AI is eating Software (?)

4% of commits on GitHub generated by Claude Code (and growing fast)

Developers & AI

Web Standards target Software Development

• Implementations (e.g. browsers, libraries)
• Integration (e.g. web app & web pages) - what and how (accessibility, internationalization, …)
• Prototypes, shims & polyfills
• Test suites, CI, validators, IDEs

AI-Generated Code & Implementations

The FastRender repo uses git submodules to include relevant specifications, including csswg-drafts, tc39-ecma262 for JavaScript, whatwg-dom, whatwg-html and more

FastRender: a browser [*] built by thousands of parallel agents

Specs are the new code

[*] although not a production ready browser by any mean

AI-Generated Code & Implementations

HTML5 is extremely well-specified, with a long specification and thousands of treebuilder and tokenizer tests available in the html5lib-tests repository.

How I wrote JustHTML using coding agents

AI-Generated Code & Integration

[It] can take years of Interop work to get a feature to the point it becomes “Baseline newly-available”. […] If it’s not in the LLM training data, it doesn’t exist.

Developer experience features (syntactic sugar, convenience APIs) are competing against established […] patterns in LLM training data

Dead framework theory

(see also Docs CG discussion on impact on documentation)

AI-Generated Code & Integration (A11Y)

The [...] results suggest that objective error count is too high to rely on LLM technology at all in digital accessibility work, even under explicit expert guidance.

AI-Generated Code & Integration (A11Y)

LLMs excel at addressing basic accessibility requirements but struggle with complex accessibility requirements, particularly ARIA-related attributes, performing worse than human developers.

Advanced prompting techniques consistently generate code with lower accessibility issues than human-written code [but with] inherent limitations in addressing accessibility through prompting alone.

Web Techs dominates code

Questions

• Impact on writing specs?
• Impact on prioritizing features?
• Impact on writing test suites?
• Impact on writing guidelines?
• Incorporating AI-generation in adoption strategy?
  • Can we influence more directly the relevant AI tools?
  • Will this move the Web vs Native needle again?
• Experimentations worth conducting?