Decagon

Choose a narrow support workflow and connect the knowledge sources and backend actions needed to resolve it. Define policies for what the agent may answer or change, which customer information it can access, and when it must escalate. Test with historical or synthetic support cases before exposing the agent to live customers. Review resolution quality, tool calls, and failure modes—not just response tone. Expand to additional workflows only after the first one is stable, and maintain human review for refunds, account security, regulated requests, or other high-impact actions.

For a first evaluation, keep the scope small enough that you can compare the AI-assisted result with a known baseline. Record the configuration, model or workflow choices that produced the result so successful tests can be reproduced. If the product connects to external systems or private data, grant the minimum permissions required and review the provider's current security and data-handling documentation before expanding access.