Module 14 · Lesson 05

Continuous Improvement — Retrospectives for AI Use

Reading time: 16 minutes Track: Claude Fluency for Teams · Lead/Manager path

The plateau problem

Most teams show rapid improvement in Claude use in the first 1-2 months. Then they plateau. They're using Claude for the same tasks in the same ways they figured out early on, not discovering new high-value applications or improving their prompting practices.

Structured continuous improvement prevents this plateau. The investment is small — 30 minutes per month — and the compounding effect over 12 months is significant.

The monthly retrospective (30 minutes)

Run this as a team meeting once a month. Keep it tight.

Part 1: Wins (10 minutes)

Each person shares one specific example of high-value Claude use in the past month. Not "Claude helped with emails" but "I used Claude to synthesize the five vendor proposals in 30 minutes instead of half a day — here's the prompt I used."

Capture these. The best ones go into the shared prompt library.

Part 2: Struggles (10 minutes)

Each person shares one place Claude underdelivered or created friction. What was the task? What went wrong? Is this a prompt quality issue, a wrong-tool issue, or a genuine Claude limitation?

Distinguish between:

• "I wrote a bad prompt" → fixable with better prompting
• "This task needs a different tool" → update conventions
• "Claude genuinely can't do this well" → update expectations

Part 3: Experiments to try (10 minutes)

Pick 2-3 things to try in the next month based on the discussion:

• A new task type to try with Claude
• An existing prompt to improve
• A convention to update

Assign an owner for each experiment. Review results at the next retro.

The prompt library review

Every quarter (not monthly — this takes more time):

• Review every prompt in the library
• Mark ones that are being used
• Archive ones unused for 3+ months
• Update ones where the prompt could be improved based on what you've learned
• Add the best new prompts from the monthly retrospectives

Staying current as Claude improves

Claude capabilities evolve. Tasks that weren't worth trying six months ago may be tractable now. Build in a quarterly "try it again" habit for 2-3 tasks you've previously written off.

Good candidates for "try it again" experiments:

• Tasks where output quality felt insufficient months ago
• More complex analytical tasks
• Tasks requiring tool use (if you've expanded your MCP setup)

The model you're using now is more capable than the one from six months ago. Your conventions should reflect that.

The prompt post-mortem

When a significant task produces poor output:

Run a 15-minute post-mortem before moving on:

1. What was the task?
2. What was wrong with the output?
3. What failure mode was this? (Use the framework from Module 11 Lesson 5)
4. What would a better prompt have included?
5. Should this become a convention update or a prompt library entry?

Post-mortems feel like overhead until you do one and realize you've written a prompt that saves the whole team an hour a week.

Building the habit

The retrospective habit is most fragile in month 2-3, when initial enthusiasm has faded but the compounding benefit isn't yet visible. Protect this time:

• Put it on the calendar as recurring
• Keep it to 30 minutes — it will collapse if it runs long
• Make the wins section first — it builds motivation for the rest

Teams that maintain a monthly retrospective for 12 months consistently report that their AI practices at month 12 look nothing like month 1 — not because Claude changed, but because their use of it compounded.