The prompt is the program - aifoc.us/prompt-is-th... I built a journalling system that is just* defined as a Claude .md file that can also adapt to how I use it. The instructions following and the ability for a journal to be proactive feels incredible *I had some bash scripts to manage backup

Regular reminder: 85,000 developers a week run the wrong commands `npm add user` instead of the correct `npm adduser` www.npmjs.com/package/user

I live inside the browser and when NotebookLM came along it changed how I thought about info management and it got me thinking... What if you combine both? Want a Slide deck of 10 different news sources from Techmeme? 👍 A study guide of research that you've bookmarked? 👍 aifoc.us/if-notebookl...

I live inside the browser and when NotebookLM came along it changed how I thought about info management and it got me thinking... What if you combine both? Want a Slide deck of 10 different news sources from Techmeme? 👍 A study guide of research that you've bookmarked? 👍 aifoc.us/if-notebookl...

FWIW, CSP is the best thing you can use today, but it's not really built for exfiltration mitigation. We're working on github.com/wicg/connect... with that specific threat model in mind.

I wanted to explore if we can build something like Claude Cowork in the browser that can work with the user's file system safely running programs against them. We can get a very long way: aifoc.us/the-browser-... Feedback and critique appreciated.

Big news! Google have sponsored the @e18e.dev community $10k via their Chrome performance fund 🎉 I'm very thankful we've managed to create a space and a community people appreciate. Sponsorships like this will benefit the ecosystem, not just this community

If you're into NotebookLM I've got a tool that I'd love you to test out and give me feedback (it requires you to bring your own API key). It's an experiment that I've made just for me but I think it's really useful - query + convert your tabs into new pages chromewebstore.google.com/detail/folio...

Still blows my mind that you can write an instruction in English and a machine has a very high chance of following that instruction

If you're into NotebookLM I've got a tool that I'd love you to test out and give me feedback (it requires you to bring your own API key). It's an experiment that I've made just for me but I think it's really useful - query + convert your tabs into new pages chromewebstore.google.com/detail/folio...

Not 42 days any more, now only 40, yes Forty, days. And only 30 In-Person tickets left…

My use of coding LLMs has increased so much that now I'm more of a web producer than a web developer.

I never thought that I'd be doing most of my work just using my voice. whether it's planning, emails, even computer programming. It's kind of bonkers.

I've been using Claude, Max and Gemini this last week to build more projects than I've ever done in a long time, but I find it hard to know when I'm actually done. It feels that taste matters because it's just too easy to keep adding one more feature

I'm not sure if my skills.md file should be licensed as Apache, or Creative Commons. It's essentially just instructions, so it feels like code.

I asked the other day about transcription tools to help me see the words I say as I type. Couldn't find a fit, so I built one using the Web Speech API which shows my utterances. Fun fact built it with my voice. chromewebstore.google.com/detail/utter...

A whole heap of Devrel will just be a SKILLS.md file. I'm expecting presentation reviews to be "could have been a SKILlS.md"

What voice transcription tools do you use on mac? Everything that uses whisper seems to not output what I say in realtime. Context, I am speaking prompts on the cli and I want to see the words come out (like Mac voice control does), but I'd like it via hotkey

I made a little Chrome extension last night that opens all the kinks in selected text - chromewebstore.google.com/detail/jjdpn... Great on @techmeme.com if you want to quickly open all the related sources.

Welcome to BlueSky @opensource.google !! Happy to see you here.

New year's resolution, engage with open source developers on our new Bluesky account. ✨

Hello Bluesky! It turns out @bsky.app has an amazing free of charge API that makes it easier for developers to interact with the platform. As a developer, I approve—that’s what totally convinced me to get an account 🤩

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).