How do I test durable agent workflows that use LLM calls?

You can validate LLM-integrated durable agent workflows using Diagrid Catalyst’s production-grade, purpose-built replayable test harnesses tailored for both deterministic and probabilistic testing. Isolate external LLM calls with schema-aligned mock responses to avoid real API interactions during local or staging testing, then replay core workflow steps to confirm predictable state transitions without unintended side effects. Take care to match mock response schemas and expected behavior closely, as mismatched data can disrupt replay validation checks.