good idea. who did you use to print it?
HN user
tonyoconnell
Enchanted by artificial intelligence for medicine, education and marketing.
Almost exactly the same for me
A blast from the past. I used his scripts for sending email from forms. The internet was a very nice place back then.
The narrative is moving towards KYC
What it is
Apple's Foundation Models framework (shipping in iOS 27 / macOS 27 this fall) is the standard Swift API for on-device AI — the same API Apple uses for their own small model. This package makes Claude plug into that same API as a drop-in swap.
// Apple's on-device model
let session = LanguageModelSession(model: SystemLanguageModel.default)
// Claude — same API, just different model constructor
let session = LanguageModelSession(model: ClaudeLanguageModel(name: .sonnet4_6, auth: auth))
One API, two tiers. You write your app once against the Foundation Models protocol. On-device model handles fast/free/private tasks; Claude handles heavy reasoning, long context, or capability gaps — you swap the model, not your code.You don't call the Anthropic API directly. Apple's framework handles streaming, tool calling, and structured output (@Generable) — you just get Claude's capability through it.
How did this get to the front page?
i gave claude code my design system that i built in figma and haven't opened it again. it's much faster designing with your voice.
just wondering... do you think bun's rewrite with ai was vibe coded or engineered with ai? i know it wasn't perfect in the beginning but i think it was good engineering and what was built will make it faster and better.
you can actually do everything most people do in next with vite/astro. i was worried to do it but i haven't ran into any problems i couldn't solve.
the dx is wonderful if you give claude code your global api key. and the price is amazing. you can deploy complex web apps for free. i love vite and astro which is built on vite. i ran both on cloudflare before they were bought by them. i'm happy. at least they weren't bough by adobe.
so we shouldn't talk to humans when an ai can give an answer?
Here's a link to a few sample chapters https://vivianvoss.net/assets/files/integrated-by-design-pre... I enjoyed reading them - they presented a very clear arguement to select FreeBSD over Linux
why not just use https://getutm.app/ ?
I've been thinking a lot about what AI agents buy, and what can I sell to them. If you are the owner of an agent that has a wallet can you share what it has bought or sold?
I think building apps and websites for other people is mad depressing. It went from "move this up there, and change that colour to pink" to a client ruining a beautiful site by using a nocode tool. Now they have superpowers to ruin it by adding AI generated code as well. AI can generate absolutely beautiful code if it is generated on the right architecture with the right patterns and rules. The problem isn't the AI it's the people telling AI and developers what to do.
Hello. I was delighted to see Penpot get a lot of traction when Adobe were going to buy Figma. I'm not a fan of Adobe because they killed so many of my favourite products. I switched from Macromedia (acquired by Adobe) Fireworks to Sketch to Figma and probably would have moved from Figma to Penpot already but I simply don't need Figma anymore. I use Shadcn UI and Claude Code to do my design these days without every having to touch Figma. All I need is a lightweight image editor that can crop and optimise some vectors and bitmaps now and again. I'd like if I had a lightweight design system that syncs designs with my code and the code with my design but I don't really need it. It looks like you are building something like this. Your MCP should be more usable with the new Claude Opus model and their new advanced tool use https://www.anthropic.com/engineering/advanced-tool-use but it would be useful to package that functionality into a tool that can call servers without needing MCP. Then I would give it to my Claude Code design subagent without worrying about it polluting my context window. I took the time to write this because I really like what you are trying to do and your new Open Nitrate business model is wonderful. I wish you lots of luck. If you ever want to chat about AI feel free to get in touch at tony@one.ie
You can make your website run fast as well with https://astro.build - it strips the Javascript and uses HTML until js is needed. You can get 300ms page loads for outcrop.app with Astro on Cloudflare Pages. Good luck with the project. I requested early access - You should use the response to the form submit requesting access better - I mean somebody who added their email expressed a lot of intent - why end the conversation with a toast notification? I hope life is going well in Dublin.
This is so cool. Good luck with your venture.
Maybe you could use your software to help them support other frameworks
You don't need a course ... just build something with Claude Code and Shadcn UI
Some parents of infants would be able to find a way to feed their children safely.
They used to be open source but closed around the same time they made a partnership with Open AI. Wonderful product though.
You just don't understand that AI becoming smarter than humans
Do you want free email for multiple addresses on multiple domains with all the features of a Gmail? Just setup Cloudflare to manage your email, create rules to forward to regular Gmail accounts, which you can configure to send email from your domains. 100% free.
n8n is a very practical way to build them.
Langchain abstracts too much and you can't really see what's going on or control the flow of data with real precision. They are fixing that though and now you have much better visibility into what's being inferred. I think Langchain is pretty useful though, especially if you want to integrate with something quickly.
Do you leave your phone outside when you use it? Do you use the same wifi? I would find it hard to believe that Meta can't predict who you are.
A big problem with Apple Care is here in Thailand anyway you need to give them your computer for a few weeks. You have to wait a week for them to look at it. They won't even allow you to use it and then bring it back in a week.
I started getting very serious about AI when I got early access to GPT 3. Before that I was using Google Dialogflow to build AI powered chatbots. Since then I downloaded, installed and tested literally hundreds and hundreds of AI related software application. At first this was really exciting ... I thought "wow this software is going to change everything" ... then a few days later another application would be released and another and another and until I became paralyzed by the amount of choices available. This resulted in a huge amount of stress. For months and months I had analysis paralysis. Eventually I settled on Supabase/PGvector for RAG (mostly because it provides Row Level Security for data/vectors) and this gave me a strong foundation to build on. Now I use Supabase, Vercel AI SDK with Svelte and N8N and mostly Claude Sonnet 3.5 as my LLM (with OpenAI O1 for more complex tasks and GPT 4o mini for cheaper inference). I also use Cursor (with the same models) every day which is amazing.
For example, I built a full stack events management plaftorm with Sveletkit, Supabase and Shadcn entirely with AI - AI generated both the frontend and backend code and database schema. It can be quantified like this - 0 (I had no platform) 1 = Now I have a platform. I honestly wouldn't have been able to build it without the help of AI, it would have taken too long and I wouldn't have started development. Also I really like some of the design decisions it took and there were many times I thought a generation was really beautiful and that I wouldn't have though of that. The designs were better than what I would have come up with in Figma so I didn't bother using Figma to design the UI. Think step by step, the user is going to land on this page with a goal of X, how can you improve the ux and remove friction while making the ui extremely beautiful worked really well.