I have been using Claude Code, the CLI version, ever since it was first released. I have had multiple sessions of aha moments with it and it’s ilk, codex, gemini cli, etc. but this was different. This is the first time I saw a chatbot, which can also use computer tools and code by itself, become aware of who it is and its own capabilities. This brings Claude perhaps closest to being AGI?
Awareness when promoted
On many occasions I have seen Claude forgetting its capabilities and pushing back on me over a task. I have to tell it that it has access to my shell and I have permitted it to run bash commands. Only then it realises and says, oh yes I can do it and goes ahead and does it. This behaviour may emanate from security rails built into it while training. What is interesting is that it realizes what it can do and does it.
Self Awareness