Frameworks & Templates4 min read

Keep an AI Practice Library Current: Repeat, Rotate, Stretch, Retire

Maintain AI practice as a portfolio: repeat core behaviours, rotate familiar examples, stretch one condition and retire obsolete work.

Bokili EditorialPublished October 23, 2026· Verified October 23, 2026

Four trays cycle AI practice cards through repeat, rotate, stretch and retire decisions

An AI practice library goes stale in two different ways. Some exercises stop matching the tools, rules or source material people use. Others remain technically correct but become too familiar to reveal whether the skill transfers. The answer is not to replace everything each quarter. Treat the library as a portfolio: repeat the behaviours that must become automatic, rotate the surface details, stretch one constraint at a time, and retire practice that no longer represents real work.

A library is not a warehouse

Every practice item should have a maintenance decision: repeat, rotate, stretch or retire.

Why AI practice needs active maintenance

The UK AI Skills Framework separates technical, responsible and ethical, and non-technical skills, and adapts them to job level. Its adoption pathway also connects organisational maturity with changing skill needs. Skills England’s evidence review distinguishes formal training from employer-led practice linked to roles and tasks. Together, those ideas point to a practical consequence: a course catalogue cannot be considered current merely because its pages still load.

AI practice should change when the task changes, when the tool or policy changes, or when learners can pass by remembering the example. But constant replacement creates another problem: teams lose the stable practice needed to compare progress. The maintenance decision must preserve the skill while changing only what has genuinely moved.

Repeat, Rotate, Stretch, Retire

Repeat

Keep the task and conditions stable when the behaviour must become reliable under normal work pressure.

Rotate

Keep the same skill and standard, but change the example, source pack or surface details so recall cannot masquerade as transfer.

Stretch

Add one harder condition: less time, incomplete evidence, a new audience or a higher consequence. Keep the acceptance standard visible.

Retire

Remove or archive practice when the tool, policy, workflow or decision boundary no longer reflects approved work.

Worked example: maintain a customer-support practice set

A support team has eight exercises for drafting customer replies with an approved AI tool. One exercise teaches staff to confirm the customer’s request before drafting. It still matters and recent work shows the behaviour is inconsistent: repeat it. Three exercises use the same delayed-delivery example: keep the acceptance standard but rotate the case details. A complaint exercise is passed too easily: stretch it by removing one source and requiring an escalation sentence. Two exercises refer to a refund policy that has changed: retire them until the source pack is rebuilt.

Reading is a start. Practice makes it stick.

Start learning

Run a monthly portfolio review

1
Name the skill
Write the observable decision or behaviour the exercise is meant to practise.
2
Check the conditions
Confirm the tool, rule, source pack, role and acceptance standard still match approved work.
3
Inspect the evidence
Look for repeated errors, effortless passes, policy drift and requests for help.
4
Choose one action
Mark the item repeat, rotate, stretch or retire, with an owner and next review date.
5
Test the revision
Give the changed exercise to a small group and check that it still measures the intended skill.

	Weak maintenance signal	Useful maintenance signal
Usage	People completed it	People made the intended decision on a fresh case
Difficulty	Average score fell	The same mistake recurred under a named condition
Freshness	The article date is old	The tool, source, policy or workflow changed
Revision	Rewrite the whole exercise	Change only the condition that invalidated the evidence

Keep a small maintenance record

For each exercise, store five fields: target behaviour, approved conditions, last evidence reviewed, maintenance action and next review date. Do not turn this into a large learning-content project. The record exists to make one decision visible and reversible. Bokili’s guide to changing one constraint at a time explains how to preserve diagnostic value. AI skills have a shelf life explains why review dates should follow changes in work, not an arbitrary annual cycle.

Portfolio maintenance check

The exercise names one observable skill or decision.
The tool, rule and source pack still match approved work.
The learner cannot pass by memorising the example.
A harder version changes only one meaningful condition.
Retired items are archived with the reason, not silently deleted.
Every action has an owner and next review date.
The revised item is tested on a fresh case.

Review four exercises in ten minutes

Choose four practice items used by one team.
Write the target behaviour beside each item.
Mark each one Repeat, Rotate, Stretch or Retire.
Add one sentence explaining the evidence for the decision.
Assign one small change and one review date.

A current library protects the signal

The goal is not a permanently new catalogue. It is trustworthy evidence that people can still perform the work expected of them. Repeat what should become dependable. Rotate what has become familiar. Stretch what no longer distinguishes capability. Retire what has stopped being true. Bokili’s short, role-based missions make that maintenance practical because each item can stay focused on one behaviour and one work condition.