Train-the-trainer guide template for AI agents

Internal trainers need a guide that gives them the exact vocabulary, session outlines, the misconceptions learners bring, exercises with answer keys, and a clear statement of what trainers do not do, such as certify anyone. This template has twelve sections to copy and fill in, followed by a sample of three sections completed for a claims team.

Why do trainers need their own guide?

Train-the-trainer means preparing a few people to teach many; the idea is explained in what train-the-trainer is. With AI agents, the risk is drift. Each trainer explains depth or verdicts slightly differently, and within a few months the business has five versions of what "held" means. A guide fixes the words and the examples, so every session teaches the same thing.

It also protects trainers. Learners ask hard questions: will the agent replace me, who is to blame if I approve something wrong, can we turn off that refusal? A trainer without prepared answers improvises, and improvised answers spread. The guide gives them tested answers and tells them when to say "I will find out".

Use the guide for any team where internal trainers will deliver part of the training, alongside the formal tracks described in the guide to training teams for AI agents.

How do you fill it in?

The agent owner and the learning lead write it together, then a reviewer from the team checks every example against the live agent. Keep it short enough to read in an hour. Pair it with a one-page quick reference guide for learners. Set a review date: whenever the agent gains a capability or a depth, the guide must change before the next session. For the programme around the guide, see how to run a train-the-trainer programme for AI agents.

The guide template

1. Purpose and scope. Which agents and teams this guide covers.
2. Who may train. Trainers hold a current certificate at or above the level they teach; reviewers teaching review must themselves be certified reviewers.
3. Fixed vocabulary. One-line definitions trainers must use word for word: depth 1 to 5; allowed, held, refused, withheld; sealed, signed, declared, external; certificate lapse; the record.
4. The agent in this team. Capabilities, depths, bounds, red lines.
5. Session outlines. For each session: audience, length, objectives, timings, materials.
6. Key messages. Three to five points every learner must leave with.
7. Misconceptions and correct answers. The wrong beliefs learners bring, and the reply.
8. Demonstration cases. Real, anonymised held actions and refusals to walk through.
9. Exercises and answer keys. Cases with the correct verdict and the reason.
10. Hard questions and prepared answers. Including the ones the trainer should pass on.
11. What trainers do not do. No certification decisions; no promises about future depth; no changes to the agent.
12. Feedback and refresh. How trainers report gaps, and when the guide is reviewed.

A sample: sections 6, 7 and 11 for a claims team

An illustrative insurer prepares three claims team leads as trainers for 40 handlers.

Section 6, key messages - The agent drafts settlements and pays small claims within bounds; every decline is held for you. - Depth is earned on our own past claims and can drop the same day a certification run fails. - Your approval is recorded with your name. Read the evidence, not just the draft.

Section 7, misconceptions and correct answers - "If the agent drafted it, it has been checked." Correct answer: the draft is the agent's work; checking is yours. - "We can ask for the refusal to be switched off." Correct answer: if the refusal comes from our own limits, the owner can review those limits; if the action is withheld, it is a platform decision that applies to every agent at every depth, and nobody here can change it. - "Once the agent is certified it stays certified." Correct answer: its certificate lapses after 90 days unless renewed, and drops at once on a failed run.

Section 11, what trainers do not do - Do not tell a handler they are ready to approve; only a completed assessment and sign-off does that. - Do not promise that the agent will pay larger claims soon. - Do not edit the agent's brief in a session, even to fix an obvious error; report it to the owner.

How often trainers themselves need refreshing is covered in how often staff should refresh AI agent training.

Where the Academy fits

Pryme Intelligence Academy can carry the formal part, leaving internal trainers to run the team-specific sessions in this guide. Lessons, labs and tests are free, every lab runs in a free Build workspace, and module tests are set in each learner's industry and role. Trainers teaching reviewers should hold a current Operator certificate themselves.