How do I write effective AI agent instructions?

Last updated: November 5, 2026

Effective AI agent instructions give the agent a clear role, workflow, rules, and fallback paths. Well-structured instructions make agent behavior more predictable across messaging and calls.

Before you start

  • Decide the agent’s purpose, audience, and main workflow.

  • List the intents the agent should handle.

  • Identify when the agent should use knowledge, tools, or escalation.

Steps

  1. Open the AI agent you want to configure.

  2. Go to the instructions section.

  3. Set the beginning message in its dedicated field if the agent uses calls.

  4. Define the agent’s identity, tone, and communication style.

  5. Write one or two sentences describing the agent’s objective.

  6. Add a numbered primary workflow.

  7. Add intent handling sections for common scenarios.

  8. Add tool and knowledge usage guidance.

  9. Add rules and guardrails.

  10. Add fallback and escalation logic.

  11. Add call ending rules when the agent handles calls.

  12. Save and test the agent in preview.

Tips

  • Use clear headings and numbered lists.

  • Keep each instruction focused on one action.

  • Use explicit language such as “always,” “never,” and “only when” for guardrails.

  • Define what the agent should do when it is confused, interrupted, or missing information.

  • Test frequently and simplify instructions if behavior becomes inconsistent.

Troubleshooting

  • The agent skips steps: Rewrite the workflow as a numbered sequence.

  • The agent hallucinates: Add guardrails that tell it not to invent information.

  • The agent ignores tools: Add a tool usage section with trigger conditions.

  • The agent gives inconsistent answers: Remove conflicting instructions and clarify priority order.