Bez Generates a Browser Engine From Specs: 0.6% of Features Done, 93.0% Untouched
Software / news
Bez Generates a Browser Engine From Specs: 0.6% of Features Done, 93.0% Untouched
The Rust project passes 227 recipe cases and nine CSS 2.1 layout rules, and Hacker News commenters doubt the method scales.

Bez, an open-source project that generates a web rendering engine from specifications and tests, lists 0.6% of features as fully generated and 93.0% as unreached. The numbers are on the project's own page, under the burrito.space account.
Bez follows the same pattern as OpenDLSS-NR, a Vulkan reimplementation of a proprietary system, and shares its question of what is actually verified. It reached Hacker News on Oct. 2, where the thread has 87 points.
What Bez says it does
The page states the goal as "Generate a web rendering engine from specs and tests". A model writes candidate implementations from specification text. Each candidate is run against Chromium, Firefox and WebKit, and code is committed only when the rendering matches the three shipping browsers. The repository is 76% Rust.
The listed aims are complete or tree-shaken builds, fast performance on all devices, easy embedding and quick updates when specs change. The portions of the page fetched here do not name the model that writes the candidates.
Where it stands on Oct. 2
| Measure | Figure |
|---|---|
| Features fully generated | 0.6% |
| Features hand-written | 0.3% |
| Features unreached | 93.0% |
| CSS 2.1 layout rules implemented | 9 |
| Recipe cases passing | 227 |
- Unreached93 %
- Fully generated0.6 %
- Hand-written0.3 %
Source: Bez project page on Tangled, accessed 2026-10-02
The three figures account for 93.9% of features, and the page as fetched does not label the remaining 6.1%. The project also reports passing WPT normal-flow pages, the Web Platform Tests category for basic block layout. It has 8 stars and about 1.2k commits across 21 branches.

The finding that is already useful
The comparison step produced one result worth reading even if the engine never ships. Across 699 of 705 comparisons, the page says, "three browsers mostly agree, and the third names the odd one out". When one browser deviates, the other two settle which behaviour counts as correct.
It also found a concrete difference: Firefox rounds with a 1/60px unit, and the page says this causes real web compatibility issues. A spec read alone would not surface that.
What commenters on the thread doubt
Hacker News user nicoburns, who describes three years of browser-engine work, wrote that AI is very far from doing this and that the generated code would be too slow to be useful. User troupo wrote that most web specifications are written in human technical language, not in a machine-readable form.
User esprehn wrote that specs leave behaviour to the user agent in places, and that real web compatibility means doing what Chrome does. That objection sits next to Bez's design, which treats the three shipping browsers as the oracle and so inherits their quirks on purpose.
None of these comments is a measurement. The project has published none that would answer them, such as render time against any of the three engines.
What comes next
The page gives no schedule, and IANA's example.com change is a reminder that behaviour on trivial pages still shifts. The next figure to watch is the unreached share, which stands at 93.0% on Oct. 2.
Sources
More in Software
- 01Zigzag, a 35 MB Mac Browser, Courts Arc Users 35 Days Before ShutdownThe WebKit-based browser imports Arc spaces and pinned tabs, but its README lists no sync, no Boosts and a partial Chrome extension API.
- 02Nimony 0.6.2 Sets a Year-End Goal for Nim 3 With 44 Days LeftThe compiler that will become Nim 3 is MIT-licensed and builds natively without LLVM, but its FAQ says it cannot yet compile Nim 2 code unchanged.
- 03Alloy CSS Runs Menus and Tabs Without JavaScript, but Dialogs Need Safari 26.2The 24 KB MIT stylesheet leans on popover and invoker commands, so its no-script promise has browser floors that differ by component.
- 04Windows 11 Telemetry Flags Which Coding Agents You Run, Researcher FindsXusheng Li traced a service-experiment feature in build 26200.9974 that matches executable names such as claude.exe and queues the result as optional diagnostic data.