Top Stories

Building and Shipping Mac and iOS Apps Without Opening Xcode

489 points · scottwillsey.com

For anyone who’s wrestled with Xcode’s sluggishness and opaque build system, this is catnip. The author walks through a full workflow — editing, building, signing, and shipping real Mac and iOS apps — using command-line tools and a lighter-weight editor instead of Apple’s IDE. It taps into a growing appetite among developers to escape heavyweight, GUI-bound toolchains in favor of scriptable, reproducible setups. The HN crowd loves anything that turns a black-box vendor tool into something you can automate and version-control, and the thread is full of people sharing their own xcodebuild incantations.

Show HN: Clawk — Give Coding Agents a Disposable Linux VM, Not Your Laptop

203 points · github.com

As AI coding agents get more autonomous, letting them run shell commands directly on your machine starts to feel reckless. Clawk’s pitch is simple: hand the agent a throwaway Linux VM instead of your actual environment, so a bad rm -rf or a hallucinated install script can’t wreck your setup. It’s a neat encapsulation of a very 2026 anxiety — we want agents to do real work, but we don’t fully trust them yet. The discussion digs into sandboxing tradeoffs, how this compares to containers and existing agent runtimes, and whether disposable VMs become standard practice as agentic workflows go mainstream.

The Git History Command

324 points · lalitm.com

A deceptively simple post that struck a nerve: a walkthrough of building (or wishing for) a better git history view than the usual git log gymnastics. Developers spend an enormous amount of time archaeology-ing through commit history, and the piece resonates because everyone has their own pile of aliases and half-remembered flags for this. The comments turned into the usual gloriously nerdy git thread — people sharing custom log formats, tools like tig and delta, and strong opinions about what a good history-browsing experience should actually look like.

Japan Develops a Method to Recover Up to 90% of Lithium from Used EV Batteries

526 points · tech.supercarblondie.com

The day’s most-upvoted story taps into a real bottleneck for the electrification push: where all the lithium comes from, and where dead batteries go. Japanese researchers claim a process recovering up to 90% of lithium from spent EV batteries, which — if it scales — could ease supply-chain pressure and cut the environmental cost of mining. HN readers are appropriately skeptical of press-release chemistry and want to see energy inputs, cost per kilogram, and whether “up to 90%” holds outside the lab, but there’s genuine interest given how strategically important battery recycling has become.

The Economics of Recursive Self-Improvement

103 points · elasticity.institute

A meaty paper applying an economic lens to one of AI’s most-debated ideas: an AI system that improves itself, then uses the improved version to improve further. Rather than treating recursive self-improvement as a sci-fi certainty or an impossibility, it models the costs, diminishing returns, and constraints that would govern such a loop in practice. That framing is exactly what the HN audience appreciates — it turns a hand-wavy debate into something with variables you can argue about. Expect a thread that swings between AGI-timeline hawks and doves, all citing figures from the PDF.

Alternatives to Run CUDA on Non-Nvidia Hardware

51 points · hpcwire.com

CUDA is the moat that keeps Nvidia dominant in AI, so any credible path to running CUDA code on other hardware is a big deal. This piece profiles Spectral Compute’s effort to “set CUDA free,” letting workloads target AMD and other accelerators without a full rewrite. Whether it succeeds is an open question — plenty of translation-layer projects have foundered on the last 10% of compatibility — but the strategic stakes are enormous given GPU shortages and the hunger for cheaper inference. HN commenters weigh in on ROCm, portability layers, and just how sticky CUDA really is.

YouTrackDB — A General-Use Object-Oriented Graph Database

132 points · github.com

JetBrains open-sourcing a graph database got attention both for the tech and the provenance. YouTrackDB is an object-oriented graph database extracted from the guts of JetBrains’ YouTrack issue tracker, now offered as a general-purpose store. Developers are curious how it stacks up against Neo4j and other graph engines, and there’s the perennial question of whether a database born inside a specific product generalizes well. The JetBrains name buys goodwill, and the thread digs into the data model, query language, and licensing.

Linux 0.11 Rewritten in Idiomatic Rust, Boots in QEMU

108 points · github.com

Rewriting historic C code in Rust is a beloved HN genre, and this one delivers: a from-scratch reimplementation of Linux 0.11 — Linus’s early kernel — in idiomatic Rust that actually boots under QEMU. It’s part educational exercise, part demonstration of how a memory-safe language handles the gnarliest low-level systems code. Commenters debate what “idiomatic” even means at the kernel level, how much unsafe is unavoidable, and whether projects like this are the best way to learn OS internals. Either way, it’s a fun artifact of the ongoing Rust-eats-systems-software trend.