Work that doesn't make people wait.
Move slow work into the background, retry what fails, and run jobs on a schedule.
Tried again, waiting a little longer each time.
Retries survive restarts, so no work is lost.
Next up.
Run work every hour, every night or on any timetable.
Nothing fails quietly.
Failed jobs are kept with their reason, ready to run again. Start jobs from any flow, or react to what happens in your system.