Chrome Releases Modern Web Guidance 1.0 to Steer AI Coding Agents Toward Current Web Standards
Chrome has released version 1.0 of Modern Web Guidance, a set of skills that feed web platform expertise, best practices, and browser compatibility data into coding agents, the AI tools that write and edit code. The stable release was announced on November 10, 2026. Chrome had previewed the project in early form at Google I/O in May.
Developers install the skills by running npx modern-web-guidance@latest install from the project directory, according to Chrome's announcement.
The 1.0 release has three main components. The first is a modernization workflow that scans an existing codebase and helps agents find places where code can be updated safely. The second is an expanded guide database covering hundreds of gaps in what AI models know. Chrome describes these guides as verified, with expert-vetted, cross-browser solutions. The third is evaluation coverage across all guides, backed by a public eval dashboard (a published set of test results) that Chrome says shows measurable improvements over agents working without the guidance.
Chrome also says every guide received human expert review, including sign-off from content area tech leads. User-facing guides also went through accessibility review.
Firefox and Edge have joined Chrome in supporting the project, improving cross-browser compatibility, and contributing to it. The project is cross-browser in its authorship as well as in its target content.
The broader context here is the gap between what a coding model learned during training and what the platform supports today. Browser features ship, get deprecated (marked for removal), and reach broad support at different times in each engine, the software that renders a page. A model working from an older snapshot can produce code that works in one browser and fails in another, and the developer may not notice until testing. The guidance is built to close those gaps and tie them to compatibility data, aiming the correction at the moment code is generated rather than in review or after release.
In my view, the detail with the most long-term weight is the participation of Firefox and Edge. A compatibility dataset maintained by a single browser vendor would carry that vendor's blind spots, and developers would have good reason to question it. Whether the shared contribution model holds up as the platform keeps changing is an open question. The public eval dashboard gives outside observers a way to check Chrome's claims about agent performance directly.
Looking at what this makes possible, the modernization workflow is likely to matter most to teams maintaining large legacy front-end codebases. Rather than relying only on periodic manual audits, such a team could have an agent propose changes that have already been checked against current guidance. For developers who would rather spend their time on product work than tracking which features are safe to adopt, moving the checking earlier in the process could be the practical gain to watch.
