Andrew Edmond

Independent buy-side technical due diligence

linkedin.com/in/andrewedmond ↗
Southern Spain · Remote & onsite

← All posts

Get Camino build log · 2 of 10

The most productive day was the day nothing got built

Four rules, written before a single screen — and why invariants matter more when features cost minutes, not sprints.

Four invariantsWritten before a single screen
  1. 01

    Deterministic engine

    Same profile in, same plan out. No model in that loop.

  2. 02

    Derived interview

    Questions come from the obligation catalog — they can't drift from the data that answers them.

  3. 03

    Lola never invents

    The guide explains, sequences, and encourages. She cannot originate a claim.

  4. 04

    Plan is a pure function

    A user's plan follows from their profile. No hidden state.

The most productive day of the build was the day nothing got built.

Before Claude Code wrote a single screen of Get Camino, we wrote the four invariants above — rules the product must never violate, no matter how fast features arrive.

Here’s why this matters more in the AI era, not less: when features cost minutes instead of sprints, the bottleneck stops being “can we build it” and becomes “will it still be true after we do.” An AI will happily generate you a confident liar at incredible speed. Invariants are the difference.

The payoff showed up in surfaces that didn’t exist when the rules were written. When Spanish localization landed, invariant #3 extended to translations automatically. When guide pages shipped, it hardened into a build gate: any page whose prose contains a number its cited source doesn’t carry fails the deploy. That lint caught its first violation before a user ever saw it.

An invariant that keeps absorbing new surfaces is how you know it was architecture, not ceremony.

Your teams are about to move faster than your review processes can follow. The answer isn’t more review. It’s fewer, harder rules — written before the speed arrives.

This is one post from a longer build log. If you're weighing an AI-built codebase, that's the work I do — or find me on LinkedIn.