Claude Code inside the Claude apps really has advanced supremely, especially lately:
- Remote control
- Connections to Telegram, other messaging apps
- Session handoffs from desktop to remote
It’s clear that’s where a ton of work + future improvements have been delivered and will continue to be delivered.
However, I continue to find I prefer to use Claude Code within terminal for a number of reasons:
- Hooks, I still haven’t found a dependable way to deploy granular hooks that federate out to various events for Claude in all situations (Claude Chat, Claude CoWork, Claude Code, in & outside of apps)
- JSONL records from chats - difficult to capture / read except
- Observability - this is especially true when using inside of IDE-based Terminal sessions
and I’ve seen other Devs prefer the same.
Let me go thru these reasons and explain them a bit.
- Hooks: These allow me to trigger automatic scripts at session start & session end, as well at various other specific points within the agentic developer flow.
- JSONL Chat Logs: We can examine these locally for semi-automated improvements in our workflows when we examine token burn rates and adjudicate for quality (not just quantity) of outputs proximate to tokens consumed.
- Observability: I realize I’m in the minority here and most people just yolo within a terminal, but for me I prefer to have my terminal operating within an IDE; I can bring in VSCode plugins for some boots in productivity / reliability + see what files my agents have edited. I can navigate other files / dependencies while my agents are operating inside of terminal sessions
