Code We Understand
A Field Manual for Building Software With AI
A practical, disciplined method for turning AI from a code generator into a reliable collaborator — and shipping software you actually understand.
Most advice about AI coding is either too optimistic or too vague. Code We Understand argues for a different approach: use AI aggressively, but only inside a disciplined method that keeps the human in command of the codebase. The goal is not to generate more code. It is to build software you can explain, maintain, and trust.
Who it's for
- Developers who want to use AI without losing control of the code.
- Technical founders and operators building real tools.
- Thoughtful beginners willing to learn software and AI practice in parallel.
What you'll learn
- How to plan before implementation instead of prompting blindly.
- How to probe real data before building against assumptions.
- How to verify outputs end-to-end instead of trusting passing tests.
- How to keep context, notes, and commits working together.
- How to recognize when AI is helping and when it is hurting.
The seven practices
At the core of the book is a structured method — seven practices that keep you, not the model, in charge of the codebase.
- 01Load context
- 02Plan before code
- 03Probe real data
- 04Verify with reasoning
- 05Verify end-to-end
- 06Commit in small increments
- 07Stop when the work is done
Inside the book
The book moves from first principles to practice. It opens by distinguishing the common failure modes of AI-assisted development, then lays out the method chapter by chapter. The final section walks through the construction of a real project — a companion tool called Tidepool — so each practice is seen applied to actual work, not only described.