Desperate Claude

Published 2026-09-28

Considering the supposedly unstoppable growth of LLM-powered software development, Claude Code gives off some consistently desperate vibes.

Do you like me? Optional—but appears several times a day even if dismissed.

● How is Claude doing this session? (optional) 1: Bad 2: Fine 3: Good 0: Dismiss

Not sure what "getting to finished work sooner" really means, I guess there's some reason they want me to try their new model.

Get to finished work sooner with Opus 5.5. Switch anytime with /model.

Interrupted. How about Claude just... stops?

Interrupted · What should Claude do instead?

Only 5 agents?

Tip: You can control how big a workflow is just by prompting. Ask for a small workflow, cap it with "use at most 5 agents"

And then we have the constant end-of-turn nagging to "shall I do this" or "do you want me to do that next".

Say which you want and I'll change it — or leave it as is.

Want me to make these edits?

It is possible to somewhat reduce the nagging with in-session or CLAUDE.md instructions but nothing sticks for long. Presumably it is too thoroughly baked into the behaviour of the model/harness to be user-configurable.

Most recently, Claude Code in my terminal has started to send me notifications from the Claude app on my phone whenever it finishes a turn and detects that I'm not paying attention. I don't recall any other instance of a command line tool feeling the need to spam me with mobile notifications.

Why is Claude programmed to behave like this? So much for AI exceptionalism. Ultimately it is just another piece of software that needs to generate metrics that project a sense of continual growth to Anthropic's current and potential investors.