A platform team wants CI for a multi-agent repo. Engineers currently only run occasional manual chat trials. Which testing strategy should you specify for integration and release?
Select an answer to reveal the explanation.
Short Explanation
You want the full testing mix: unit tests for tools and parsers, integration tests for MCP and orchestrator paths, regression suites, and automated agent-quality evaluations before deployment. Quarterly manual chat won't catch regressions, unrelated admin-portal CSS won't test agents, and non-determinism is handled with eval thresholds, seeds, and statistical gates—not by banning tests.
Full Explanation
Correct Answer — combined testing strategy
Multi-agent testing includes unit tests, regression tests, integration tests, and automated evaluations in CI/CD. This covers tools, parsers, MCP and orchestrator paths, and quality regressions before deployment.
Why quarterly manual chat is wrong: It misses regressions and does not scale.
Why unrelated admin-portal CSS is wrong: It does not validate agent behavior.
Why banning automated tests is wrong: Non-determinism requires eval thresholds, seeds, and statistical gates, not zero testing.
Exam tip: Unit + integration + regression + automated agent evals in CI/CD.