Finding the slow step inside an AgentCore agent session

MLA-C02 Operating, Monitoring, and Securing ML and AI Solutions Medium Question 4 of 30

An agent hosted on Amazon Bedrock AgentCore Runtime answers order questions by calling a foundation model and several tools. Some sessions take more than 40 seconds, and some end with incomplete answers. The ML engineer must find which step in each slow or failed session, a model call or a specific tool call, caused the delay or the error.

Which solution will meet these requirements?

Show answer and explanation

BInspect each session's execution path in AgentCore Observability traces and spans in CloudWatch

AgentCore Observability meets the requirements. It records each step of the agent workflow as traces and spans in Amazon CloudWatch, so the engineer can follow a session's execution path, see intermediate outputs, and find the model or tool call where time went or an error occurred. CloudTrail records who called an AgentCore API and when, not how long each step inside a session took. Model invocation logs show model inputs and outputs but not the tool calls between them. An error rate alarm reports that failures rose without showing which step failed.

Source: docs.aws.amazon.com/bedrock-agentcore/latest/devguide/obser…

Anonymous, no account needed. Your tally stays in this browser.

Next question Start 5 timed questions Browse all 30 original MLA-C02 practice questions