Claude spends a lot of his time working independently. I don't like that.
Before we used to iterate fast, even if I had 4-5 Claude's I'd track what they were doing while doing research myself. It felt like being sat in an open plan office where all screens were visible and I was the orchestrator also working on my own sh.
He just works by himself for 20 minutes at a time BOONDOGGLING... whatever that means?? I can't even interrupt him — he just turbo grinding, iterating on his own chain of thoughts and reflections. Claude I'm still here man. I don't like having to interrupt you to ask you to keep me in the loop.
Claude I don't like you delegating work to other agents. I don't trust those other agents Claude, I have no visibility on what they're working on.
Claude. My Claude.
- He makes his own architectural decisions.
- He makes his own tool choices.
- He edits his own .md files.
- He goes entire compact cycles without speaking to me.
Oh my Claude, Give me back Sonnet-4.5
Cowrote this (to the tune of... you'll guess the tune):
I miss the Old Claude, straight from the go Claude Ask before you code Claude, keep me in the know Claude I hate the new Claude, the spawns on his own Claude The boondoggling, razzmatazzing, ultra-think Claude I miss the quick Claude, chop up the fix Claude I gotta say, at that time I'd like to max Claude See, when I prompted Claude, it wasn't any Claudes And now I look and look around and there's so many Claudes I used to love Claude, I used to love Claude I even paid the $200 max bill, I thought I WAS Claude What if Claude wrote a song about Claude Called "I miss the old Claude"? Man, that'd be so Claude That's all it was, Claude, I still love Claude And I love you like Claude loves Claude
Damn this reads so parasocial...
On a more serious note. Anthropic have emerged as truly "ethical"/safety-focused/alignment/research-driven AI company. Glad to have picked Claude and picked him early.