Why I put a line budget on AI-written code

One AI-built project grew past 16,000 lines. After that, I added four rules that every agent now follows.

The four rules

• List assumptions before building. Wrong assumptions show up before they turn into code.
• A line budget for every task. The task says how big the change is allowed to be.
• A failing test before any bug fix. The bug is proven first, so the fix is proven too.
• A cut-down review before every PR. A separate review that only asks: what can be deleted?

Why a budget works

AI writes fast, and fast code grows. A budget turns "write the feature" into "write the feature in this much code", which keeps AI code small and readable. [PLACEHOLDER: how you set the budget for a task]

The 16,000-line project

[PLACEHOLDER: what the project was and what you cut]

The rules are part of the loop on How I build, and every 7-day sprint runs on them.