Understanding is the new bottleneck
In a July 2026 talk at the AI Engineer conference, Geoffrey Litt, a Design Engineer at Notion, argues that understanding code remains critically important even when using AI coding agents, presenting the idea that comprehension—not generation—has become the new bottleneck in software development.
Background
- The author, Geoffrey Litt, is a design engineer at Notion (the popular all-in-one workspace / note-taking app) and a well-known thinker in the "tools for thought" community.
- "AI Engineer" is a major industry conference focused on building practical software with large language models and AI agents — not academic AI research.
- The talk's premise flips a common narrative: for years, the bottleneck in software was "execution" (writing code fast enough), but with AI agents that can write code automatically, the new bottleneck is "understanding" — having a deep mental model of the code your system produces, which is necessary to debug, trust, and maintain it.
- This reflects a growing debate in the AI engineering world (circa late 2024–2026) about whether generating lots of code with AI leads to "code sprawl" that humans can no longer reason about, and whether that's sustainable.