TestMu Kane CLI Exploration: AI writes tests with no spec

AI coding agents ship code faster than anyone writes it down. Then the test agent gets a URL and nothing else. On October 30, TestMu AI (formerly LambdaTest) launched Kane CLI Exploration, an AI agent that maps a whole web app from one URL and turns the map into tests. For a small team with a legacy app and no spec, that is a real starting point. It is not a finished test suite.

What actually happened

Per the Business Wire release, you give the agent one URL. It walks every page and flow it can reach. For each page it records the controls, the routes in and out, and the states the screen can be in, for example a form before and after a validation error. It flags UX, security and accessibility problems on the way, including pages your Playwright, Selenium or Cypress suites never reached.

Kane CLI then reads that map like a requirements document. It writes use cases and acceptance criteria, runs the tests in real browsers, and saves an Evidence Pack for each run. A person who knows the product reviews the tests, and the approved ones run in CI on each pull request.

The specifics:

• Each exploration runs on "dozens" of parallel browsers in TestMu's cloud. TestMu says a large app maps in hours.
• Web apps only for now. Kane CLI itself, launched in April, also covers mobile.
• It is part of the KaneAI Max plan, with three exploration sessions per license.
• The release gives no price.

Why it matters for your business

No spec is the normal case. Plenty of small businesses run an app built by a contractor who has since left. Nobody wrote acceptance criteria. An agent that builds them from the live app gets you from zero tests to a draft in a day.

A map of what the app does is not a map of what it should do. If checkout has a bug today, the agent may write a test that locks the bug in. That is why the human review step matters. Read every generated assertion before it goes into CI.

Three sessions is a budget. Point it at staging with seeded test data, not production with real customers and real payment buttons. Pick the flows that make money first: sign-up, checkout, quote request.

Evidence Packs are your audit trail. Keep them. When a client asks if the release was tested, you show them, you don't just say so.

Sources: TestMu AI via Business Wire, Business Wire original.