Post

In response to our objection to the Prompt API, some folks replied "well, fix it. Propose something better". This is something we end up doing a lot at Firefox…

…I'm currently finishing a redesign of popover="hint" github.com/whatwg/html/..., we're proposing an alternative to the translation API after finding security issues in the Chrome API github.com/mozilla/stan.... We provided a lot of fixes to the design & spec of anchor position & the nav API…

…some of this is just part of participating in web platform design, but in some cases it's a little unbalanced. Given the same resources, I think we'd be in a better place if Chrome was working on fewer features, but fully-baking them. I don't think it's on Mozilla to fix every half-baked proposal.

Also, the lack of an alternative proposal is not a good reason to ship the half-baked thing.

Genuine question: are you happy with the role that the web plays on mobile devices today in comparison to native apps? If not, is it possible that the industry shift to AI might be an iPhone-level moment and put the web on the same position on desktop too (say, if openclaw and MCP takes on?)?

I'd like the mobile web to be better in the future than it is today. I don't think the Prompt API achieves that.

Taking a step back for a second (out of the specifics of the Prompt API), don't you think AI (as mobile was) is such a fundamental shift that is hard for any of us to predict how things are going to play out?

I think it's important to keep to the specifics of this API. Mozilla hasn't objected to it just because of AI - we've been clear on that. It's the API itself that's bad. One of the specifics I'm curious about is why Chrome doesn't use the already-there OS model in cases where one is available.

yup that's exactly where i'm at. i want an API that exposes a local model at some point, but i definitely *do not* want an API that forces me to use google's model

This is what I struggle with "Given the same resources, I think we'd be in a better place if Chrome was working on fewer features, but fully-baking them.": haven't we (browsers) been wrong multiple times agreeing on what was the one thing to work on (e.g. semantic web, arguably WCs)?

There is a prototype in Chrome of the Prompt API powered by Android's ML Kit (built-in model) and folks have looked at doing the same for iOS using the Foundation Models framework. Though lack of engine choice on that platform makes it awkward.