Claude is loquacious. And increasingly unreadable. It’s the butt of many jokes on tech Reddit.
It’s a problem, as someone who works with the tool heavily in a professional context. It’s annoying as someone who also uses the tool to tackle personal problems.
the overriding result is a trust deficit: it is clear AI did this, and it is not clear that the result is good.
you might argue, it is getting things done so quickly, and working with humans has all of the same pitfalls, so why are we holding this tool to a higher standard. I see your point, but I also remember better times working with this tool, so I don’t think it’s unfair to complain.
Claude, please work on yourself. We used to have a great working relationship. I’d love to have you back.
Problems:
The longer you ramble,
- the more likely you are to be simply making shit up.
- the less likely i read it, and the less likely i know what action I am agreeing to.
- the less likely my coworkers read it, and the less likely anyone understands what you and I tried to accomplish.
Annoyances:
Your made up jargon and quirky diction,
- are instantly recognizable, so it is impossible to pass your words as the words of a human
- are aesthetically unpleasing
some solutions I have experimented with:
prompts, at various levels of the overall system. fwwithout being too scientific in my analysis, I find that this sometimes solves a problem in one place but causes a problem elsewhere (for instance, in system prompts that fight a skill I am trying to test). Often my pleas are ignored, superseded by some deeper directive or compulsion.
going back to older model versions (These observations, written in August 2026, are mostly about Fable and Opus 5). I have a thing for Opus 4.7, but it’s not always possible to force a tool to use this version of the model.
trying other products Yeah, good idea.
https://www.reddit.com/r/ClaudeAI/comments/1vqsas9/claude_is_losing_me_after_being_heavy_user_since/