Top
Best
New

Posted by WXLCKNO 12 hours ago

Claude Code is being dumbed down?(symmetrybreak.ing)
806 points | 542 commentspage 5
runjake 10 hours ago|
> “Read 3 files.” Which files?

> “Searched for 1 pattern.”

Hit Ctrl-o like it mentions right there, and Claude Code will show you. Or RTFM and adjust Output Styles[1]. If you don't like these things, you can change them.

Like it or not, agentic coding is going mainstream and so they are going to tailor the default settings toward that wider mainstream audience.

1. https://code.claude.com/docs/en/output-styles

peacebeard 10 hours ago||
My biggest beef in recent versions is the automatic use of generic built in skills. I hate it when I ask a simple question and it says "OK! Time to use the RESEARCHING_CRAZY_PROBLEM skill! I'll kickstart the 20 step process!" when before it would just answer the question.

You can control this behavior, so it's not a dealbreaker. But it shows a sort of optimism that skills make everything better. My experience is that skills are only useful for specific workflows, not as a way to broadly or generally enhance the LLM.

ndespres 6 hours ago||
It makes sense that any product written after the advent of these AI code generators, including the AI code generators themselves, will get worse as it starts to eat itself.
ukuina 11 hours ago||
It's clear we're seeing the same code-vs-craft divergence play out as before, just at a different granularity.

Codex/Claude would like you to ignore both the code AND the process of creating the code.

ergonaught 11 hours ago||
If you've got a solution to the problem of bad decisions made by people who shouldn't be empowered to make them in the first place, you'll solve more than Claude Code.
jgb1984 8 hours ago||
I like claude models, but crush and opencode are miles ahead of claude code. It's a pity anthropic forces us to use inferior tooling (I'm on a "team" plan from work). I can use an API key instead but then I'll blow past 25$ in an hour.
supermatt 9 hours ago||
And they hate that people are using different agents (like opencode) with their subscription - to the extent that they have actively been trying to block it.

With stupidity like this what do they expect? It’s only a matter of time before people jump ship entirely.

singularfutur 9 hours ago||
Anthropic is optimizing for enterprise contracts, not hacker cred. This is what happens when you take VC money and need to sell to Fortune 500s. The "dumbing down" is just the product maturing beyond the early adopter phase.
bennydog224 4 hours ago||
It feels like CC got nerfed after the 4.6 drop.
madrox 10 hours ago|
I have noticed, if I hit my session quota before it resets, that Claude gets "sleepy" for a day or so afterward. It's demonstrably worse at tasks...especially complex ones. My cofounder and I have both noticed this.

Our theory is that Claude gets limited if you meet some threshold of power usage.

More comments...