What used to include just your laptop screen and its keypad with an MS Word draft open has now transformed into multiple screens, .md files, and workflow triggers. However, some key decisions are still for the human brain to make: 1. Comparison between the effort required to do something manually and fixing an output generated by AI. If an agent you have designed to capture screenshots takes one in the time you would take 10 manually, why use the agent? 2. Is building something from scratch better-suited with prompting or writing it yourself and then feed it to an .md context file? The latter does not only train better, but also acts as a quality assurance check on your own work.

3. Minor scope changes are flagged by AI agents on JIRA and new stories are created. The same cannot be replicated while writing release notes. Sometimes, 3 changes impacting the same workflow are presented as 3 separate stories. You do that in release notes, and it loses meaning while deteriorating readability.