Millan Kaul

less than 1 minute read

Written by - Millan Kaul

The AI answer looked great. It still failed the test.

Image showing LLM outputs with some being not compliant with rules
A beautiful AI answer can still break production.

What I did today:

I checked whether every response returned the required JSON field. I ignored whether the wording was impressive and focused only on the contract.

What surprised me:

Several responses were useful to a human but unusable by the application.

My QA takeaway:

  1. For AI systems, product quality includes format, schema, and integration—not just smart words.
  2. A beautiful AI answer can still break production.

What “boring” requirement has caused your team the most pain?

#AIEvals #APITesting #AIEngineering

You may also enjoy

Five examples are enough to start - day 04

less than 1 minute read

You don’t need thousands of test cases to begin. Five carefully chosen examples are more than enough to boot up your AI evaluation pipeline.

I tested a longer prompt - day 03

less than 1 minute read

Increasing the complexity of prompt inputs to see how the LLM judge responds and analyzing its limits when prompt length grows.

My first AI eval was boring - day 02

less than 1 minute read

Setting up a basic LLM evaluation only to find the initial results predictable, but discovering why simple tests are the best place to start.