Skip to content
Release: Australia · Updated: 2026-03-12 · Official documentation · View source

Guidelines for writing AI agent instructions

Review the guidelines to write affective instructions for the AI agents to complete your Agentic Playbooks activity.

GuidelineDescription
Provide step-by-step instructionsBreak down complex operations into individual steps.If you need to perform operations like count, high, or low, write each operation out, step by step.
Verify that all columns can be queriedCheck the Knowledge Graph to confirm that the columns can be queried
Use specific column names- Make sure column names are correct in your instruction. - If agent doesn't understand the column value that you're giving, try to give an example. For example, if an agent returns a sys\_id instead of a number, you could tell the agent to Retrieve Incident numbers like INC000081.
Write conversationallyWrite instructions in plain English.
Use meaningful names and labels for activities and other nodesRename tables and other nodes for clarity. For example, if a table named abcd contains employee data, rename it to Employees.Note: This is especially important if you are using custom Knowledge Graph for the workflow.
Test and revise instructionsTest your instructions and check for accurate AI agent behavior for your use cases.
Emphasize critical stepsHighlight important steps to ensure they are not skipped. Use keywords like Mandatory, Required, or Important.
Specify query operationsUsing the keyword Query helps our data gathering AI agent to pick the more effective Knowledge Graph over the AI search.
Custom agentsTo maintain accuracy, add a maximum of 5-6 custom agents to an activity.