(This relates to my note at the end of https://simonwillison.net/2026/Jan/27/one-human-one-agent-on... )

The things that make me think this is still a huge project include:

1. JavaScript and the DOM. There's a LOT there, especially making sure that when the DOM is updated the page layout reflows promptly and correctly.

2. Security. Browsers are an incredibly high-risk environment, especially once you start implementing JavaScript. There are a ton of complex specs involved here too, like CORS and CSP and iframe sandbox and so on. I want these to be airtight and I want solid demonstrations of how airtight they are.

3. WebAssembly in its various flavors, including WebGPU and WebGL

4. It has to be able to render the real Web - starting with huge and complex existing applications like Google Maps and Google Docs and then working through that long tail of weird old buggy websites that the other browsers have all managed to render.

I expect that will keep people pretty busy for a while yet, no matter how many agents they throw at it.