Agents
What documentation does an AI agent need to understand your business?
By Julián Medina · Director of SourcingUp and creator of CommerceUp
An agent needs current rules, shared definitions, named owners and examples of past decisions. Useful documentation explains what to do in a specific situation and makes it possible to distinguish current instructions from historical context.
Organize by decisions
Hundreds of files do not guarantee useful context. Start with questions that arise at work: which price to use, what a status means or who approves an exception. Identify the main source for each answer. When documents conflict, give the agent a clear rule for stopping or asking for help.
Include examples and the reasoning behind them
A past decision is more useful when you preserve both the reason and the conditions that made it appropriate. For example, a delivery exception for one customer should not become a general promise. Documenting its scope keeps a special case from becoming permanent policy.
Assign responsibility for keeping it current
Each area should know what documentation it is responsible for reviewing and what event requires an update. A price change, a new tool or a modified policy are specific triggers. Then try questions affected by that change; updating a file does not prove that the agent is retrieving the correct version.
Put it into practice
- Identify main sources by topic.
- Identify the owner, scope and effective date.
- Keep the reason for the exceptions.
- Test information retrieval after updates.
The thinking behind this guide
A guide by Julián Medina. The scenarios are illustrative, not measured customer outcomes. My work at SourcingUp.