AI Architecture
How I build software by orchestrating AI instead of hand-writing it. The approach is the same on every project: settle the architecture up front and write it down as enforceable rules, let agents apply those rules on a schedule, and stay the escalation path for the cases the rules do not cover. The decisions that are expensive to reverse get made before any code is written, not discovered halfway through. Click any tile to see the mechanism.
Human in the loop
The rules and CI gates are robust enough that tight low-level PRs auto-merge once they pass. I review the high-level and architectural PRs by hand, resolve escalations from the decisions ledger and accept or reject each auto-call in the weekly review. Human QA testing catches whatever the rules don't. The rule set grows exactly where my judgment is exercised.