Turns out that OpenRouter does report input/output token ratio and it's mind-blowing. Input token volume is far bigger than output. This has a massively on reduces their revenue and margin. Weekly revenue $3.7m, margin: ~$200k for the top 20 tools for programmers, academics, tech and science

OpenRouter's has the only public data I've seen about token growth and the shift of devs to using models. I tried to calc their rev/week for models likely used in dev tools. I have them at ~$1m margin on ~$18m rev per week docs.google.com/spreadsheets... Would love some eyes on this to validate

I had a productive year exploring the intersection of the Web and LLMs so I documented some experiments: aifoc.us/projects/ One of my wishes is that more people experiment with how we might advance the web and browsers with tech like LLMs other than "just generate some content for me".

I've been using a Mac for the last 20 years and I've only just realised that you can do OPT-Click on the desktop to hide the current windows... I've also just learnt that you can hide windows and not minimize. What other hidden magic is there? I'd like to know.

I don't know why I didn't publish my HTML5 Router to NPM over a decade ago, but there's no time better than now. github.com/PaulKinlan/l... www.npmjs.com/package/levi...

Anyone know any blogging platforms that offer the ability to gift a subscription? I'd like to set a blog up for my son as a Christmas gift (he's doing sports journalism course and he will likely need a blog in the future).

🤔 What are request priorities, and why do we even need the HTML fetchpriority attribute? Check out my new video to find out! www.youtube.com/watch?v=yoa7...

Some kind folks in Korea translated my "Dead Framework Theory" post handhand.tistory.com/entry/dead-f...

I use Gemini Deep Research all the time. I use Email all the time. The new Deep Research API seemed like a great opportunity to integrate with email. I made [email protected] as a demo that you can mail and get a report delivered to your inbox. Just put a prompt as the body of the mail

Can anyone with local dev knowledge of China help me understand this data in statscounter? Explosion in growth of Win in last few of months: gs.statcounter.com/os-market-sh... Also Chrome gs.statcounter.com/browser-mark... I'm wondering if there's another change (i.e locked user agent string etc)

I'm looking for some feedback on a Chrome Extension that I made. I wanted to explore the user trails concept from the Memex. Trails - chromewebstore.google.com/detail/trail... - allows you to create your own links between pages entirely locally and have them maintained across navigations.

I'm looking for some feedback on a Chrome Extension that I made. I wanted to explore the user trails concept from the Memex. Trails - chromewebstore.google.com/detail/trail... - allows you to create your own links between pages entirely locally and have them maintained across navigations.

It was fun to build flickity.val.run I turn each article linked from hackernews into a short video by summarizing the post, taking a screenshot and then combining them into a script for veo3.1 Video gen could be better, but surprisingly good.

It was fun to build flickity.val.run I turn each article linked from hackernews into a short video by summarizing the post, taking a screenshot and then combining them into a script for veo3.1 Video gen could be better, but surprisingly good.

If you're in to double iframes. I have a test custom-element called<generate-html /> that aims to let you securely* run HTML generated by LLMs in the browser. generate-html-element.paulkinlan-ea.deno.net github.com/paulkinlan/g... * this is a demo and example and should not be used in production

If you're in to double iframes. I have a test custom-element called<generate-html /> that aims to let you securely* run HTML generated by LLMs in the browser. generate-html-element.paulkinlan-ea.deno.net github.com/paulkinlan/g... * this is a demo and example and should not be used in production

When OpenAI launched their AppsSDK I was impressed by the double-iframe sandbox. To test out the AppsSDK I built a simple shim and Chrome Extension that replicated major parts of the API. Was fun to get OpenAI Apps hosted in a Chrome Extension and just work... github.com/PaulKinlan/o...

I explore 30 year old tech (Content negotiation in HTTP) and how we might use it on the web with LLMs. If you have the ability to transform any content into any content we can get a web that is built to be personalised to every single person in their preferred medium. aifoc.us/hyper-conten...

I explore 30 year old tech (Content negotiation in HTTP) and how we might use it on the web with LLMs. If you have the ability to transform any content into any content we can get a web that is built to be personalised to every single person in their preferred medium. aifoc.us/hyper-conten...

The headless stopgap - aifoc.us/headless-sto... LLMs and generative technology can change the way we think about creating and managing content. Specifically, we can apply *layout and functionality* to pages through prompts how CSS helped separate content from the style.

My #WebAI Summit talk "Built-in AI in the wild: A Mastodon translation success story" is now live: youtu.be/-l2BUUdBDS8?.... I had so much fun giving this talk, and in it I teach you the German word for waving back at someone who wasn't waving at you: Fremdwinkzurückwinkungspeinlichkeit 🇩🇪 👋 😳.

The headless stopgap - aifoc.us/headless-sto... LLMs and generative technology can change the way we think about creating and managing content. Specifically, we can apply *layout and functionality* to pages through prompts how CSS helped separate content from the style.

It's still frustrating that after 6 months my site aifoc.us is still not showing in search... No need to share with our search folks, I'm just venting.

Whither CMS? I think the CMS industry is up for a massive change with the introduction of LLMs. Historically CMSs tend to be used at the long-tail of the web where there is little traffic, it's also the space where people might be more willing to use new tools. aifoc.us/whither-cms/

Token slinging: aifoc.us/token-slingi... I've been thinking about the effects of tokens and how it impacts direct web development. Feels like there is a change in that we are now back to paying for our IDEs (even if it's indirectly).

📰 Dead Framework Theory 💬 React's reign might be over as LLMs could liberate us from its compilation woes. Mixed feelings on whether this is great or dreadful. 🤷‍♂️ https://news.ycombinator.com/item?id=45843220

Moon through the flames

I love autumn

Any recommendations of a KVM switch? I'd like it to be mostly usb-c (I use it for display) and for 2 monitors and a laptop

We've taken the new version of the Core Web Vitals Tech Report out of Beta and now consider this the report to use: httparchive.org/reports/tech... Anyone still using the old Looker Studio-based report, it's time to check out the new goodness with much quicker response times and improved UX!

I had fun last night. I got a Chrome extension to talk MCP via the Omnibox. It can even render and host OpenAI's Apps SDK runtime "Skybridge" - I have a small shim that attempts to polyfill the API.

Agree. I think LLM tools likely have to output React by default today. However LLMs need to work with any framework in brownfield, so enabling other outputs as an option becomes somewhat easy to implement - enterprise user might choose due to corporate policies and hobbyists for any reason.

Last but not least, the incumbent always had a strong advantage (nobody gets fired for buying IBM) - which didn't in the past stop innovation from happening or players to come up. Our belief in the team is that innovation will continue until UIs are a solved problem, which we think is far from true!

at least when the developers are interested in it. Also new features/API might matter if they're better performing then generated patterns, and even more ergonomic syntax might be useful to humans AND LLMs in all but greenfield use cases (if only to save tokens on maintenance).

Dead framework theory aifoc.us/dead-framewo...

Such a stunning morning in North Wales

I've been so annoyed by AI tools and CSS that I made a CSS MCP that: * consults with MDN for docs * reports browser compat * uses project Wallace to analyze your CSS github.com/stolinski/cs... Also made claude code agent for css but still perfecting it.

Little @codepen.io demo: single `img` squircles + grainy inset shadow codepen.io/thebabydino/... (for `img`, the actual image is content & gets painted on top of shadows; the earlier workaround has been a sibling/ parent's pseudo used as a "cover" that gets the shadow; none of it here 😼) #CSS #SVG

Was nice to play with Github's copilot to scaffold a web app in the creation of the repo. It's awesome to have a software that is what I want and not worry if anyone else really wants to use it paulkinlan.github.io/HackerNewsDe...

If we started Chrome Dev Summit again, would you be interested in it? (huh - apparently can't do polls here...), Please reply, Yes or No, or any other answer in between.

Playing around with Live streaming an LLM generated web - www.youtube.com/watch?v=HZBd...

Lovely weather in Chester today.

Stunning morning in North Wales.

Announcing our public preview of Chrome DevTools MCP! Experience the full power of DevTools in your AI coding agent→ goo.gle/4pDE6Tk With Chrome DevTools MCP, your AI agent can run performance traces, inspect the DOM, & perform real-time debugging of your web pages.

This is such a great enhancement. Being able to have any tool that talks MCP have the ability to deeply introspect your pages is huge.

Tonight I integrated an LLM into every part of the request/response life-cycle of the the browser to see what I could do. aifoc.us/interception/ It was a lot of fun, in the demo linked I can rewrite a web page as before it's rendered by the browser to give me just a summary of the contents.

Tonight I integrated an LLM into every part of the request/response life-cycle of the the browser to see what I could do. aifoc.us/interception/ It was a lot of fun, in the demo linked I can rewrite a web page as before it's rendered by the browser to give me just a summary of the contents.