Decagon Product Foundations

Launching a new workflow without testing is risky and even a small failure can derail the customer experience. That’s why testing is essential. It ensures your Decagon agent behaves as intended before customers ever see it, and QA helps you keep improving once it’s live.

Testing begins with a core set of baseline cases that cover your most important workflows. Typically, these fall into four categories:

1. Brand and tone — making sure every response matches your voice guidelines.
2. Escalation guardrails — verifying the agent hands off to humans in the right scenarios.
3. Knowledge retrieval — confirming accurate answers from your knowledge base.
4. AOP logic execution — checking that workflows run as designed, in line with your policies.

Decagon provides a testing suite for running these checks consistently. Tests can be created from scratch or directly from real customer conversations, which automatically carry over metadata and tool outputs. Within each test, you add assertions, specific checks like whether the correct AOP was triggered, a tool was called, or the response content matched expectations.

To simulate edge cases safely, you can mock tool outputs. For example, instead of triggering a real refund, you can mock “refund successful” or “refund declined” and confirm the agent responds correctly in both situations. Running tests multiple times also matters, since LLM outputs can vary.

Testing doesn’t stop at deployment. QA in production helps catch issues that structured tests may miss. Teams can manually review conversations in Decagon, customize QA fields, and over time, turn recurring patterns into reusable tests. 

By maintaining a robust test suite and QA process, you can confidently roll out new workflows, knowing your AI agent will perform reliably across scenarios.