pools -> Settings.
There you can define automated procedures for cleanup on your agent pools.
In my setup, I changed the days to keep unused working directories to 20.
The working directories of the agent are some folders with specific numbers inside C:\agent\work.
Every time a new build is initiated a new folder for this specific run is created. If the same pipeline runs more than one time, then the same working directory will be kept and the files will get overridden. For example lets say my pipeline A is bind to the folder 5. Then every time the pipeline A runs, then the folder 5 will be used for the sources (git repositories) builds, artifacts etc. All previous data hosted there will be deleted and written again.
The maintenance jobs will remove any directories than are not used for x period of days. In my example I had set up 20 days for that task.
You can configure agent pools to periodically clean up stale working directories and repositories. This should reduce the potential for the agents to run out of disk space. Maintenance jobs are configured at the project collection or organization level in agent pool settings.
You can check your history under Organization Settings -> Agent pools -> Maintenance History.
You can also download the log and figure out how much data have been deleted.
Video tutorial on YouTube: