Besides rampant failures in communication and skills allocation, wild U turns of requirements were (sometimes, not even real business requirements) were holding back corporate environments doing a decent job.
With AI, I can only see the rate of such changes sky rocketing due to expectations wildly misaligned with reality. Hence we are unlikely to see any meaningful improvements.
Sorry, missed this post. I don't have any write ups to recommend I'm afraid, for me it was a lot of trial and error, but what really made the whole thing click for me was setting up a linux box and not being able to remember all the mad incantations and flags for everything I need to do on the semi-regular. So I started just putting them in a user-global mise.toml as tasks with nice descriptions to help me remember what they do, and gradually, over time, I'd think "it would be really helpful if this task also did x", so I'd add that. They're basically superpowered aliases with a vastly better user experience.
Then I realised how powerful it was that I could create tasks with dependencies (ie: when a task requires the user to have jq installed, you can add that to the mise.toml) which makes the tasks beautifully shareable across a team. The only tool they need to have installed is mise, and mise handles everything else for them.
"Science moves forward one funeral at a time" - Max Planck
reply