> In the culture of automation, we wiggle and iterate among four overlapping phases: (1) document the steps; (2) create automation equivalents; (3) create automation; and (4) create self-service and autonomous systems.

Notably missing is the most important, difficult and time-consuming (thus potentially defeating the whole purpose) phase: _maintain_ automation. The code will break, it will make incorrect assumptions, it will fail in rare edge cases, it will have unintended side effects, people will use it wrong, etc. "It works right here right now" is like 10% of the whole journey.

Not only maintenance of automation can cause more work than it saves. It has the potential to cause unpredictable, burst-like work in the worst possible times (for the maintainer or the users), which is a problem in itself. A steady background level of predictable, simple manual workload may be preferable to sudden bursts of complex troubleshooting.

I do not say "never automate", I say "life has a surprising amount of detail" so be smart about when to automate.

You are making a big assumption that it is your decision to automate or not to automate and that everything is in your control. In large orgs, even small workflows touch teams that are not under one command. They will have disparate technologies, different release schedules, processes and complex ownerships. Forget automation, if it works by manual methods, then it is doing great.

(2018) In 2019 (193 points, 46 comments) https://news.ycombinator.com/item?id=20504471