What is ai agent?

Why it matters

The shift from deflection to resolution is the whole commercial story of AI in support. An agent that can actually close an issue changes the unit economics of a support team, because the marginal cost of the next routine answer approaches zero while quality stays constant at three in the morning.

What to know

• The quality ceiling is set by your content. An agent grounded in stale documentation gives confident stale answers.
• Action-taking, such as issuing a refund or changing a plan, needs explicit permission boundaries and an audit trail.
• Escalation design matters more than model choice. A clean handoff with full context is what customers actually notice.
• Agents should be evaluated on resolution rate and reopen rate together, never on deflection.
• Coverage of your worst-documented topics is where real deployments break, not on the easy questions in the demo.

An example

Example: a customer asks whether their plan includes single sign-on and how to enable it. The agent retrieves the plan matrix and the setup guide, confirms the plan does include it, and returns the exact three steps. No ticket is created and nothing is escalated.

Common mistakes

• Launching before auditing the knowledge the agent will answer from
• Giving action permissions without limits or logging
• Measuring success on deflection instead of resolution and reopens
• Treating the escalation path as an afterthought

Questions