PET BRAINS

An indie publication for technical builders shipping with AI agents. Tools, methodology, build films — no demos, no fluff.

// LATEST // TOOLS // METHODOLOGY // BUILD FILMS

01
AI demos stop at the prompt. They never show the second-day work — the verifying, the rework, the regression a junior would have caught.
02
Generic AI workflows produce generic UI. Cyan-on-dark, gradient text, three-column feature card rows — the visual default since 2023, regardless of brand.
03
Specs rot the moment code is written. Most teams treat the document as input, then never reconcile what shipped against what was specified.
04
Open-source AI tooling is scattered across READMEs. There's no canonical place to read the methodology behind the tools, no editorial line that connects them.

// methodology · 03 / 03 documents

Most AI-coding content stops at the demo.

Pet Brains is the indie publication for the gap between AI capability and real engineering work. We publish the methodology — Document-Driven Development — and the open-source tools that implement it.

Document-Driven Development treats the LLM as a strong reasoner and an unreliable worker. Specs generate code, not the other way around. TDD cycles guarantee tests, self-review catches the agent before it commits broken work, and session memory keeps context across runs.

Three tools ship in v1: MVP-builder for greenfield projects, Design-builder for production-grade UI, Designlib-MCP for design knowledge over Model Context Protocol. Read each below — or skip ahead, subscribe, and we'll send the methodology breakdowns as they ship.

// 01

MVP-builder

MVP-builder is the Claude Code workflow we use for greenfield projects. Document-driven, ships a structured chapter list before any code lands — TDD cycles and a self-review loop keep the agent from committing broken work.

// 02

Design-builder

Design-builder is the Claude Code plugin that produces production-grade UI without the AI slop. Anti-pattern filters, motion principles, HIG references — four commands cover setup, create, improve, review.

// 03

Designlib-MCP

Designlib-MCP is a Model Context Protocol server giving AI coding agents access to a curated catalog of palettes, fonts, inspiration pages, and patterns. Works with Claude Code, Cursor, Copilot.