How ZeroWeb tests AI-autonomous software engineering

A browser gives AI development a long-running test case: architecture, code, debugging and verification must work together.

What the experiment asks

ZeroWeb explores whether AI can keep a large systems project under development. A human sets direction, makes key decisions and verifies results; AI analyzes tasks, changes code, adds tests and investigates failures. The project is experimental and is not yet an everyday browser.

Why a browser?

A browser connects HTML parsing, CSS, layout, fonts, graphics, scripting, network requests and platform integration. A change that compiles can still break page behavior or rendering. These dependencies make it a useful test of sustained engineering rather than isolated code generation.

How changes are checked

The repository uses module boundaries, specifications, review rules and CI. Checks include unit and integration tests, Web Platform Tests, pixel comparisons and browser comparisons. The appropriate check depends on the changed behavior; one green test does not prove every browser path works.

What evidence to read

Read the development updates for completed work and remaining tasks, and the test results for measured trends. Updates link to commits and goal documents. Test results include corpus versions and measurement units, because changing the denominator can change the apparent pass rate.

What the results do not prove

Repository size and commit count describe activity, not software quality or human productivity saved. Closed goals record their scoped acceptance evidence; they do not imply complete web standards support or production readiness.

Sources: project README, quality gates, engineering rules.