Start with one agent, not a fleet
Give the learner a clear training mode before assigning a job. Add real knowledge in small, inspectable files instead of hiding everything inside one enormous prompt.
Practical agent training
How to train an AI agent with your own knowledge
Turn documents, examples, corrections and decisions into files an agent can inspect, cite and carry into its next project.
Process before you trust
Let the agent inspect the new material as a set. Conflicts, missing authority and unclear exceptions should become questions rather than invented answers.
Approve durable memory
Teacher explanations become visible memory and proposed rules. Only reviewed files become active; generated work remains a draft until a human approves it.
The practical rule