Coding with LLMs in the summer of 2025 – an update

A good way to get a model to answer questions about a codebase without overwhelming it or exceeding its token count is to: 1. just give it the directory structure 2. ask it questions based on that 3. after it answers a question ask it if there are any specific code files it needs to better answer the question you asked 4. attach only those files so it can confirm its answer and back it up with code

you can upload a zip of a feature files as they are in the project and ask to implement it on another platform on chatgpt 4.5, works pretty good.