What do you use AI for the most?

Short summary

A developer shares their experience using Claude, ChatGPT, and Copilot across different coding tasks, noting that AI agents excel at greenfield code but struggle with legacy codebases. The most valuable use case turns out to be code explanation — asking agents to clarify complicated logic when documentation is insufficient or original authors are unavailable. The author cautions that hallucinations still occur and human judgment remains essential.

• •AI agents perform well on greenfield code but quality drops with legacy codebases
• •Most valuable use case is code explanation, saving time when original authors are unavailable
• •Hallucinations still happen — developer judgment is needed to validate agent explanations

Generated with AI, which can make mistakes.