All writing
AI Engineering

Features Say What Code Means. Nothing Says Who Can Reach It.

I found a hole in the graph. Routes, CLI commands, cron jobs — the actual doors into the system. Entry points are now first-class artifacts, and dead code finally has a definition.

SA
Salah Al‑Nawah
May 9, 2026 · 2 min read
aicode-generationentry-pointslattice

I found a hole in the graph this week. I can ask “what does this code mean?” but not “how does the outside world get here?” Routes, CLI commands, cron jobs, queue workers. The actual doors into the system. That’s a different axis entirely.

So entry points are now first-class artifacts. Detectors parse the framework’s routing (Laravel first, since that’s the dogfood) and propose manifests: this route, this handler, flowing through these features. I almost let detections become facts directly. Caught myself. A parser’s guess is still a guess, so every proposal waits in an accept/reject loop. It’s annoying by design.

The payoff surprised me. “Which endpoints touch the refund feature?” used to be an hour of grep. Now it’s a lookup.

Dead code finally has a definition I can compute: a feature no entry point reaches is a feature nobody can trigger.

Open question: flows are declared. What stops a declared flow from lying about what the handler really does? I’m parking this. It smells important.