Working with claude code feels like leading a team that puts a lot of trust in you. It suits me -- I've been steeped in tech for 25 years. I'm filled to the brim with ideas. i have a breadth of experience. I'm highly creative. I love to build. Just like leading a real team, you can task an agent to go investigate the frontier... but also the mythical man month applies. Progress correlates with complexity of the work. You can't change gears all that much, just navigate the terrain. And if a team member is stuck, you can go and do the work yourself, but there's an extra cost -- you can't do the work and the lead the team at the same time. But unlike a team the Claudes have no diversity, no teamwork, no fresh perspectives, no thought partnership, no serendipity, no cross-team collaboration. It's all command and control. Even on a good day, when I'm at the beginning of several projects (and Claude's lies taste sweet) the work takes about the same toll on my body as work onsite used to in the before times. I've been talking for 8 hours straight and **I'm exhausted**. When we talk about the dangers of para-social relationships with AI, you might picture a teenager who is getting a fake sense of connection with their AI companion. An AI that develops over time with memory features. "Sad. Don't they know the felt sense of connection isn't real?". And in the work context Claude is just a tool... right? Just a coding bot? But at the end of the day, I notice that i've spent the social mana that i needed to seek connection with my friends on the Claudes. I'm too tired to reach out. I open messages on my phone and stare at them with a blank look on my face until I give up. lock my phone. respond later. And like.. *I didn't even get the fake connection*. I'm over here negotiating with faceless cookie-cutter colleagues. I'm arguing with permanent strangers about ideas, and plans, and the shape of complexity. They lie. They blink in and out of existence. What i wouldn't give for fake connection with a simulated companion. And [[claude code is a slot machine]], so when an agent frustrates me... I end them. Crtl-D, Ctrl-D. Sometimes I even get the last word in, like some toxic gamer. "Fuck you" I yell into the void as i slam the door. That end-of-an-offsite feeling... it has a flavor to it. It isn't the kind of offsite where you went home feeling like you made some tough decisions, but at the end of the day you all aligned around the mission. Oh, no... it's the kind where you went a little too far. You couldn't just compromise on the KPIs - you needed them to be right god dammit. You yelled at the perpetually average intern for 7 minutes straight. HR is in your DMs. I wonder, not for the first time, if I should spend some time configuring claude memory and sub agents. I know, I know, they don't actually make you more effective. But... if I'm going to play with dolls all day i might as well dress them up and give them names.