Since LLMs have been available, I've been doing a lot of coding using agents. Unlike shoving AI into every app, replacing customer support with chatbots, generating endless streams of slop, etc., agentic coding is actually useful. I still use VS Code as an editor, but now I do agentic coding itself using CLI-based tools that differ depending on the model I'm using. It's actually entirely changed the way I use LLMs for agentic coding. This might by fine for your personal gh use, but I wouldn't trust a coding agent with that access.