Team Sprint Retro Kit

Pre-Retro: Data to Gather

Before the meeting, collect:

| Metric | Source | Target | |--------|--------|--------| | Tasks completed with AI | Sprint board | All AI-tagged items | | Estimated vs. actual time | Ticket history | Δ calculation | | Review cycles per task | PR/doc comments | Compare AI vs. manual | | TLX scores (if collected) | Survey/Slack | Team average |

The 5 Retro Questions

Use these verbatim. They're designed to surface friction that teams avoid discussing.

1. "Which AI-assisted task took longer than expected?"

What you're looking for: Tasks where generation was fast but review/revision was slow.

Follow-up: "How much of the total time was fixing AI output versus your own work?"

Action if common: Add review time estimates to AI task planning.

2. "Where did AI output require the most editing?"

What you're looking for: Patterns in AI failure modes (formatting, accuracy, tone, completeness).

Follow-up: "Could a better prompt have prevented this, or is this task unsuited for AI?"

Action if common: Build prompt templates for high-edit tasks or reclassify them as manual.

3. "Did anyone feel more mentally drained using AI than expected?"

What you're looking for: Hidden cognitive load—checking output, maintaining context, managing prompts.

Follow-up: "What specifically caused the fatigue?" (Switching contexts? Verifying accuracy? Prompt iteration?)

Action if common: Implement micro-TLX checks after AI tasks.

4. "Which task should we stop using AI for?"

What you're looking for: Honest admission that some AI applications aren't working.

Follow-up: "What made you realize it? Time, quality, or frustration?"

Action if raised: Remove from AI workflow immediately. Test again in 2 sprints with better prompts.

5. "What's one AI use case we should expand?"

What you're looking for: Genuine wins that can scale.

Follow-up: "What made it work well? Can we template that approach?"

Action if raised: Document the pattern. Add to team playbook.

Tracker Fields to Copy

Add these to your sprint board or project tracker:

AI_ASSISTED: [Yes/No]
ESTIMATED_TIME_MIN: [number]
ACTUAL_TIME_MIN: [number]
REVIEW_CYCLES: [number]
DELTA_PERCENT: [calculated: (estimated - actual) / estimated * 100]
TLX_SCORE: [0-100, optional]
AI_NOTES: [free text - what worked/didn't]

Sample Ticket Entry

Task: Draft Q4 planning doc
AI_ASSISTED: Yes
ESTIMATED_TIME_MIN: 30
ACTUAL_TIME_MIN: 55
REVIEW_CYCLES: 3
DELTA_PERCENT: -83% (took 83% longer than expected)
TLX_SCORE: 72
AI_NOTES: AI draft missed key context from Q3.
          Had to rewrite intro and add 4 missing sections.
          Prompt: "Draft Q4 planning doc based on Q3 template"

Retro Template (Copy to Miro/FigJam)

┌─────────────────────────────────────────────────────────┐
│  AI SPRINT RETRO - [Sprint Name]                        │
│  Date: ___________  Facilitator: ___________            │
├─────────────────────────────────────────────────────────┤
│                                                         │
│  📊 THIS SPRINT'S DATA                                  │
│  ┌────────────────────┬────────────────────┐           │
│  │ Tasks with AI      │ _____              │           │
│  │ Avg Δ              │ _____%             │           │
│  │ Avg TLX            │ _____              │           │
│  │ Time saved (claimed)│ _____ hrs         │           │
│  │ Time saved (actual) │ _____ hrs         │           │
│  └────────────────────┴────────────────────┘           │
│                                                         │
│  ✅ KEEP DOING (AI wins)                               │
│  ┌─────────────────────────────────────────┐           │
│  │                                         │           │
│  │                                         │           │
│  └─────────────────────────────────────────┘           │
│                                                         │
│  🛑 STOP DOING (AI friction)                           │
│  ┌─────────────────────────────────────────┐           │
│  │                                         │           │
│  │                                         │           │
│  └─────────────────────────────────────────┘           │
│                                                         │
│  🧪 TRY NEXT SPRINT                                    │
│  ┌─────────────────────────────────────────┐           │
│  │ Task: _______________ Owner: __________ │           │
│  │ Hypothesis: __________________________ │           │
│  │ Success metric: ______________________ │           │
│  └─────────────────────────────────────────┘           │
│                                                         │
│  ⚠️ ACTION ITEMS                                       │
│  □ _______________ Owner: _____ Due: _____            │
│  □ _______________ Owner: _____ Due: _____            │
│  □ _______________ Owner: _____ Due: _____            │
│                                                         │
└─────────────────────────────────────────────────────────┘

Warning Signs to Watch

• ✓
• Δ trending up over 3+ sprints — Team is getting less calibrated, not more
• TLX above 65 for multiple team members — Burnout risk, even if velocity looks good
• Review cycles increasing — AI is generating more work for reviewers
• "We don't have time to track" — Measurement debt is accumulating
• Same person always editing AI output — Hidden bottleneck forming

Apply Now

Team Resources

• Delta Logging in Sprints — Full guide to Δ tracking in agile workflows
• Micro-TLX Fatigue Check — The 2-slider protocol for workload monitoring
• When AI Slows You Down — Diagnostic framework for negative ROI tasks
• Fair Trial Protocol — How to run valid AI vs. manual comparisons