Draft with AI

Draft with AI, above the postmortem editor, writes every section nobody has written yet. It is a first draft for the people who were there to correct, not a postmortem written for them.

What it reads

Only what the person pressing the button can see in Jira:

• the work item's summary and description;
• the incident's severity and times, with the durations worked out by the app (the model is told to use them as given, not to calculate its own);
• the timeline, naming people by their Jira display name and marking the incident commander;
• the corrective actions already raised;
• the sections the team has already written, so the draft agrees with them.

It never sees email addresses or account ids, and nothing from a space or comment the person cannot see.

What it writes

• Only empty sections. Anything you have written is left exactly as it is, including a section that only lists @mentions.
• In the language of your Jira. Section headings stay as they are.
• Where the record does not say enough, a line like [TODO: how many support cases were raised] instead of a guess. A section of nothing but TODOs still counts as unwritten, so pressing the button again after the ticket says more fills it in.

The draft lands in the editor as an unsaved change. Read it, correct it, then save. It usually takes under a minute; you can keep working meanwhile.

Where it runs

On Forge LLMs: Claude models hosted by Atlassian. The incident does not leave Atlassian's platform, there is no third-party AI provider in between, and Afterbrief keeps nothing beyond the draft until someone saves it. This keeps the app eligible for Atlassian's Runs on Atlassian programme.

Plans and limits

A draft that fails before the model answers is not counted. Pressing the button again while a draft is being written picks up that draft rather than starting another.