AI-written postmortems your team can review

After an incident, the facts are scattered across an alert history, a chat channel, deploy logs, and people's memories. Assembling them into a timeline is slow work, and it often waits until everyone has moved on. An AI agent can draft the postmortem within minutes of resolution: timeline, root cause, impact, and follow-ups. Your team then reviews that draft, corrects it, and agrees on the actions.

Offprint is where that review happens. The agent publishes the draft as a page, the people who were there comment on specific lines, and the agent revises it. Every revision records who wrote it, so the final postmortem shows which parts an agent drafted and which a person changed.

See an example postmortem drafted by an agent for a fictional incident.

What you need

• An AI agent connected to Offprint. See setup for every client.
• Access to the incident's raw material: the incident channel, alerts, deploy history, and relevant logs, through the agent's other connectors or pasted in.

1. Draft from the record, not from memory

Ask the agent to work only from the sources and to mark what it's unsure of:

Draft a postmortem for today's checkout latency incident. Use the #inc-checkout channel, the alert history, and today's deploy log. Include a summary, an impact table, a UTC timeline, root cause, what went well, what went poorly, and follow-ups with owners. Where the sources don't say, write "Unknown — confirm".

Keep the tone blameless: describe what systems and processes allowed the failure, not who made a mistake. Say so in the prompt if your team follows that practice.

2. Publish it as a draft

Publish this to Offprint in the Incidents project as "Postmortem: checkout API latency spike". Mark the status as "Draft, in review".

For anything involving customer data or security, publish privately:

Publish it as a private Offprint document.

Private documents open only for you and the collaborators you add.

3. Review it together

Add the incident lead and the people who responded as commenters or editors: use Share this document in the side panel, or share the whole Incidents project. They comment on the exact line that's wrong ("We rolled back at 14:24, not 14:30") or missing ("Mention that support was told at 14:20").

Then ask the agent to work through the review:

Read the comments on the checkout latency postmortem. Correct the timeline and root cause where reviewers pointed out errors, add anything they say is missing, publish a new revision, and reply to each comment saying what changed, then resolve it.

The agent uses list_comments, update_document, reply_to_comment, and resolve_comment. The link stays the same, and the history shows each revision and its author.

4. Close it out

Once reviewers agree, ask for the final pass:

Change the status to Resolved, make sure every follow-up has an owner and a due date, and publish.

Link the final postmortem from your incident tracker. The permanent short link (offprint.io/d/…) keeps working even if the title changes.

Tips

• Draft fast, decide slowly. The agent's job is the first draft while details are fresh. The root cause and actions are decided in review.
• Keep follow-ups in a table with owner and due date, so they can be copied into your tracker.
• One project for all incidents. It becomes a searchable record you can share with new on-call engineers.

Related

• Example: Postmortem, checkout API latency spike
• All guides