Users support testing AI coding agents on preserving intent while navigating large codebases rather than just generating code, as this better matches real-world demands.
Based on 1 visible X reactions from 2 accounts; directional sample.
Ask a question below.
Published answers will appear here.
@davidbau This is the right test shape for coding agents: not "can it generate code?" but "can it preserve intent while navigating a large, hostile legacy surface?" The reading/comprehension bottleneck is where most agent demos quietly stop being demos and start being engineering.
Is it possible to write 100,000 lines of code well, if you do not read it? Let's go Hunting Zombies! https://davidbau.com/archives/2026/07/16/hunting_zombies.html In this post I dive into the code of two AI agent contestants in the Teleport coding challenge to learn their secrets. Very fun. And also instructive.
This Spring I'm going to wade into these waters by teaching a course on large-scale vibe coding: on writing more code than you can read... Looking for ideas. Post here with specific lessons that you think we should be teaching the next generation of computer science students.
Users support testing AI coding agents on preserving intent while navigating large codebases rather than just generating code, as this better matches real-world demands.
Based on 1 visible X reactions from 2 accounts; directional sample.
Ask a question below.
Published answers will appear here.