Two publications on Medium
Switch between The Builder’s Playbook and The Pull. Playbook is selected by default — recent stories, then the full piece on Medium.
40 articles in ReactBD
Oct 30A key in .env goes public when Expo bundles it. I built 52 screens on mocks and kept Claude on the server.
Oct 30Note 8 puts serving at 14.6 cents per revenue dollar. On my Claude Code logs, 3 June to 15 October, that’s $2,019 against $900 paid.
Oct 29Trending is a score, not usage. The downloads field is a 30-day count, per Hugging Face’s docs. How to sort by what people actually pull.
Oct 29My Mac has llama-cli 0.5.0 and no .gguf files. Homebrew’s current build is 0.7.0. One -hf flag is the step the install skips.
Oct 28My model list is dated 30 September. A week later Anthropic shipped a cheaper Haiku and halved Sonnet’s cache reads. My file got neither.
Oct 28gpt-4o-mini lists at $0.15 in and $0.60 out. GPT-6 Luna is $0.10 and $0.50. The old string is now the pricier guess.
Oct 27I measured every install and build cache in my GitHub folder. Six repos hold most of node_modules, and five .next caches are free space.
Oct 27Agent tutorials start with a framework. I wrote the loop by hand on Claude, and the real risk was the deploy.
Oct 27Three footnotes, three different tests, all run in August on an M5 MacBook Air. My MacBook Pro runs Chrome 155. What each claim measured.
Oct 26SDK 58 shipped for iOS 27 with React Native 0.88. A non-interactive create now pins the newest SDK, which store Expo Go may not open.
Oct 04Every route to an AI catastrophe, checked against this year’s record. Then I counted my own AI assistant’s tools: 200, and 39 that act.
Oct 03PHP lost as much of the web in the last 21 months as in the ten years before. Where the new code went, and why mine went to TypeScript.
Oct 03A beginner-to-advanced guide built on a real build: what each tool does, what it costs, the prompts that work, and what changed between…
Oct 03Next.js 16 moved next dev into .next/dev. The turbo.json most of us copied caches it with every build, and 10 of my 11 repos still do.
Oct 02The “fired within a week” figures come from résumé-site polls. Official data show a different problem: fewer entry-level hires, and less…
Oct 02Court papers, a leaked plan and three consumer regulators show how Microsoft is selling AI.
Oct 01
Oct 01The brain-rot claims don’t hold up. The real risk does: you lose the skills you hand over. So I counted how much I hand over, in 1,141…
Oct 01It’s in System Settings, one click deeper than most people look.
Sep 30Handing out work, checking it and answering for it is management, and management has a limit on how many reports one person can supervise…
Sep 20Apple shipped fm in /usr/bin — a local model, a token counter, and an OpenAI-compatible server. Every subcommand refuses until someone…
Sep 19English is a UI for intent. TypeScript, Python, and Rust still do the real work — and the builders who win design the seams.
Sep 18It doubled since May, changed hands, grew a plugin and a product pitch, and has 97 open pull requests. What the four rules mean and how to…
Sep 18Auto-invoke skills, /init CLAUDE.md rot, listing budgets that mute new skills, and a “slash-only” flag that sometimes lies back. A setup…
Sep 17iOS 27 ships Siri AI with Apple Intelligence on the new Pros. The product is personal context plus privacy — not another cloud chatbot…
Sep 17macOS 27 shipped Monday without Intel. The toolchain left first, and one GitHub Actions decision explains why.
Sep 16Passkeys, 2FA codes, breach alerts, shared vaults, Chrome and Windows support. Preinstalled on every Mac, and most people run it at 30%.
Sep 16Progressive disclosure only works if you author for it: keep skill bodies thin, references in side files, and descriptions as selectors…
Sep 16Cursor, Claude Code, Midjourney, ElevenLabs, and Photoroom did not win by shipping demos. They won by owning a paid workflow. Here is the…
Sep 15The 5× posts keep the same 120 ms table. I measured a 183-byte object instead. The useful number is not a speedup. It is which tax you are…
Sep 13
Sep 12Leaderboards do not disagree because agents are mysterious. They disagree because each bench measures a different verb. Pick the verb, or…
Sep 12Progressive disclosure hides the body. It does not hide the rent. Here is how to see which skills are earning their keep, and which ones…
Sep 11I gave the same ticket three seats — planner, worker, reviewer — on different models. They argued. That argument is the architecture.
Sep 11Not Claude-only. Portable skill patterns that earn rent on Claude Code, Cursor, Codex, and anything that reads SKILL.md — plus the stack…
Sep 10
Sep 10The gray bar is a janitor’s closet macOS cleans on its own schedule.
Sep 09Why the people who build software over-choose a minority operating system: the three-legged design Windows and Linux each only half have…
Sep 09Shortcuts quietly gained a Use Model action that hands you Apple’s local model as a building block. Seven builds, exact prompts, and the…
Sep 08Apple offers lunch money for it. Left on a shelf, it replaces cloud storage, camera plans, a password manager, and chunks of your…
ReactBD