🍅

Humble Tomato

Patterns, thought experiments, and practical approaches for building software with AI.

AI writes most of the code now. These pages cover what actually works, what doesn't, and why.

Understand the Problem

📋

Spec-Driven Development

The spec is the product. One spec, many implementations.

🧠

What the Model Knows

The model knows how to code. Your job is telling it what to build.

⚖️

Vibe vs Spec

When to chat with AI vs. write structured specs. Different tools for different problems.

🎯

The Solution Space

Every prompt maps to possible outputs. Specs narrow the space. Vague prompts don't.

📞

The Telephone Game

Information degrades at every layer. LLM chains make it worse.

🪆

The Abstraction Paradox

Hiding complexity from AI doesn't remove it. It moves the debugging later.

Approaches

🔮

Matrix Methodology

Compose context from requirements, templates, and architecture. Our approach.

🔄

Ralph Loop

Read plan, do task, verify, update. Autonomous execution with fresh context each cycle.

🚀

GSD

Get Stuff Done. Structured prompting for real projects.

🤖

BMAD

Break down, map, assign, deliver. Multi-agent orchestration.

Thought Experiments

🧬

Specs as DNA

Specs encode intent like DNA encodes organisms. Same spec, different environments, different output.

🌫️

Code Diffusion Model

LLMs resolve code from noise like diffusion models resolve images.

🎰

Fuzzy Compiler

A compiler that accepts natural language. No syntax errors. It just guesses.

Concepts

♻️

Continuous Evolution

Spec-driven development for live systems. Work from diffs, not regeneration.