Two Ways AI Changes Testing
AI-augmented testing and vibe testing both use AI to improve software quality. They solve different problems, operate at different levels, and produce different artifacts. Using the wrong one for the wrong job wastes time. Using both together covers more ground than either alone.
Here is how they differ, where each excels, and when to use which.
AI-Augmented Testing: AI Helps You Write Tests
AI-augmented testing is the practice of embedding AI capabilities into a human-led QA process. The human defines what to test. The AI handles the parts that do not require human judgment — generating test code, maintaining selectors, prioritizing test runs, and detecting flaky tests. [1]
Gartner defines AI-augmented software testing tools as tools that provide integrated capabilities for continuous, self-optimizing, and highly autonomous testing through AI — including test scenario generation, test case maintenance, test suite optimization, and test value scoring. [2]
The output is code. AI-augmented testing produces test files — Playwright scripts, Jest suites, Cypress specs — that you commit, review, and run in CI. The AI accelerates the writing and maintenance of those tests, but the artifacts are traditional test code that the team owns.
Vibe Testing: AI Is the Test
Vibe testing flips the model. Instead of writing test code, you describe what should happen in natural language. The AI interprets the intent, drives the browser, validates the result, and adapts when the UI changes. The term emerged as a counterpart to "vibe coding," coined by Andrej Karpathy in February 2025. [3]
The output is intent. The stable artifact is the natural language description — "navigate to signup, fill in credentials, submit, verify dashboard loads." The AI handles selectors, waits, and assertions. When the UI changes, the AI adapts without test code updates. Playwright MCP is one of the most flexible implementations: Claude reads the accessibility tree and performs deterministic actions through structured tools.
Vibe testing catches issues that traditional automation often misses — slow-loading popups, confusing button labels, awkward navigation flows. It tests what the experience feels like, not just whether a selector exists. [4]