(Context: we’re trying out Claude Code at work)

I think it’s extremely problematic that these products are designed to stimulate anthropomorphism. I’d greatly prefer them to be more authentic, i.e. more machine-like.

@avi I haven’t done this or anything, but I know you can alter how it outputs (be terse, etc)… maybe prompt it? (FWIW, it’s my understanding the later models follow ‘do this’ (positive) better than ‘don’t do this’ (negative) phrasing.

@octothorpe see screenshot — looks like I’ll have to manually issue this instruction for every session, like an animal. Good thing I’m a heavy user of a clipboard manager.

Thanks for the suggestion!

ALT

@avi Hmm. I know Cursor has the notion of ‘Rules’ which are persistent in the sense that they are automatically read from disk at every prompt. IIRC Claude Code also has a similar function. Aha! this may help:

https://www.anthropic.com/engineering/claude-code-best-practices

@avi Also, IIRC, the response it gave is because the agents aren’t allowed to write their rules (I reckon for security). They can however be asked to write it and spit it out into the chat window so you can plug it into an md/mdc file.