Designing for Forgotten Context

Anti-fragile patterns for LLM-native codebases. A 5-part series on building systems that get stronger — not weaker — the more AI sessions touch them.

APRIL 2026  ·  5 PARTS  ·  30 MIN TOTAL

Every team using AI coding tools hits the same wall. Greenfield is magic. Edits break things. The standard advice is better prompts, more context, longer memory. That advice is wrong.

The problem isn't the model. It's that your codebase was designed for humans who remember, and LLMs don't. The fix is architecture — structural patterns that make forgetting safe, interchange contracts that prevent silent integration failures, and a meta layer that catches drift before it compounds.

This series comes from operating a production system with 75+ tools across three machines, modified by AI sessions daily for over a year. These aren't patterns I think would work. They're patterns I run on.

The Series

Part 1 — The Problem

Why LLMs Are Great at Greenfield and Terrible at Your Codebase

LLMs don't forget your codebase. They never knew it. Here's why that distinction matters — and why the fix isn't better prompts.

5 min read
Part 2 — Structural Patterns

One File, One Job: Design Patterns for LLM-Native Codebases

Four structural patterns that make your codebase safe for AI modification — not through documentation, but through architecture.

7 min read
Part 3 — Integration Boundaries

Interchange Formats: How Your Tools Should Talk to Each Other

Most AI-generated code fails at integration boundaries. Contracts, event logs, and manifest-driven dispatch fix this structurally.

6 min read
Part 4 — The Meta Layer

The Session Problem: Living Documentation and Architectural Health

Even good architecture rots when AI sessions keep touching it. End-of-session hooks, nightly reviews, and architectural health monitors keep the system honest.

6 min read
Part 5 — What It Means

What This Means for Your Team

The gap between “we use AI coding tools” and “we design for AI collaboration” is about to become the most important differentiator in software.

6 min read

I build these systems for companies.

If your team is using AI coding tools and hitting the wall described in this series, let's talk about what your architecture actually needs.

Book a 30-Minute Call