Friday, March 21, 2026

Simon Willison published two pieces today: a "mildly dystopian" experiment profiling HN users from their comments, and a practical guide to using Git with coding agents. The Tinybox from George Hotz's tinygrad hit the front page hard with 597 points. Someone ran an LLM on a PlayStation 2, which is the kind of Show HN energy we live for — a PS2 running AI in 2026. And Atuin v18.13 shipped with AI features for your shell history.

Things People Built

Show HN: I ran a language model on a PS2

This is the good stuff. Someone got a language model running on a PlayStation 2 in 2026. No practical reason to do this. That's what makes it great.
Hacker News

Show HN: AI SDLC Scaffold – repo template for AI-assisted development

A GitHub repo template that sets up your project structure for AI-assisted software development. If you're trying to get coding agents to be useful on real projects and keep hitting the "agent doesn't understand my repo" wall, this is a structured starting point.
Hacker News

Guides & Practices

Profiling Hacker News users based on their comments

Simon Willison calls this a
mildly dystopian prompt
— feed an LLM someone's last 1,000 HN comments and ask it to profile the user. A public API makes pulling anyone's comment history trivial. Simon tested it on himself first, which is the responsible way to do something slightly creepy.
Simon Willison

Using Git with coding agents

Part of Simon's growing Agentic Engineering Patterns guide series. The core idea: coding agents are good at Git, so let them handle the version control busywork while you stay focused on what you're actually building. It becomes a safety net — easy to undo anything the agent gets wrong.
Simon Willison

How to attract AI bots to your open source project

Andrew Nesbitt on making your open source project AI-agent-friendly. 181 points on HN, which suggests this hit a nerve. If agents are going to be submitting PRs and filing issues, the projects that make themselves legible to those agents will get more contributions. Whether that's good or not is left as an exercise.
Andrew Nesbitt

Tools & Hardware

Tinybox – A powerful computer for deep learning

George Hotz's tinygrad project has the Tinybox — a dedicated deep learning computer. 597 points on Hacker News, which is a lot of interest. The pitch is a self-contained box purpose-built for running AI models, not a cloud bill.
tinygrad

Atuin v18.13 – better search, a PTY proxy, and AI for your shell

Atuin is a shell history replacement that syncs across machines. Version 18.13 adds AI features, improved search, and new terminal features. If you're not using Atuin yet and you work across multiple terminals, it's one of those tools that immediately feels essential once you try it.
Atuin Blog

Also

    Today’s Sources