This browser is no longer supported.
Upgrade to Microsoft Edge to take advantage of the latest features, security updates, and technical support.
A team is designing a customer service agent. They want to measure its effectiveness, but they haven't yet tracked how long their current manual process takes or how often responses require revision. What should they do before finalizing success metrics in the design?
Document the metrics they want the agent to achieve and establish baselines during the first week of production.
Skip formal baselines and use the agent's first-month performance as the initial benchmark.
Measure the current manual process before finalizing the metrics, so each success criterion has a corresponding before-state value.
An agent's workflow includes a step that receives a ticket category ID and writes a routing record to a business system. The routing rule is explicit: 'shipping' maps to the shipping queue, and 'returns' maps to the returns queue. Should this step use agentic or deterministic logic?
Agentic logic, because the agent needs to interpret the category ID and decide where to route the ticket.
Deterministic logic, because the routing decision follows explicit rules with no need for AI judgment.
Agentic logic, because all agent workflow steps must use AI to maintain consistency across the agent's behavior.
An agent monitors incoming support requests, classifies each request into a category, and drafts a proposed response — all without a user initiating each step. It surfaces its work for human review and approval before any response is sent. Which interaction pattern best describes this agent?
Conversational
Autonomous
The agent spans both patterns and can't be classified as either.
A knowledge agent retrieves policy documents and returns excerpts to a triage agent via A2A. A security review finds that certain documents contain embedded instructions designed to redirect the triage agent's behavior when processed. At which intervention point in the agent's processing pipeline should the team place a control to address this risk?
User input
Tool call
Tool response
An agent runs without a user present to categorize incoming requests and send escalation emails from a shared mailbox. When planning this agent's identity, which configuration is required?
An interactive pattern with delegated permissions scoped to an admin account.
An autonomous pattern with application permissions, plus an agent's user account.
An autonomous pattern with application permissions only.
You must answer all questions before checking your work.
Was this page helpful?
Need help with this topic?
Want to try using Ask Learn to clarify or guide you through this topic?