Field notes

AI coding · August 18, 2026 · 6 min read

How we build fast with AI coding — the workflow behind these apps

These products were built by a small team moving quickly with AI in the loop. Speed like that only works with guardrails. Here's the actual workflow — specs first, tests always, human review at the edges.

AI coding tools make it possible to build a lot, fast. They also make it possible to build a lot of unmaintainable, half-working code fast. The difference is entirely in the workflow around the tool. Here's ours.

Spec before code

We write the technical spec first — the data model, the route map, the integration points — before generating a line. A model given a clear spec produces clean, consistent code; a model given a vague ask produces a plausible mess. The spec is the steering wheel.

Tests are non-negotiable

Every project ships with a real test suite, and our marketing dashboard alone runs hundreds of automated tests. When AI writes a lot of code quickly, tests are how you keep it honest — they catch the regression the model didn't foresee, and they make the next fast change safe.

Type-check, lint, build — every change

TypeScript, ESLint, and a production build run on every change before it merges. This trio catches the class of errors AI is most prone to — a wrong type, an unused import, a route that doesn't compile — in seconds, not in production.

The human owns the edges

AI moves the middle of the work fast: scaffolding, wiring, refactors. Humans own the edges that matter — the architecture decision, the security boundary, the thing that goes live. That division is why the pace doesn't cost quality. The same principle that governs our products governs how we build them: automation does the volume, a person owns what's irreversible.

How we build fast with AI coding — the workflow behind these apps — Kruzeniski Studio · Kruzeniski.ai