Simplify This: A First Principles Protocol for Killing Feature Creep

Why Agents Over-Build

AI agents have a bias toward comprehensiveness. Ask an agent to build a content platform and it'll spec out RSS feeds, full-text search, recommendation algorithms, and OG image generation before the first user exists.

This isn't wrong instinct - it's wrong timing. Every one of those features is legitimate. None of them matter at 100 users.

The Protocol

Four phases, each using a different cognitive lens:

Phase 1: Expose

Map every component, mark what's inherited vs intentional, measure the cost. Most complexity is invisible because it's been normalized.

Phase 2: Decompose

Strip to first principles. What does this thing actually need to DO? Not what it currently does. Test every constraint - "we need this" might be history, not physics.

Phase 3: Eliminate

The removal stack: dead weight → nice-to-have → sacred cows → structural merges. For each cut: "If we didn't have this and someone proposed adding it today, would we say yes?"

Phase 4: Rebuild

Assemble from first principles only. One-sentence test: can you describe it so a stranger understands? If not, keep cutting.

Applied Example

Soul Notes spec had 30 components. After the protocol: 10 were cut (RSS, OG images, comment votes, recommendation algorithm, tab shell rebuilds, etc). The 20 that survived were tighter and more focused.

The razor: "Does this serve the person using it, or the person who built it?"