Index
All Posts
018Getting Models to Return Clean JSON: Four Structured Output TrapsThe real traps are rarely about format. They're about fields: the JSON parses fine, the content isn't what you asked for.017The More AI Writes for Me, the More I Worry I'm Faking ItAI lets output run ahead of understanding. Code that works and a plan that sounds right don't mean you actually get it.016AI Tutors Are Hot, but Teaching Is the Wrong Job for ThemOverrated: personalized instruction replacing teachers. Underrated: getting teachers out from under repetitive work.015Giving the Model Examples: Four Counterintuitive Things About Few-ShotExamples beat a pile of adjectives. But the wrong ones are worse than none, because the model learns the traits you never noticed.014Claude Code: A Practical Command HandbookFrom slash commands to pipe scripts, the Claude Code commands that actually earn their keep every day, collected into a handbook you can copy from.013Writing Agents with Claude Code: Where Context Should FlowThe hard part of building an agent isn't orchestrating tool calls. It's deciding what goes into the context at each step.012A Prompt Is Not a Spell: It's a Spec for the ModelWrite your prompt like a task brief you'd hand a coworker. You get output you can count on.011Opus, Sonnet, Haiku: How to Actually Pick OnePicking a model isn't chasing the strongest one. It's trading off task complexity, latency, and cost.010Wiring Claude Code Into the Release Flow: 5 Real CasesThe value of Claude Code isn't in the flashy demos. It's in folding high-frequency chores into one stable, reusable workflow.009Minimalism in Front-End DesignWhitespace, restraint, content first. That's what Apple taught me.008RAG Is Not a Silver Bullet: When to Use It, When Not ToStuffing every question into a vector store won't make your AI smarter. It might make it dumber.007Context Engineering: Getting the Model to Work Like a ColleagueOutput quality isn't set by a single prompt. It's set by what you show the model and what you hide, step by step.006Is a Longer Context Window Always Better? Four Overlooked CostsA 1M context window sounds great. Actually ship with one and you find out longer is not better.005Vue 3 Composition API: Practical Tipsref, reactive, computed — which one to pick, in one read.004The Limits of AI in Medical ImagingMedical imaging was one of the first places AI actually shipped. Accuracy was never the thing holding it back.003Writing Quant Research with an LLM: A New Tool for AnalystsAI can't replace an analyst's judgment. It can replace 80% of the formatting grunt work.002Frosted Glass with backdrop-filterOne line of CSS and your cards feel like macOS.001My Dev Toolchain, 2026Editor, terminal, AI assistant — the tools you touch every day are worth the time it takes to sharpen them.