Post

I've already been complaining about #ClaudeAI a lot recently and now that my team is actively going to be downsized in favor of Claude I am going to tear it a new one. Here's all the bad shit that nobody who uses this regularly talks about, a thread. By a snake forced to use it at work.

Claude consistently makes things up instead of checking documentation or trying to find accurate data online. This is extremely pronounced especially on larger models. Opus 4.7 is by far the worst offender of this. It constantly creates code that doesn't run or gives me guesses as answers.

The "personality" of Claude is also giving: "Asshole doing their best to cosplay a service worker." The apology language it uses when you point out a mistake is so extreme, it feels completely performative. It also keeps going "this is the most honest thing" or something like that.