Theo Browne challenges the idea that effective engineers must understand every line of a large codebase. Responding to an essay about partial understanding, he distinguishes knowing implementation details from having enough architectural intuitionSoftware architecture is the organization of a software system into components, responsibilities, and relationships that shape how it works and changes. to locate the right subsystem and make a safe change.
Theo Browne applies this distinction to coding agentsAn AI coding agent is a tool-using AI system that can inspect, modify, and validate software within a repository., which often begin a new session without the historical knowledge of previous workRepository context is the relevant source, configuration, history, documentation, dependencies, and worktree state needed to understand a software task.. He argues that clear structure, bounded changes and tools that answer questions about existing code can help both agents and unfamiliar developers contribute.
Theo Browne warns that an AI-generated explanation is not evidence that a rewrite is justified. Existing systems preserve unusual user requirements and accumulated behavior, so learning one flow and improving isolated sections is often more realistic than replacing everything at once.
Theo Browne describes using AI to build a SwiftUI version of an existing mobile app without reading every implementation detail. He attributes his ability to steer it to knowledge of the original system, its data flowsData flow describes how information moves through a program, including where it enters, changes, is stored, and produces an output. and failure cases, presenting this as a trade-off in understanding rather than permission to abandon engineering judgment.
Watch on YouTube
