Niklaus Wirth’s classic Project Oberon system has been migrated from the bespoke RISC‑5 FPGA CPU to the open‑source RISC‑V ISA. The repo delivers a complete stack, kernel, compiler, and graphical UI, so the historic OS can now be explored on any RISC‑V hardware or emulator. This opens a practical teaching platform and preserves a seminal OS design for modern developers.
The GitHub project compiles the original 1989 Word for Windows 1.1a source into a native 64‑bit Windows executable, replacing 16‑bit assembly with modern C/C++ and mapping Win16 APIs to Win32. It runs the classic UI unchanged, includes a full test suite, and preserves a piece of software history for developers and researchers.
os8088 boots from a floppy into a fully windowed desktop, offering overlapping windows, menus and pre‑emptive multitasking on an original IBM PC/XT with just 256 KB RAM. Written in real‑mode NASM assembly (AI‑assisted) and open‑sourced on GitHub, it proves a modern GUI is possible on 8088 hardware.
us-vs-them is a CLI tool that diff-analyzes a git-tracked file to assign a human-authorship score to each line, flagging which parts were written by a person versus an AI agent. It lets you enforce human-owned “islands” in code or docs, preventing downstream models from overwriting critical sections. Works on plain text, no markup needed.
The author set out to own a portable AI coding‑agent harness, only to discover that OpenAI’s Codex Desktop offered a far more practical solution. The post weighs the trade‑offs between full control and lock‑in, showing why portability often means accepting a proven, integrated tool. It’s a reality check for anyone eyeing a DIY agent stack.
VCs are inflating the idea that AI will wipe out whole service sectors to hype a new investment story. In reality, buyers see AI as just another software layer, demanding transparent pricing and costs that undercut human wages, not a magical cost‑free replacement. The gap between hype and demand reshapes where AI actually adds value.
Anthropic switched auto mode to the default for Claude Code’s Pro, Max, and Team plans, betting the agent can edit files and run commands without constant user prompts. Internal tests showed auto mode blocked 89% of risky actions that humans approved, and a third‑party audit recorded zero successful prompt‑injection attacks.
A randomized trial by METR gave 16 experienced developers access to Cursor Pro and Claude models for real open‑source tasks. They estimated a 24% time saving, but logged data showed they actually spent 19% more time. The gap shows self‑reported AI productivity gains can be wildly inaccurate.
Writing to a modern database may feel instantaneous, but the engine shoves compaction, indexing, and replication into background jobs. That trade‑off boosts latency for the next write burst and can surprise operators when resource limits are hit. Understanding this shift helps you tune throughput and avoid hidden performance cliffs.
The post shows how starting with project constraints, static content with pinpoint interactivity, lets you avoid heavyweight SPA frameworks and ship near‑zero JavaScript. By using tools like Astro’s islands of interactivity, you cut bundle size, boost performance and keep the codebase lean.
They moved oversell protection from Redis to MySQL using per‑unit rows and SKIP LOCKED, achieving required throughput during Black Friday 2025's $5.1 M/min sales, while improving consistency. The shift shows MySQL can handle high‑contention inventory workloads, challenging the “use a cache” dogma.
An old CMF Phone 1 now runs Surf, a finance tracker, screen‑sharing and several web apps, surviving reboots and network switches. Its 8‑core ARM CPU, 8 GB RAM, Wi‑Fi 6 and built‑in battery replace a small Hetzner VPS, and the author shares what hardware tricks work and which Linux‑flashing attempts fail.
You’re caught up.
That was all 12 stories in the 9 August issue.