15 articles
The AI testing tool landscape sorted into five jobs — spec-driven generation, traffic recording, self-healing UI, authoring assistants and LLM evals — with what each is genuinely good at and what to verify in a trial.
How to gate AI-assisted development — why line coverage stops meaning anything, the four checks that hold, and where human review should be spent when volume outruns it.
What agentic QA actually means, where it works today, the four questions that separate a usable tool from a demo, and how to put one in a pipeline without losing determinism.
How to build an LLM eval suite that actually catches regressions — dataset design, scoring functions, LLM-as-judge calibration, thresholds, CI gating and the tooling landscape.
A practical method for testing LLM-backed features — deterministic shell versus probabilistic core, invariant tests, golden-set evals, regression on prompt changes, and what to run in CI.
The MCP threat model and the test cases that cover it — tool poisoning, confused deputy, token passthrough, session hijacking, DNS rebinding and excessive agency, with runnable assertions for each.
MCP versus REST from a testing point of view — protocol, contract discovery, caller behaviour, error channels, security surface and what your existing API test skills carry over unchanged.
A practical guide to testing Model Context Protocol servers — the handshake, tools/list and tools/call, schema assertions, error mapping, transport differences, and the security cases unique to MCP.
How enterprises test APIs and applications backed by non-deterministic AI models. Property-based assertions, statistical thresholds, golden-set evaluation, and the governance pattern that satisfies audit without forcing the model to be deterministic.
