02 · Workday
Eight minutes to two
Agentic incident comms on Gemini, PagerDuty and Slack — plus a Claude runbook agent.
8 → 2Comms, minutes
How I ran it
The sequence on the night.
- 01
PagerDuty fires the P1. The room exists; the broadcast does not.
- 02
Gemini drafts who is hit, what is down, who is on the bridge, what happens next.
- 03
Slack ships it. Eight minutes becomes two — executives act while engineers still pull logs.
- 04
A human still sends it. The agent does not talk to customers alone.
- 05
Claude retrieves the runbook for that service, then parses MTTD / MTTI / MTTR so the PIR is not a memory test.
At Workday I own global P1/P2 major-incident and problem governance for a SaaS platform used by more than half of the Fortune 500. Restoration still depends on engineers and vendors. What used to waste the first eight minutes was the broadcast: who is impacted, what is down, who is on the bridge, what happens next.
I built an agentic workflow on Google Gemini, PagerDuty and Slack that writes and ships those notifications. Delivery time dropped from eight minutes to two. That is not a demo. That is the difference between executives guessing and executives acting while the technical room is still pulling logs.
Alongside it I trained a Claude agent on our operational runbooks, so the team can get step-by-step recovery guidance for the specific service or infrastructure that is down — not a generic chatbot, the book we already wrote, retrieved under pressure. A second Claude model parses major-incident data for MTTD, MTTI, MTTR, owners, gaps and problem trends, so the PIR is not a memory test.
With the automation team I also put proactive ticket alarming into PagerDuty and Jira. Mean time to identify fell 14% over three quarters. The point is not that we use AI. The point is the clock moved.