A surprising percentage of that routine work that can be machine assisted is already automated. Even an agentic AI LLM doesn’t do anything a clever bash script couldn’t. (And bash does it without hallucinations.)

So the opportunity is that a we can automate things that were not worth automating, before.

But the challenge is that anything with crazy good returns on efficiency is likely to already have been automated better.

So we have a really expensive tool hunting for huge efficiency gains among the scraps leftover from earlier better automation solutions.