What is spaced repetition?

Spaced repetition is a learning technique in which material is revisited at increasing intervals, so that each review comes just as the memory starts to fade. For people overseeing AI agents, it keeps the rarely used rules, such as what a withheld action means or why an external document cannot authorise a payment, ready for the day they are needed.

Why does spaced repetition matter here?

Agent review is mostly routine. A reviewer may clear hundreds of ordinary drafts between two cases that test what they learned about red lines. Memory does not care that the rare case is the important one; it keeps what is used and drops what is not. One round of training, however good, decays fastest on exactly the points that carry the most risk.

Spaced repetition fights that decay cheaply. A few minutes every week or two, on the right questions, does more than a full day of retraining a year later. The article on what people forget soonest after agent training lists the usual suspects.

How does it work in practice?

The method is simple; the discipline is in keeping it going:

1. Pick the items: short questions or cases on points that are rare and costly.
2. Schedule reviews: first after a day or two, then a week, then a few weeks, then longer.
3. Adjust by result: a question answered wrongly comes back sooner; one answered well waits longer.
4. Keep it short: two to five items per session, a few minutes at most.
5. Refresh the pool: when the agent changes, retire old items and add new ones.

Items work best as small scenarios rather than definitions. "A held payment is backed by an emailed invoice; the amount matches. Approve, send back or escalate?" tests the judgement, not the vocabulary. That links spaced repetition to scenario-based learning.

An example

A collections team trains reviewers on an agent that drafts payment-plan offers. After initial training, each reviewer gets three questions every Monday. Week one covers the customer's limits and hardship flags. Three weeks later, the hardship question returns for anyone who got it wrong. By the third month, the questions most often missed are about what happens when the agent's certificate lapses and it drops to depth 1. The team lead adds a short session on that, and the misses stop.

How is spaced repetition different from refresher training?

Refresher training is a scheduled session that brings people back up to standard, often on what has changed. Spaced repetition is a continuous technique, applied to individual items at intervals set by each person's results. A refresher might happen quarterly for the whole team; spaced repetition happens weekly, per person, in minutes.

It also differs from microlearning, which describes the size of a unit. Spaced repetition describes when units return. The two combine well.

Which terms are related?

• Retention check: a test that shows what has faded.
• Microlearning: short units well suited to spaced review.
• Training transfer: whether learning shows up in real work.

The case for small, frequent effort is made in the cheapest control in an agent deployment is a trained person, and what makes agent training stick puts the technique in context. For the full approach, read training teams for AI agents.