Who hurts
On-call engineers at companies with a postmortem culture and no time for it.
The problem
The write-up is owed within 48 hours and takes three hours of archaeology across PagerDuty, Slack, deploys and dashboards — so it gets written badly, late, or not at all.
What you build
On incident resolve, pull the alert chain, the Slack thread, the deploys in the window and the graphs that moved, and produce a draft with a real timeline, contributing factors and open questions. Humans do the analysis; the machine does the archaeology.
Why now
Every source is already an API. The missing piece was correlating them into one timeline, which is exactly what long-context models are good at.
Validate it this week
Do it manually for three public incidents from companies that publish postmortems. Compare your draft against the real one, publicly.
Why you'd keep winning
Timeline correlation quality across a specific tool stack. Also the least defensible part of the pitch — lean on workflow fit instead.
The honest risk
Postmortems are a cultural practice. Companies that don't do them won't buy, and companies that do them well already have a template and a habit.
The prompt is written to make an AI argue with you before it writes code — that first round of pushback is worth more than the scaffold.