Working with AI means accepting no bottlenecks. I don’t always exercise that instinct as much as I should. But a few days ago, I did! I had a local workflow problem: how to parallelize work when a single repository is ~50GB? I wanted multiple parallel workstreams, but I definitely did not want 10 full copies of the repo when I barely had disk space for one. Worktrees don’t solve it. So I built Mirage. It leverages APFS to clone a folder with virtually zero upfront disk cost, and then only pays as files are actually edited in a sweet CLI API. Suddenly BANG! I can spin up a bunch of “worktrees” fast and cheap. Now to the next bottlekneck... Github repo here: renanliberato/mirage

We are making our discount permanent! 🎉 Enjoy building with DeepSeek-V4-Pro and bring your innovative ideas to life! 🚀

Hey @justsisyphus what's the recommended model setup for oh-my-opencode using only opensource models?

Claude code leaks process if you close terminal tabs without exiting Claude first, until you either kill the manually or restart the machine

I’m using HAIKU because Sonnet is SO SLOW these days… I prefer to guide haiku than wait for Sonnet to reason every 10 files lines read or so

Computer scientist: Well, binary search is the most practical way. So for a value of 1 to 100 I’ll say … 50? Woman who asked me to guess her age:

isn't Opus on Claude Code thinking today? :sad: I'm on xhigh under a subscription and I see no thinking traces

Spent 1 hour trying to setup a @runpod pod to serve some coding model with no success, even with an LLM as copilot. Jesus.

when you're 20 minutes from reset and you have 50% usage left

any suggestion of markdown readers? It's weird to have to open vscode to read a .md generated by an LLM. Also my terminal windows are usually small, so nvim even thought with formatters isn't what I expect

RT @galligator: Me prompting my coding agents at 3am

claude code down?

Hey @VictorTaelin I’ve heard some buzz about “quint lang”. Is it something from the same domain of Bend/HVM etc?

Hooks are coming to codex. That’s all I wanted to say.

Wild times

🚨BREAKING: Alibaba tested AI coding agents on 100 real codebases, spanning 233 days each. the agents failed spectacularly. turns out passing tests once is easy. maintaining code for 8 months without breaking everything is where AI collapses. SWE-CI is the first benchmark that measures long-term code maintenance instead of one-shot bug fixes. each task tracks 71 consecutive commits of real evolution. 75% of AI models break previously working code during maintenance. only Claude Opus 4 stays above 50% zero-regression rate. every other model accumulates technical debt that compounds over iterations. here's the brutal part: - HumanEval and SWE-bench measure "does it work right now" - SWE-CI measures "does it still work after 6 months of changes" agents optimized for snapshot testing write brittle code that passes tests today but becomes unmaintainable tomorrow. Alibaba built EvoScore to weight later iterations heavier than early ones. agents that sacrifice code quality for quick wins get punished when consequences compound. the AI coding narrative just got more honest: most models can write code. almost none can maintain it.

Shipped games in both. Godot is good, but remember you won’t get Unity’s (IMO): - 3D runtime scene editor - better profiling tools - better runtime performance (instantiation etc) - some 3D model features - better UI system - better mobile build tools

Macbookがアコーディオンになるアプリ作った #メイカーズ長岡まつり