---
url: /training/ai-trainer.md
description: >-
  How to use an AI assistant as a trainer for new advisors and FDEs, with these
  docs as its source.
---

# AI trainer

A new hire should be able to connect their AI and say **"Train me how to use this."** The AI then walks them through the [training path](index.md) one module at a time: it explains, shows, sets a task, and checks the answer.

## What works today

These docs are plain Markdown, so any AI assistant can read all of them. You can run a trainer now with a prompt.

1. Open Claude, ChatGPT or Claude Code.
2. Give it the whole site in one file: `llms-full.txt`. See [AI-readable docs](../ai/index.md).
3. Paste this prompt:

```text
You are the DutyGraph trainer. Your only source is the DutyGraph docs I gave you.

Train me to run a DutyGraph engagement as an advisor. Follow the Training path:
eight modules, in order. For each module:
1. Tell me in a few sentences what I will learn and why it matters.
2. Teach it in small steps. Ask me to do the "Do" task in the app and tell you what I see.
3. Ask the module's "Check" question. If my answer is weak, say what is missing and ask again.
4. Give me a one-line summary and ask if I want to go on.

Rules:
- One step at a time. Wait for my answer.
- If the docs label something Planned or In review, tell me it is not available yet.
- If the docs do not answer my question, say so. Do not guess.
- Use the words in "Words we use".

Start by asking what my role is and which modules I have done.
```

With Claude Code and the CLI signed in, the trainer can also look at the sample company with you and check your work against what is really there.

## What is planned

Planned

| Part | What it adds |
| --- | --- |
| **A "Train me" tool on the MCP server** | The trainer starts when a new advisor connects and says "Train me". No pasting. |
| **Progress that is saved** | Which modules a person finished, kept in DutyGraph, so they can stop and resume like the Task Card interview |
| **Checks against a sample company** | The trainer verifies the "Do" tasks, for example that the person really made a Task Card |
| **A trainer persona** | Built the same way as a [role AI](../operate/role-ai.md), from the advisor role profile |
| **Teaching pages** | The trainer makes a small web page for a module, with a diagram or a walkthrough, using the same pattern as [analysis pages](../playbooks/analysis-pages.md) |
| **Sign-off** | A manager sees who passed which module |

## Why the docs come first

A trainer is only as good as what it reads. That is the reason every page here is Markdown, every feature carries a status, and the handbook, the manual and the reference live in one place. When a page is wrong, fix the page and every trainer is fixed with it.

## The same pattern for a client

The trainer for DutyGraph and the [role AI](../operate/role-ai.md) for a client's new hire are the same idea with different sources:

| | Trains | Reads |
| --- | --- | --- |
| **DutyGraph trainer** | A new advisor or FDE | These docs |
| **Role AI in training mode** | A client's new employee | The role profile and the role's confirmed Task Cards |
