
It’s damn good. It understands my code base, uses, how the things work. If it doesn’t it will search for examples, it will search the internet, github, get source code, look at different options. I want you to understand how good it is. So, I am attempting to create a map editor like StarCraft 2’s level editor, using Rust language and the Bevy game engine 0.16.1 (the latest version). Bevy is a relatively young engine with fast paced development and it is all code, no editor, and it has breaking changes constantly. Every 3 months a new version comes out. I went and overspent a bit too much on AI’s and so I have both Claude and ZenCoder. When Claude Sonnet 4.0 makes mistakes and can’t figure it out, ZenCoder is the one that can figure it out and fix them, explaining along the way. It’s come along quite good very quickly. ZenCoder hasn’t been trained on this bevy engine code, but it reads the code base, understands the code, the context. Highly recommended. I have found one little niggle weird. If I gave it rules such as use code that is only compatible with Bevy engine 0.16.1, it would use older variations of bevy engine code incompatible with bevy 0.16.1. But when I give it no rules, it uses code compatible with bevy 0.16.1. I have found the same to be true with all other LLMs as well so it’s not just a ZenCoder thing. No rules is the better option.
What's great
context aware (1)complex problem solving (3)AI coding agent (4)codebase understanding (2)
Report
58 views



