Autonomous Site Discovery: How AI Maps Your App and Generates Test Coverage
Point AI at your running application. Autonomous discovery maps the pages, forms, and flows it can reach, then generates Playwright tests.
Validate.QA discovery explores your running web app in a real browser, maps what it finds, and writes Playwright tests from that map. You give it a URL and, if the app has a login, test credentials. This post covers what discovery does, what it produces, and where it stops.
How discovery works: explore, plan, generate
1. Explore
An AI agent drives a real Chromium browser through Playwright tools over MCP. It is not a static crawler: JavaScript runs, client-side routes resolve, and the agent clicks, fills and navigates the way a user would. As it goes, it records:
2. Plan
The agent groups the pages into features (for example sign-in, projects, billing, team) and plans scenarios from real user flows, preferring multi-page journeys over single page-load checks. A second pass can review the plan and drop or strengthen a weak expected outcome.
3. Generate
Each scenario becomes readable @playwright/test code: describe blocks, setup, and assertions written from what the app did during exploration. Before a test is promoted from draft to active, it is run against the live app. Tests that do not pass stay as drafts, and assertions that do not prove out are dropped rather than guessed. Tests the generator could not fully verify carry an @unproven tag so you know which to review first. A generated test looks like this:
Survey, deep and full runs
Run time depends on the size of the app and is not fixed, so there is no promised duration.
What you get: the site map
After a survey, Site Intelligence shows every page the agent reached, its type, its elements and forms, and the transitions between pages, plus the feature groups it proposed. This is an illustrative example of the shape:
The map feeds the planner, and you can read it on its own as a picture of your app's test surface.
Other things captured along the way
Because discovery drives a real browser, one pass also records traffic and page health:
Limits
Plan and next steps
Survey and discovery use Validate.QA's own AI, so they are included in Solo Dev and locked on MCP Army, where you bring your own AI. See pricing for the current terms and how it works for the pipeline in more detail.
Topics: Autonomous Testing, AI Discovery, Test Coverage, AI QA.
Read the full article · Get Started Free