Module assessment

1.

In Fabrikam's multi-agent system, the orchestrator agent depends on the analysis agent's tool schema. After updating the analysis agent to add a new required output field, what must happen before the orchestrator agent proceeds through the pipeline?

The orchestrator agent's pipeline should proceed independently since the new field is additive and shouldn't affect existing functionality.

Contract testing must validate that the orchestrator agent can consume the analysis agent's updated schema, and the analysis agent must be deployed before the orchestrator agent.

The analysis agent pipeline should wait for the orchestrator agent pipeline to complete first, since the orchestrator initiates agent calls.

2.

Fabrikam's canary deployment of the new code review agent routes 10% of requests to the new version. After 24 hours, evaluation scores for the new version drop eight percent below the predeployment baseline. What should the pipeline do?

Continue the canary rollout while the team investigates—a single 24-hour window may not be statistically representative.

Stop the canary rollout and trigger rollback to the previous stable version, since the quality drop exceeds the defined threshold.

Switch immediately to 100% traffic on the new version to gather more data points for a statistically valid evaluation.

3.

After an automated rollback in Fabrikam's multi-agent system, only the security analysis agent was rolled back. The orchestrator and three other agents remain at their newly deployed versions. What determines whether this partial rollback is safe?

The rolled-back security agent version must still be compatible with the API contracts expected by the agents at their current versions.

All agents must be rolled back together to ensure the system returns to a fully consistent prior state.

A partial rollback is always safe as long as the agent being rolled back is not the orchestrator.

피드백