Derk Nolte

@SebAaltonen @cryptopunk7213 Interesting, which model(s) are you referring to that could run on m5 max?

Yes, and that's why Apple is a nice safety bet if you invest in AI companies. Apple is a big winner if AI software bubble bursts. They have a lot of money in bank waiting to purchase AI software companies for cheap. And they have very good chips and edge AI. M5 GPU has tensor cores now (4x fp16 rate). 8x rate fp8, 16x rate fp4 coming in next generations (like Nvidia)? New open source LLMs running locally on 128GB M5 Max are actually very good right now (between GPT 4.5-5.0 quality). No token cost, no licensing costs.

i find it fucking hilarious how Apple "failing" at AI is now the exact reason they're about to win it: - watched everyone else burn $1.4T+ building models... then picked the winner (gemini) to use for... $1B - while everyone fights to grow users, apple flips a switch and 2.5 billion devices get AI siri tmrw. - $150B to splurge on the device / app layer. zero competition (because everyones spent their cash). - while openAI charges $200/mo subscriptions, Apple lets you run models on-device (cheaper, faster, private, personal) - while openAI struggles to build an AI device, Apple just dropped 5 powered by the best AI chips for hand-held devices. they "lost" the model race because they didn't need to win it in the first place greatest to (accidentally) ever do it.

@sabine_s_ is it now "I know how to prompt and QA"?

"full stack" used to mean "I know React and can copy from Stack Overflow"

@badlogicgames @RobertClaypool @geirfreysson @nicopreme Ooh cool, I didn't know there was a NES emulator plugin too. That's awesome haha

@RobertClaypool @geirfreysson if in doubt, check if there's an extension for MCP support in pi :) @nicopreme has a nice one with bells and whistles. pi.dev/packages

got an email from a pi user with a link to his blog "Things change too fast in this space for anything to be permanent. But right now, the thing that calls itself a shitty coding agent is the least shitty tool I’ve used." i agree to both onuruzunismail.com/blog/the-engin…

@mitsuhiko @simonw Interesting, it seems like that opencode's compaction doesn't retain the last messages. I'm about to try Pi. @mitsuhiko I guess you like the way Pi does compaction?

@simonw BTW I find the docs on pi's compaction quite useful here to understand how to structure this: github.com/badlogic/pi-mo…

Here's a compaction edge-case I hadn't known to worry about before

@dillon_mulroy @itsnoahd That sounds nice. I'm also going to give it a try.

@itsnoahd i like it so far, it's a lot like nvim

pi

@dillon_mulroy Ooh awesome. How "big" is the feature/work that sprints were required? Or is this just an easy way to have a task breakdown structure in a sequential way with sprints?

so maybe the product owners were on to something

@mitsuhiko That’s awesome. I’ve been looking for a way to set up a web browser skill. I think I read somewhere you let the agents create the skill, right? If you’ve published your Claude/Pi skills or prompts anywhere, I’d love a link. Obviously totally fine if it isn’t :)

I think it's kinda cool. I asked it to debug what's going on with the shadows and now I can just scroll up and see what it tried :)

It's a small thing, but I absolutely love that Pi renders images. Particularly when my web-browser skill goes and makes screenshots itself it can be hard otherwise to see what it actually snapshotted.

@kdy1dev @opencode what made you switch?

I partially switched to @opencode recently. opencode for my business and Claude Code for other businesses. I may switch fully to opencode soon.

@kettanaito That's awesome, thanks for sharing ❤️ I don't know much about service workers so I was curious. It also benefits compared to HTTP browser caching with cache control header?

@DerkNolte This: github.com/kettanaito/sw-… I wouldn't recommend deploying this as-is, it's a WIP to illustrate how to write a service worker from scratch. But it yields benefits even at this stage.

A 15-LOC service worker that only caches images can yield up to x6 times faster page load time. Service workers are pretty good and you should use them more btw.

@ScriptedAlchemy This is awesome! Any plans to share more details on this to public? No pressure, just super curious —especially how you orchestrate agents/subagents and switch models.

Today is the day the new Claude Max limits kick in. Just in time. Ive built out a UI for opencode which allows for bulk management of concurrent agents and subagents. Ill use this to more intelligently switch between other models for various tasks to offload pressure from my Max plan. For example, using AI elements in the frontend to dispatch cheap or free subagents for egress tokens, and preserve claude for management and ingress. Still some kinks to work out. But im liking it, will add a agent editor ui as well to configure agents on the spot

@mitsuhiko Wow, that must’ve felt surreal! What happened next—did the car just keep going?

Still one of my weirdest SF experiences was riding a Waymo alone through the tenderloin with a drug addict falling in front of the car. Waymo slows down and just waits for a good minute. Alone in a self-driving car, and outside visible human suffering. x.com/AndrewYatzkan/…

@samhuckaby @codegirl007 I'm also very interested to follow along :) I think it is very interesting when people pivot like this. And congratulations @codegirl007 with new big step! 🔥🔥🔥

@codegirl007 I’m looking forward to following along! I love a good origin story.

For 11 years, I’ve lived in the comforts and “safety” of working at a large tech company. All that is about to change. I’ve hashed out plans and back up plans and backup plans to those backup plans. But as someone once said, we’re doing it scared. Don’t know who said it.

@kettanaito Congratulations!!

Huge news! I've written a book. This is a story of light and dark that won't leave you indifferent. Illustrated, atmospheric, and meaningful. The kind that stays with you. Please enjoy this cover reveal for Loggerheads: zakarcher.com/blog/cover-rev…

@theo @chemicalheadset Or it is time for phased rollouts?

@chemicalheadset If I learned anything from this failure, it’s that we can’t ship risky shit on weekdays. This was our lowest traffic time of week, and a 1% edge case broke hundreds of people still.

I’m so tired. Will try again tomorrow, maybe.

@victorsavkin @typescript Congrats on the Nx 21 release! 👏 I’m really excited about the TypeScript DX improvements—could you share what’s new? The link seems focused on Nx 20 changes, right? Any details coming for 21? It would be cool if the deps in a (p)npm workspace would also be synced/installed

Huge @typescript DX improvements. Whether you're using ⚛️ React, 🅰️ Angular, or 💚 Vue, expect notably faster builds across the board. nx.dev/blog/new-nx-ex…

Nx 21 is now available with major upgrades across the board. Three key themes. Thread 👇 @NxDevTools #monorepos #ai #smartrepo

@enggirlfriend Very symmetric plating of the meat, love it

on vacation but still engineer wifey

@neogoose_btw Ah, I should give it go then. No bugs ruining the game experience?

Civ 7 is not that bad btw

@feedthejim Thats great to hear 💪

@DerkNolte our goal is to make the opennext maintainers life easier, not to replace it!

dear self-hosted users of Next.js, what can we do to make your life easier?