Every week at Convoy, the on-call engineer wrote a retro summary of every incident from their shift. The raw material lived in scattered Slack threads generated by PagerDuty, so the write-up meant hunting down each thread, reading through the comments, and reconstructing what happened — 2 to 3 hours of an engineer's week, every week.

A Python automation on the Slack API that collected every thread and comment belonging to each incident and assembled them into an organized draft, ready for the engineer to review and finish.

2–3 hours of manual work became 20 minutes — roughly an 85% reduction, every single week

The engineer stayed in the loop: the automation did the gathering and organizing, the human did the judging

This is the exact blueprint of the message-triage systems I now install for e-commerce brands: scattered messages in, organized digest out, human approval kept in the loop. The tools changed — Claude instead of a Python script — but the discipline is the same, and so is the result: hours of weekly grind turned into a short review.