If you know software and know how to steer them, models are good at helping improving CI efficiency (up to the extent of github being github). a few tricks are less obvious, like leveraging Azure VM's temporary local disk more effectively.
Yeah. Honestly, this is mainly due to tests which had a 30m timeout they were waiting on :) bad edge case that happened to trigger a few times on a few runners
Likely, I'll be fine next month.