The reading version
Tuesday afternoon. Sprint retro in fifteen minutes. Nobody is excited. The last three retros produced the same action items, and nobody followed up on any of them. Half the room talks too much, half says nothing, and you're in the middle — trying to synthesise everything in real time while also facilitating.
The retro falls apart because you're doing three jobs at once: running the conversation, listening for themes, and writing the artefact. AI takes the third job. The first two are still yours — and they're the ones that actually matter.
The pattern: collect, prompt, review
Collect the raw feedback however your team already does it (sticky notes, a Miro board, a shared doc). Paste it in along with last sprint's action items. The model groups, prioritises, and flags what never got resolved. You read the output and decide which themes the team actually wants to commit to — that's the judgment layer only someone in the room can add.
- Shape raw updates into sections
- Group wins by theme
- Enforce word count
- Draft the headline
- Decide what the single headline is
- Rank the blockers
- Name the risk you're worried about
- Sign your name to it