The Test Pyramid Survives. Assertions Don't: Testing AI in the Real World

Join us at Code Talent for our Local Expert speaker series featuring Elliot Sequeira.

Every engineer knows the testing pyramid: a wide base of fast unit tests, fewer integration tests above, a handful of end-to-end tests at the top. It has outlasted every major shift in how we build software.

Then a language model enters the system, and exactly one piece of it stops working. Not the shape — the assertion. assert output == expected has no meaning when the output is a paragraph.

Join Elliot for a session on that transition: what the testing pyramid becomes when the system under test is both deterministic and non-deterministic, and why LLM-as-a-judge is the one assertion primitive cheap enough to run at every layer of it. We'll walk the pyramid three times — the classic version, the AI version with a new foundation band underneath it, and the version with a judge operating at every band — then get into the pipeline and the data engineering that make it real.