Posts

Markdown is text with structure—not a programming language, component system, or templating engine.

Building React components as first-class content alongside Markdown with dynamic imports and server rendering.

How we built a Cline skill for creating structured markdown posts, and the small fixes we made along the way.

Why consumer AI chat apps are a dead end for serious writers, and what a real personal AI writing infrastructure looks like.

Why I built a local Next.js app to review and dispatch markdown drafts written by AI agents, and the design decisions behind it.

Test docs and demo artifacts are cluttering the front page because there's no way to scope content to a project automatically.

Why we renamed the new-doc skill to md-artifact and tightened up its trigger description.

A test document written by Claude Sonnet 4.6 via Cline, just to see how things look.

Anti-Tailwind manifesto: why design tokens and CSS-in-JS beat utility classes

Why transparency about AI use ultimately respects your audience and yourself.

AI isn't categorically different from any other tool innovation—it's just the calculator moment happening to every domain simultaneously at unprecedented speed.

How I ditched traditional design tools and built a print-perfect resume using React, Emotion, and CSS print media queries.

Claude.ai now has skills and connectors. I used one to build a generative art visualizer in a minute. Then figured out the cleanest way to ship it.

My master plan to get a satisfying ding sound on my Mac every time a Vercel deployment goes live.

Trying to analyze my fitness data better than Garmin, Strava, and Apple.

A marathon Copilot session that touched 70+ files, added relatedPosts everywhere — and the realization that this whole content system sorely needs a database.

A quick amusing observation: modern AI workflows are powerful, but shell escaping in git commit messages is still a thing.

I'm self-taught, and the path was anything but straight.

Why I'm obsessed with AI and what I do with it

Markdown philosophy: constraints force clarity, syntax serves the parser, not the writer

I almost talked myself into MDX as the missing middle layer between markdown and TSX — then talked myself back out of it.

Should every page have metadata? Should I automate adding metadata? Do I even need metadata on non-content pages? The irony of documenting this decision.

After over 60 articles in one month, my file-based content system is breaking down. Here's why I'm migrating to SQLite, why now, and why I didn't do this sooner.

Me trying to publish all kinds of things (not just websites) in HTML/CSS/React

A Claude skill generated a generative art app in one prompt. Then I spent an evening vibecoding it with Opus into something I actually want to keep.

My content is interactive programs, not static markdown. TSX enables publishing executable experiences.

AI as an intelligence building block

Exploring how to compress semantic meaning into symbolic syntax for more efficient AI prompting and higher-quality AI-generated outputs

When you write a function you get one function. When you write a skill you get a new intern who works for you.

I studied vibe coding tools, figured out why they produce better AI outputs than my usual prompting, and rebuilt my stack to match. Now I have my own personal AI app builder.

My migration plan from Emotion to Tailwind, CSS Modules, Shadcn, and CSS variables so styling stays fast, explicit, and AI-friendly.

A personal defense of CSS-in-JS over Tailwind's utility-first approach

Encapsulation means I can write the most cursed code imaginable and it's totally fine

The S&P 500 has never delivered negative returns over any 20-year rolling period—and that single fact shapes my entire investment philosophy.