Internal Comms & Meetings

How to run a team retrospective with AI

Internal Comms & Meetings3 AI tools7 steps6 friction points

Want an agent to handle this on your live data? See the Starch version →

A team retrospective is the structured meeting where a team looks back at a recent sprint, project, or quarter — what went well, what didn't, and what to change next time. For most operators running small teams, retros sit in an awkward middle zone: important enough to schedule, easy enough to skip. Without a consistent format, they devolve into venting sessions or get canceled the moment the week gets busy.

The appeal of using AI here is obvious. A retro has a clear shape: gather input, surface patterns, generate action items, archive the output. That structure maps cleanly onto what language models do well — synthesizing scattered text, grouping themes, drafting summaries. It feels like exactly the kind of repetitive, format-heavy meeting prep that AI should be able to take off your plate.

ChatGPT, Claude, and Gemini can genuinely help with several parts of this workflow today. You can paste in raw team responses and ask for theme clustering. You can prompt for a structured summary with action items. You can generate retrospective question templates tailored to your team's context. Where they fall short is everything that requires memory, live data, or continuity across multiple retros.

Skip the manual work → See the Starch version

Internal Comms & Meetings3 AI tools7 steps6 friction points

AI walkthrough

How to do it with AI today

A practical walkthrough using ChatGPT, Claude, and other off-the-shelf LLMs — what they're good at, what you'll have to do by hand.

Tools that work for this

ChatGPTClaudeGemini

Step-by-step

1 Before the retro, use ChatGPT or Claude to generate your question set. Paste in a one-sentence description of the sprint or project and ask for 8–12 retrospective prompts across 'what went well,' 'what didn't,' and 'what to change.' Tweak the output to match your team's language.

2 Send those questions to your team asynchronously via Slack, Notion, or a Google Form. Give people 24–48 hours to respond in writing before the live meeting — this surfaces more honest input than going around the table in real time.

3 Collect all responses into a single text file or document. Paste the full set into Claude or ChatGPT with a prompt asking it to identify 3–5 recurring themes, flag the most actionable items, and note anything that appears to be a blocker.

4 During the live retro, use the AI-generated theme summary as your agenda. Run the meeting against those clusters rather than reading through every response. Use the AI output as a facilitator's guide, not a replacement for the conversation.

5 After the meeting, paste your notes or a rough transcript into the LLM and ask it to produce a structured summary: key decisions made, action items with owners, and open questions to revisit next retro.

6 Copy that output into whatever you're using for documentation — Notion, Google Docs, Confluence. Assign action items manually in your task tracker.

7 Next retro, repeat from step one. There's no memory between sessions, so you'll be starting fresh each time unless you manually paste in the previous summary for context.

Prompts you can copy

We just finished a 2-week product sprint. Generate 10 retrospective questions across 'what went well,' 'what slowed us down,' and 'what we'd do differently.' Keep questions concrete, not abstract.

Here are 8 team responses to our retrospective questions. Identify the top 3–4 themes, note any patterns that appear in multiple responses, and list the 5 most actionable items mentioned.

Turn these raw retro notes into a structured summary with three sections: key wins, key friction points, and action items. For each action item, suggest a likely owner based on the context in the notes.

We're a 6-person product team. Our last retro revealed recurring issues with unclear handoffs between design and engineering. Generate a focused set of retrospective questions to dig deeper into that specific problem this sprint.

Here's last retro's summary and this retro's raw responses. What themes carried over? What's new? What action items from last time appear to still be unresolved?

Reality check

Where this gets hard

The walkthrough above works — until your numbers change, the LLM hallucinates, or you have to re-paste everything next month.

No memory between retros — you have to manually paste the previous summary every time if you want continuity, and most people don't bother.

Action items live in the chat window. Moving them into your actual task tracker requires a manual copy-paste step that often gets skipped, so items disappear.

Theme clustering is only as good as what you paste in. If responses come in over Slack threads, Google Forms, and email, you're assembling that manually before the AI can touch it.

Output structure drifts between sessions. The clean format you prompted in week one won't look the same in week six unless you re-paste your exact instructions each time.

No searchable history. After six months of retros, there's no way to ask 'what has the team consistently flagged as a problem?' — every retro is an island.

Assigning action items to real people requires the LLM to guess from context. It doesn't know your org chart, who owns what, or who's already overloaded this week.

Tired of the friction?

Starch runs the whole workflow on live data — no copy-paste, no hallucinated numbers, no re-prompting next month.

See the Starch version →

Starch alternative

The same workflow on Starch

Starch is an agentic operating system — for this workflow, that means an agent builds a persistent retrospective system connected to your actual team tools, so retro summaries accumulate, action items land in your task tracker automatically, and nothing lives only in a chat window.

The Meeting Notes starter app transcribes your retro in real time, generates a structured summary with key decisions and action items after the call ends, and archives every meeting in a searchable history — so 'what did we decide last quarter?' has an actual answer.

Action items extracted from the retro summary flow directly into the Project Management app. Describe it once: 'when a retro ends, create tasks for each action item, assign them based on who was mentioned, and tag them as retro follow-ups.' Starch builds that automation.

Connect Notion or Slack from Starch's integration catalog, and the agent can pull team async responses directly from those tools — no manual copy-paste before the AI can process them.

Every retro summary is stored and searchable through the Knowledge Management app. Over time, you can query across all past retros to spot persistent patterns — recurring blockers, themes that never get resolved, items that keep reappearing.

Describe the recurring workflow in plain English — 'every two weeks, remind the team to submit retro responses, run the summary, create tasks from action items, and post a digest to our team Slack channel' — and Starch runs it on that schedule without you re-triggering it manually each sprint.

Get closed-beta access →

Toolkit