New: the QA x AI radar, the glossary and 8 ready-to-copy QA prompts.See the radar →

scan('radar')

QA x AI radar

Testing tools and techniques in the age of AI, sorted into four rings: adopt, trial, assess, hold. Opinionated, based on real projects, updated every quarter.

Autumn 2026 edition Next update: January 2027

QA x AI radar: 17 tools and techniques across 4 quadrants and 4 ringsAdoptTrialAssessHoldTechniquesToolsAI for testingTesting AI1. Test pyramid (Adopt)12. Contract testing (Pact) (Adopt)23. Test impact analysis (Trial)34. Mutation testing (Assess)45. Fixed waits (sleep) (Hold)56. k6 (Adopt)67. Playwright (Adopt)78. AI-assisted visual testing (Assess)89. Record and replay without maintenance (Hold)910. AI-generated unit tests (Trial)1011. Playwright MCP (Trial)1112. AI merge request review (Assess)1213. Self-healing selectors (Assess)1314. AI-generated tests without review (Hold)1415. LLM-as-a-judge (Trial)1516. promptfoo (Trial)1617. LLM red teaming (Assess)17
Square: new in this edition. Click a dot to read the verdict.
  • Adopt Proven, use it by default.
  • Trial Promising, try it on a real project.
  • Assess Worth watching and exploring, without committing.
  • Hold A false good idea, or outdated.

Techniques

  1. Test pyramidAdopt

    Lots of unit tests, fewer integration tests, a few end-to-end tests. Still the foundation of a healthy strategy.

  2. Contract testing (Pact)Adopt

    Check that an API's consumer and provider speak the same language, without spinning up the whole environment. Official website

  3. Test impact analysisTrial

    Only run the tests affected by a change: CI stays fast as the suite grows.

  4. Mutation testingAssess

    Inject small bugs into the code to check that the tests catch them. The best judge of a suite's quality.

  5. Fixed waits (sleep)Hold

    Waiting 5 seconds “just to be safe” slows the suite down and makes it flaky. Wait for a state, not a duration.

Tools

  1. k6Adopt

    Load tests written in JavaScript, easy to plug into CI. Official website

  2. PlaywrightAdopt

    The reference end-to-end testing framework: cross-browser, auto-waiting, debug traces and parallel runs. Official website

  3. AI-assisted visual testingAssess

    Compare screenshots while ignoring noise (fonts, anti-aliasing) so only real visual changes are reported.

  4. Record and replay without maintenanceHold

    Recording clicks gives fragile tests that become impossible to maintain as soon as the UI changes.

AI for testing

  1. AI-generated unit testsTrial

    Code assistants suggest unit tests in seconds. A real gain, as long as you review every assertion.

  2. Playwright MCPTrial

    Give an AI agent control of a real browser to explore an application and write tests. Official website

  3. AI merge request reviewAssess

    An agent comments on an MR's risks and missing tests. Useful as a safety net, not as a replacement for human review.

  4. Self-healing selectorsAssess

    The tool fixes a broken selector by itself. Handy, but it can hide a real regression: keep a record of every repair.

  5. AI-generated tests without reviewHold

    Hundreds of tests that pass but assert nothing useful. An unreviewed test is false reassurance.

Testing AI

  1. LLM-as-a-judgeTrial

    Having one model grade another model's answers, with an explicit rubric. Powerful, but calibrate it on hand-labelled cases.

  2. promptfooTrial

    Regression tests for prompts and LLM applications, driven by case files and runnable in CI. Official website

  3. LLM red teamingAssess

    Deliberately attack your own AI application (prompt injection, data leaks) before someone else does.

A tool to add to the radar?

Using an AI testing tool that deserves a spot, or disagree with a placement? Tell me: the radar changes with every edition.

Suggest a tool