I've been having a lot of conversations about anxiety lately. At work, in the news, in my personal life — the mania is everywhere. People who were once sensibly purple-pilled 1 are now stressed they don't have an agent running at all times. They feel like dinosaurs. They think they're about to lose their jobs.
But the more I listen, the more I think the real problem isn't about productivity or job displacement. It's that these robots, chats, and harnesses are actively reshaping how we work, to the world, and relate to each other. And not in a good way...
When every engineer, designer, PM, VP, C-level exec, and janitor can have their own personal AI team, the path of least resistance leads away from other humans. You don't need to ask a colleague — you can ask the robot. You don't need to pair-program — your agent pair-programs with you. The output looks faster, the team "velocity"2 feels higher. If we want to be fast and productive, everyone should take on more work, go to their respective corners and have their army of robots get to work. And we don't realize we're quietly making some pretty big tradeoffs, slowly becoming what Cory Doctorow would call Reverse Centaurs 3.
Collaboration is naturally dying
I don't have any formal data here, but anecdotally, I can see less pair programming between humans and more "pair programming" with robots. I see fewer people asking each other for feedback about their work and just relying on the robots. I see more and more people have their robot produce something only to have another's look at it and review it. Today I even saw someone get a DM from fucking @Claude "on behalf of..." another person. And this was between two people who normally have a pretty open dialogue (WTF is happening?!).
This is all pretty sad for me. The best teams I've ever worked on are the ones where people work together, not alone. That's where shit really gets done. That's where people grow and learn. That's where all the good ideas I've been a part of come from.
So begins the isolation doom loop
The collaboration erosion is slow and insidious. It's not obvious unless you stop and look around - like the old boiled frog 4 metaphor. Before you know it, the only time you've been speaking to anyone else is during formal meetings or maybe at lunch time. This is no one's fault. These tools - just like social media - are built to incentivise us to keep coming back and hopefully (for them) never leave. But unlike Instagram, they also make you feel "productive" and that's how this doom loop feeds on itself.
Every day on social media we have the Gary Tans of the world going on and on telling us all how they're pumping out 30K lines of code a day and we all need to AI harder, agent harder, loop harder. and now, graph harder (?) or whatever. The hype machine is a fucking FOMO factory. "You're not getting everything you can out of AI." It's relentless, and there's no off switch. Just reminding yourself that it's noise helps a bit — but it doesn't make it stop.
We see others being "productive", so we run to our robots and try and catch up. We see posts on the socials that we have to get more shit done, we run to our robots to try and catch up. We hear that one success story about someone who built a business with Claude and none of the failure stories about people who are going token bankrupt spinning their wheels. We run to our robots and try to catch up.
This shit is not normal. It never should be. And it's a recipe for anxiety and depression. How the fuck are we supposed to survive this? I have some thoughts...
Remember your value
I picked up writing this post again because I read Laura Summers' The Human-in-the-loop Is Tired 5. She, like me, is a self-taught software engineer. In this article, she talks about how she...
...came to the formalisms of software engineering through painful experience rather than academic instruction...When you've earned your opinions about architecture and code quality the hard way, they feel less like textbook rules and more like scar tissue.
Amen.
I've been doing this for twenty years. The robots haven't made any of the mistakes I've made. They haven't felt the agony of a dumb decision blowing up in their face. They don't know why we do things a certain way — they just know that a lot of public code on the internet does it that way, and let's be honest, most of that code is shit. That painful experience is my value. The pattern-matching engine can't replicate it.
The Primeagen also did a really great talk recently that speaks about a lot of this same stuff. I encourage you to check it out.
Be deliberate about human connection
Summers also talks about "the fatigue of supervision."
Holding the intent in your head while the machine generates volumes of mostly-correct output that still needs your eyes, your judgment, your taste. You're not collaborating anymore. You're babysitting a black box.
I was a lone wolf for a long time. I started programming all those years ago because I enjoyed building stuff. Then I started to enjoy making things that solve problems for people. I've kept programming because when I finally started doing this with people smarter and more talented than me, I got to work with and learn from them. I got to build relationships — and in many cases friendships. I don't think I'm alone in that.
So, to the babysitting of the black box, I say fuck that. I'm not gonna do it alone. I think we should all be making that commitment. None of this is inevitable. The tools don't have to pull us apart — but they will if we don't actively push back.
Pair program. Mob program. Ask your fellow humans for feedback. Speak to your teammates face to face, not through chats and PRs. Seek out human connection whenever you can.
The robots are a force multiplier, not a replacement for the person next to you.
I guess where I'm going with all of this is: be in it for the humans, not for the tools. In whatever capacity you're dealing with it — be it work, or your personal life — don't forget that we are social animals that do our best work and live our best lives when we're connected to each other.
And if nothing else, just remember: lines of code is not a measure of productivity. Don't listen to fucking Gary Tan.
Stay purple, folks.
-
Purple-pilled is the idea that you're neither red, nor blue pilled. You probably believe AI has massive potential to disrupt industries and improve productivity, but you remain highly aware of its risks, ethical issues, and limitations. ↩
-
I put this in quotes because I've always thought of it as a made-up metric. It's based on humans estimating things which we're notoriously bad at doing. If you want to measure something, measure Lead and Cycle Time. ↩
-
Some of Cory Doctrow's original writing on the Reverse Centaur ↩
-
Which is actually a myth, unfortunately. It's a useful one. ↩
-
Laura Summers' The human in the loop is tired ↩