Retrospective AI
The retro drafts itself from what actually happened.
Retrospective AI analyzes a sprint's real history (items, blockers, comments, metric signals) and drafts the retrospective before the meeting starts: clustered themes with evidence, a health score, and action items that are tracked to completion.
Running an analysis
Pick a sprint and run the analysis. The AI extracts themes from two kinds of signals: text clusters (what people wrote on items and in comments) and metric signals (what the numbers say: cycle-time spikes, blocked time, delivery rate). Analysis counts toward your monthly AI actions.
Themes with evidence
Themes are grouped into what went well and what needs improvement, each with sample quotes from the underlying items so the discussion stays grounded in what actually happened instead of what is loudest in memory.
Recurring themes
Themes that reappear sprint after sprint are flagged as recurring. This is where systemic problems stop hiding: a theme that survives three retros is a process issue, not bad luck.
Health score and delivery rate
Each sprint gets a 0–100 health score and a delivery rate (planned versus completed), so you can see across retros whether things are actually improving.
Action items
Capture agreed actions with an assignee and due date. Items can be pushed directly to Jira or Linear, are tracked to completion, and appear in each person's Assignments page. The next retro shows how many of the previous actions were actually completed.
Org memory
Every retro, theme and action is searchable afterwards. Before repeating a discussion, search the org memory. Chances are the team has been here before, and what was decided is on record.
Team pulse
A team pulse asks the same three quick questions every week or every two weeks, so you can see how the team is doing between retros. A project manager starts it from the Team pulse section at the bottom of the Retro page: keep the suggested questions or write your own, and pick weekly or every two weeks. It can be paused and edited later.
Answering the pulse
Everyone who can open the project sees a card on their dashboard until they have answered the current round: three answers from 1 to 5 and an optional comment. Answers are anonymous. Nobody, including admins, can see who answered what.
Reading pulse results
The Team pulse section shows the average for each question per round, colour-coded, and the latest comments. To protect anonymity, a round only shows averages and comments once at least three people have answered; until then it shows how many answers are in.