
Artificial
AI coding agents are getting better at writing code, but I'm not convinced they're getting better at understanding codebases
I've been using Claude Code, Cursor and a few other coding agents quite a bit recently. One thing that keeps standing out is that generating code isn't really the bottleneck anymore. Understanding the codebase is. Agents can usually find the relevant file. The problems start when the change depends on: historical decisions undocumented relationships ownership boundaries files that always change together Bigger context windows help, but I'm not sure they solve this problem compl...
Read the full discussion on Reddit
This article was aggregated from Reddit. Click to join the conversation.
View on Reddit