Modern Web Guidance 1.0: Set a Baseline Target for Your AI Agent

What is Chrome's Modern Web Guidance 1.0?

Modern Web Guidance is a set of skills that embed web platform expertise, best practices and browser compatibility data directly into coding agents. Chrome announced the stable release, version 1.0, on 10 November 2026, after an early preview at Google I/O in May ^[1].

The pitch is simple. Chrome's documentation says AI coding agents tend to default to older solutions for modern web problems, and those solutions often use JavaScript where a native feature would do ^[2]. The skills tell the agent which native feature to reach for instead.

How much it covers depends on which page you read. Chrome's guide says it includes best practices for over 100 web development use cases ^[2], while the Show HN post says 200+ guides ^[5]. No page explains the gap between the figures, and the README lists 250 modern web features ^[6], so this article does not give one total. Count the guides that match your stack after you install it.

What changed between the preview and 1.0?

Chrome lists several additions in the release post ^[1]:

• A modernization workflow that scans an existing codebase (HTML, CSS, JavaScript and package.json) for places to upgrade safely.
• An expanded guide database aimed at hundreds of verified gaps in what AI agents know.
• Expert review of every guide, including sign-off from content area tech leads and a dedicated accessibility review on user-facing guides.
• Evaluation coverage across all guides, with a public eval dashboard.
• Firefox and Edge joining Chrome to support the project and improve cross-browser compatibility.

The repository describes the software as Apache License 2.0 and the guide content as CC-BY 4.0 ^[6].

How do you install it and try it?

The README's quickstart is one command, npx modern-web-guidance@latest install, which the README says runs an interactive wizard ^[6]. You can also try it without installing: the README shows a search command for finding guides, and a legacy-scan command that scans a codebase for legacy patterns ^[6].

One limit is worth knowing before you expect a big diff. Chrome says the first release of the modernization workflow uses static analysis to check for a limited set of patterns, not a full audit ^[1].

What is a Baseline target, and why does it matter?

Baseline is a shared label for how widely a web feature works across browsers. Chrome's guide defines three levels: limited availability (not interoperable), newly available (interoperable within the last 30 months) and widely available (interoperable for 30 months or more) ^[2].

The target is the one setting you control. Chrome says you can choose a Baseline target and add it to your AGENTS.md or CLAUDE.md file, for example "This project's Baseline target is Baseline 2024" ^[2]. LogRocket's July walkthrough of the preview says the default target is Baseline Widely available ^[3], while freeCodeCamp says the tool defaults to the newest Baseline features ^[4]. Chrome's current guide page does not state a default in its text. With three accounts that disagree, set your target explicitly rather than rely on a default.

How does it handle features that not every browser supports?

Chrome's guide says some guidance covers features with limited availability, and that the guides then tell the agent how to keep older browsers working. In many cases the fallback is progressive enhancement, and any polyfill is loaded conditionally so only unsupported browsers pay the cost ^[2].

Two opt-ins widen the guidance. Volatile features, which exist in only one browser engine and may change, need npx modern-web-guidance@latest set-allow-volatile-features. Features in an origin trial need set-allow-origin-trials ^[2]. Chrome says origin-trial guidance is not made available to your agent by default ^[2], and the README says volatile-feature guides are gated until you consent ^[6].

What did independent tests show?

Two independent write-ups tested the preview, so treat them as examples, not benchmarks. LogRocket compared a guided and an unguided agent on a task app. The guided version used a native dialog and container queries, where the unguided one used custom modal code with JavaScript state and timing ^[3].

LogRocket's article also warns that the tool does not replace code review, browser testing or product trade-offs ^[3].

freeCodeCamp tested a tooltip and a toast. The guided tooltip needed no JavaScript, but the article notes that the interestfor attribute is available in Chrome behind a flag, with a polyfill ^[4]. Neither article reports timings or bundle sizes.

Why are some developers pushing back?

The Show HN post reached 284 points and 87 comments, and the page showed it as nine hours old when fetched ^[5]. Its author says the effort started in Chrome and shipped 1.0 in collaboration with Firefox, MDN and Edge ^[5].

A recurring objection in the thread is browser support. One commenter listed CSS features the skills point agents toward that are stable in Chrome but missing from Firefox and Safari ^[5]. A project contributor replied that the guides document how to use these features and give the agent a browser support table for anything not widely available, and that the agent then decides based on the site's browser support ^[5]. Others worried about Chrome's influence over web standards ^[5].

How reliable are the accuracy numbers?

Treat them as vendor figures. Chrome's guide says early results show an average 37 percentage point improvement in adherence to modern best practices, and that results may vary with your project, model, prompts and tools ^[2]. The tests, per the same guide, are Playwright checks run against a correct reference implementation and a deliberately flawed one ^[2].

The release post adds customer figures. One company reported pass rates on modern web checks rising from 61% to 94% across 180 runs on its own production codebase ^[1]. These are each company's own numbers, and the pages reviewed for this article include no independent replication.

Should you install it today?

If you already use an agent on a web codebase, trying it costs one command. Do these steps in order:

1. Run the legacy-scan command on a branch and read the results before accepting changes ^[6].
2. Write your Baseline target into AGENTS.md or CLAUDE.md before the agent writes new code ^[2].
3. Leave the volatile and origin-trial opt-ins off unless you ship to a single browser. This is the author's recommendation, not Chrome's.
4. Test the generated pages in every browser you support.

If you must support older browsers, choose an earlier Baseline year in your instruction file, such as Baseline 2024 ^[2], and judge the output yourself.