Notes from the field on test automation, agentic QA, and shipping quality software.
Shipping an LLM feature without eval gates is shipping blind. How to turn eval scores into CI blockers before a bad model change reaches production.
Consuming MCPs gets an agent started quickly. Building one that exposes your test framework, fixtures, and CI signals is where real authority begins.
Production data in test environments is a liability. How to use LLMs to generate realistic test fixtures at scale without touching real customer records.
Load testing REST APIs with JMeter: modeling realistic load, the metrics that matter, assertions that mean something, and CI integration.
Pre-release testing finds what you thought to look for. Production telemetry finds what actually breaks. How to build a feedback loop between the two.
When an agent does the wrong thing, it looks correct from outside. How to trace why your agent made that call, with concrete tooling and span design.
A grounded look at the Level 1 to Level 3 shift in test automation and what it means for people who write tests for a living.
A single QA agent that does everything is a generalist that does nothing well. How to design specialist agents that divide and conquer test coverage.
When the thing under test is a model, pass/fail stops working. How eval-driven development brings real quality control to AI features.
The EU AI Act is not just a legal question. High-risk systems must demonstrate adversarial robustness by August 2026, and that is a QA problem to solve.