Engineering
AI work automation: measure one task before you automate ten
· 5 min read
Last month we argued for starting with one repeated task. This month: how to know whether that task is worth automating at all.
Time it honestly
Do the task by hand three times and write down how long each run takes, including the small corrections. If the task takes two minutes a week, automation will cost more than it saves. If it takes an hour a day, a reviewed AI plan starts to make sense.
Define the outcome, not the clicks
AI work automation works best when the goal is an outcome — 'this report reaches this folder' — rather than a script of clicks. Outcomes survive small changes in the tools; click scripts break on the first redesign.
Keep the record
Whatever you automate, keep a readable log of what ran. ORION's activity history exists for exactly this: automation you cannot audit is automation you cannot trust.
