I like the idea they mentioned of "a browser made up of composable components, protocol handlers separate from primary document renderers separate from attachment handlers", and I had the same idea. (Not all browsers will have to be implemented in this way, and they are not necessarily all the same, but this can be helpful when you want this.)
There can be two kind of extensions, sandboxed VM codes (e.g. WebAssembly) and native codes; the app store will only allow sandboxed VM codes, and any native codes that you might want must be installed and configured manually.
There is also the issue of such things as: identification of file formats (such as MIME), character sets, proxies, etc.
I hade made up Scorpion protocol and file format which is intended to be between Gemini and "WWW as it should be if it was designed better". This uses ULFI rather than MIME (to avoid some of the issues of MIME), and supports TRON character code, and the Scorpion conversion file can be used to specify a way to handle unknown file formats (there are several ways that this can be specified, including by a uxn code).
So, an implementation can be versatile to support things that can be useful beyond only MIME and Unicode etc.
Adding some additional optional specifications to WWW might also help, e.g. a way to specify that certain parts of the document are supposed be overridden by the user specifications in the client when they are available (although in some cases the client could guess, e.g. if a CSS only selects by HTML commands and media queries and not by anything else (or no CSS at all), then it should be considered unnecessary and the user's specifications of CSS can be used instead if they have been specified). Something like the Scorpion conversion file would be another possibility to have, possibly by adding a response header.
The previous "Google is killing the open web" article also mentions some similar things, but also a few others:
> in 2015, the WHATWG introduces the Fetch API, purportedly intended as the modern replacement for the old XMLHttpRequest; prominently missing from the new specification is any mention or methods to manage XML documents, in favor of JSON
Handling XML or JSON should probably better be a separate function than the function for downloading files, though. (Also, DER is better for many things)
> in 2024, Google discontinues the possibility to submit RSS feeds for review to be included in Google News
This is not an issue having to do with web browsers, although it is related to the issues that do have to do with web browsers (not) handling RSS.
> in 2025, Google announces a change in their Chrome Root Program Policy that within 2026 they will stop supporting certificate with an Extended Key Usage that includes any usage other than server [...]; this effectively kills certificates commonly used for mutual authentication
While I think they should not have stopped supporting such certificates (whoever the certificate is issued to probably should better make their own decision), it is usually helpful to use different certificates for client authentication anyways, so this is not quite as bad as they say, although it is still bad.
(X.509 client authentication would also have many other benefits, which I had described several times in the past.)
> in 2021, Google tried to remove [alert(), prompt(), and confirm()], again citing “security” as reason, despite the proposed changes being much more extensive than the purported security threat, and better solutions being proposed
Another issue is blocking events and JavaScript execution (which can sometimes be desirable; in the case of frames it should be better to only block one frame though), and modal dialog boxes potentially blocking other functions in the browser (which is undesirable). For the former case, there other other things that can be done, though, such as a JavaScript object that controls the execution of another JavaScript context which can then be suspended like a generator function (without needing to be a generator function).