Stop Pretending You Understand. Your Codebase Is Too Big.

The video argues that believing you must fully understand a large codebase is a misunderstanding of how modern software works. For systems with millions of lines of code and high turnover, completeness is impossible and counterproductive. The real skill is having an intuition for where things belong, not memorizing every line. This is supported by

examples from working at Twitch and Amazon. The speaker defends 'partial understanding' as the default state for engineers on large systems. The key tradeoff is between pure engineering ideals (maintaining a perfect mental model) and the impure reality of shipping features on time. Good codebases are designed so that incomplete knowledge doesn't block work, relying

on good architecture and tooling like type systems. The discussion extends to AI, noting that coding agents experience '100% turnover' every new thread, which makes them a test case for working without historical knowledge. The speaker finds this both challenging and rewarding, as it forces engineers to focus on steering outcomes rather than micromanaging implementation

details. The core tension is between the desire for a perfect theory of the program (from Peter Naur's paper) and the practical reality of large-scale systems where nobody has one. The conclusion is that modern engineering is about managing partial understanding, not achieving omniscience. This is a useful reframe for leaders dealing with legacy systems