AI for on-call: cut the writing, not the thinking
On-call costs you the context reload, the status update, the handoff and the writeup. AI cannot take the page. It can take most of the writing wrapped around it.
On-call fatigue is not mostly about the number of pages. It is about what each page costs you: the context reload, the status update, the handoff, the writeup after. AI cannot take the page for you. It can take a real chunk of everything wrapped around it.
The escalation brief. You need to pull someone in and they are starting from zero. Give the AI what is broken and the impact, when it started, what has been tried and the result, your current theory, and what you need from them. Ask for under 150 words, nothing that does not change what they do next. You get a brief that respects the fact that they were asleep.
The status update. Same facts block, different audience. Impact, start time, current status, what you know, what you are doing, next update time. Regenerate it for leadership, for the whole company, for a customer. It stays calm and drops the jargon each audience will not know.
The handoff note. Bullet points at end of shift: what is still open and the next step, what you are watching, what you resolved, any heads-up like a freeze or a known-flaky alert. Out comes a structured note the next person can actually act on, with the likely-to-page items on top.
The runbook you keep meaning to write. You have fixed the same thing three times from memory. Describe how you handle it to the AI and ask for a runbook someone else could follow at 3am: how to confirm it is this and not something similar, the fix steps with exact commands, how to verify recovery, when to escalate. Then you clean it up instead of writing it from scratch.
The weekly review. At the end of your rotation, feed it your notes and the alert list. Ask for total pages, actionable versus noise, the three noisiest alerts with a specific tuning for each, and a couple of concrete follow-ups. That review is how noise actually gets fixed, and it is usually skipped because writing it is a chore.
None of this removes judgment from on-call. It removes the transcription.
Get the prompts
Five of these, written out in full with the fill-ins and guardrails already in them, are free on GitHub. No email needed.
A sixth, the postmortem prompt with a complete worked example, comes free by email.
Get the postmortem prompt See the five free prompts
All 42, organized by the job in front of you, are $9: the Cronhaven Ops Prompt Library.