How to tell if an automation is worth building
Not every manual task should be automated. A quick framework for deciding what's worth the engineering time.
Automation has a seductive pitch: do it once, save time forever. But automation also has a cost — it has to be built, maintained, and debugged when the inputs change.
The math is simple. Multiply how often a task happens by how long it takes, then weigh that against the build and upkeep cost. High-frequency, error-prone tasks are the obvious wins.
The hidden factor is reliability. A task humans do imperfectly but recover from can become a silent failure when automated. Build observability in from day one.
Our rule of thumb: automate the boring, repetitive, well-defined tasks first. Leave the judgement calls to people — at least until the AI is genuinely good enough.