@RaulJuncoV So true. People tend to put AI inside their software and try to make everything deterministic with complex loops they manage. Codex and Claude Code already hve this. Far better to build a CLI to some registry and let Claude just drive that with instructions and a clear objective.
@PhilipJoubert@AnthropicAI I combat this by constantly asking the AI to push back and or be ruthless Elon Musk. The models are obviously designed to make humans feel good, otherwise people won't hit the slot machine. ๐ฐ
@xai there is a bug in @grok. When scheduling a task, the button to click is invisible. I couldn't find it, and then realised it's a very light colour.
@elonmusk please build a version of Cursor but with Grok.
The Grok interface is way better that Grok in Cursor, but it's hard to use and keep in sync. ๐
@mckaywrigley@elonmusk@grok@xai We need a generic CLI that can do what Claude Code does, but work with any AI model. That way, you can easilys switch between models without swapping tooling every time.
@Suhail Run 4 Claude code terminals on the same codebase. Let one of them drive as project manager, and the rest as developers. Since it can access all the native tools on the machine, it's significantly faster at building things when compared to Cursor. Use Cursor as a window only.
@OfficialLoganK@geminicli Nice strategy with free requests. I instantly dropped Gemini when Opus launched and moved my entire flow to Claude Code. But I'll give this a go as well.