Karl Marx

announcing: Happy Coder, native mobile client that connects to Claude Code. Fully Open Source, MIT license github.com/slopus/happy Control Claude Code from your phone - Push notifications - End-to-End Encryption - Input from either device appears immediately on both

Yep

I believe I have one upped the designers at Notion and frankly everywhere. Instead of putting the toggle control on the left of the parent, why not underneath? Hover reveal means I have to memorize the location of the control anyways.

TIL about in-basket tests (manager assessments from the 1950s) you just prioritize/respond to a pile of mail, calls & memos. this seems incredibly doable for LLMs. has anyone benchmarked this? asking for my manager

@providenceluvr Very low. You can make more money doing door to door solar sales.

is this… considered a low wage in the US?

Uber is the one who designed and distributed the backpack. Why don't they just make an umbrella attachment? Seems at least partially their fault the employee got wet

The irony is I think people know this, which is exactly why they grind so hard on the idea/code. They're aware the social stuff matters, so they're desperately searching for a bar they can clear where it'll be good enough to explain itself,

AI will not replace humans AI will not replace humans AI will not replace humans AI will not replace humans

I hope this is not true

@alexeykozy Few know that models are already trained using this kind of debugger data. See this HN post from last year.

Been hacking on a new kind of debugger — one where the model sees the runtime. Early signals are strong. Watching it dig through the toughest bugs in a massive codebase in real time is wild 🤯

@var_epsilon Always start the line with a comment, then remove the comment. Or even better, measure twice, cut once. select count(*) from (select * from x where y) then edit that line into delete * from x where y

“DELETE FROM x WHERE y” is insane syntax because there’s a brief moment while typing that where you can wipe a table if you hit enter

the cambrian explosion is happening. you're just not seeing it because of how credibility works on here. if you have any reputation, you don't blast every tool you build. you show 2-3 people. they say "too specific for me." you stop sharing. that's how curation works when you're not anonymous. the stuff exists. Speaking for myself, I learned sharing AI-assisted solutions to personal taste problems = getting judged by your worst output. want proof? don't ask for github links. ask about disk space. before AI my macbook had ~9gb of projects. 500gb was fine. hundreds of gb free. now i'm constantly hitting the limit. hundreds of projects with isolated dependencies. hundreds of gb in npm, cargo, react native builds. i had to write cleanup scripts that i run weekly just to keep space. ~/.yarn, ~/.pnpm, build caches everywhere. if devs are suddenly running 300gb+ source directories, that's your signal. the software exists. it's just not worth the attention tax of posting. As an example: I wrote my own disk usage visualization, showed it friends, one used it, 2 stars on github. No one cared.

i will say that i expected a cambrian explosion of custom software and passion projects that hasn’t really materialised.

@eatonphil Sign up for Palo Alto Coding Agents Coffee Club. First 20 people who respond to the monthly email will get a calendar invite forms.gle/DRZESFjbvj8awH…

Just read @eatonphil's post on running a coffee club and it inspired me. Obviously my interest is around coding agents (working on Happy Coder after all). Did a deep research looking for coding/tech coffee meetups in the Bay Area near Palo Alto... found nothing. So I'm starting one! ☕️ Cafe Borrone - local gem, tons of space. Has tea & orange juice & more for non coffee drinkers. First meetup coming soon. Who's in?

@Cr4zy0_0 @VraserX have you seen the compute as teacher paper showing exactly how models can themself to be better?

@VraserX 1 million GPUs ≠ 1 million IQ points. Compute is scaling like crazy, but intelligence isn’t just FLOPs, it’s feedback loops, data freshness, and architecture leaps. ❗️The real acceleration happens when models teach themselves better, not when we burn more silicon. ❗️

GPT-5 was trained on 180,000 GPUs, a true powerhouse. By the end of this year, OpenAI will have over 1 million GPUs, according to Sam Altman. My bet: the GPT-6 training run starts right after that… and by mid-2026, we’ll be looking at AGI. The acceleration is insane.

@VraserX I'd double that to 2027 because things always tend to take longer than you expect.

@thsottiaux I'm really getting use out of the claude-code /plugin

Codex team is doing a weeklong hackathon starting next week, what should we build? Other than CLI auto-update, we're on that already

@ain3sh @testingcatalog Yeah they have taken the codex mobile app approach of running these tasks in the cloud, while Happy Coder lets you run claude code on any internet connected computer you have access too. On the other hand, they have raised the UI bar so we will be busy this weekend

@testingcatalog @PeoplesGrocers

BREAKING 🚨: Anthropic is preparing Claude Code to be released on the mobile app! Users will be able to connect Claude app to GitHub and run their coding prompts on the go. Claude Codex 👀

@Plinz Important context needed: Is this before or after initial screening? 20% working submissions from an open application form = surprisingly good. 20% working submissions after resume review + phone screen = concerning. Which scenario are we looking at here?

Evaluating coding submissions from prospective engineers; 80% are clearly vibe coded, 60% don’t actually work, 40% of the documentations do not make sense; this generation is ngmi

@forgebitz I wish these AI products would take another look at Elm. Just a beautifully simple language for modeling data and visualizing that data.

we need to go deeper we need an abstraction for this, like a language to write this code in like a coding language

@skirano There's loops in ChatGPT's agent builder too. OpenAI just simplifies the opening day messaging for billions of users vs Anthropic talking directly to developers. Different audiences, different language 🤷

It’s funny to see how much the concept of “agent” is diverging between OpenAI and Anthropic. OpenAI seems to define it as a strict, linear pipeline, while for Anthropic, an agent is an AI in a loop with its tools.