7 posts
Fundamentals First, Part 7 — the capstone. Six posts, six fundamentals, and the same quiet move at the end of each one: I kept it in a skill. On why almost none of the leverage in these codebases is mine, the skills pulled straight off one public shelf against the single one I wrote by hand, and the last fundamental — knowing which judgment to encode yourself and which to borrow.
Fundamentals First, Part 6. When the feedback loops finally work, the AI ships faster than you can read — and the bottleneck slides off the machine onto the one thing that can't scale: your head. On the working-memory ceiling, the gray box that saved my brain, the tinkforge Split module I stopped reading, and why the door is a convention held by tests, not a fence held by the compiler.
Fundamentals First, Part 5. The AI builds the right thing and it still doesn't work — because it does too much before it checks. On the rate of feedback as your speed limit, separating the loops by latency, the tdd skill, and the archive change that inverted this blog's own test suite.
Fundamentals First, Part 4. The AI steps into your codebase with no memory and sees a web of modules that can all import each other — not the map you hold in your head. On Ousterhout's deep modules, the codebase-design / design-an-interface skills, and a .NET orchestration stack collapsed behind one door.
Fundamentals First, Part 3. The AI keeps re-explaining your own domain back to you, and the good words you finally agree on evaporate when the session ends. On Eric Evans's ubiquitous language, the domain-modeling skill, and tinkforge's CONTEXT.md — the glossary that stops the AI from talking past you, and from building things that shouldn't exist.
Fundamentals First, Part 2. The AI keeps building something other than what you pictured — because no one, you included, knows exactly what they want until they're made to say it. On Matt Pocock's grill-me skill, Fred Brooks's design concept, and earning a shared picture before you write a line of code.
Opening Fundamentals First — a series on building with AI. The premise: AI didn't make the codebase matter less. It made architecture, domain modeling, and judgment matter more.