Developers are Turning Into AI Vampires. I’m One of Them.
Inside the addictive, high-adrenaline loop of running AI agents, and the cost of forgetting to touch grass.
I slept with my laptop last night. Open, on the bed, with a Claude session still running.
For the last week and a half I’ve carried it everywhere. It rides shotgun in my car, hooked to my phone’s hotspot, so I can keep working at a red light.
I’m not telling you this to brag. I’m telling you because I think a lot of you are doing the exact same thing right now, and nobody wants to say it out loud.
Marc Andreessen Calls Us AI Vampires
Marc Andreessen has a name for people like me. He calls us AI vampires.
His version goes like this: The best programmers in Silicon Valley now run five or more coding agents at the same time, reviewing one while the others write. They’ve stopped sleeping, because going to bed means all those agents stop working and you lose money every hour you’re out. The cost of sleep got too high.
Then he said the quiet part out loud: productivity is now outrunning comprehension. People are shipping way more code than they can actually understand.
And it isn’t just him. Andrej Karpathy, the guy who coined “vibe coding,” told one podcast he’s been in a “state of AI psychosis” since December and now spends 16 hours a day directing swarms of agents. Garry Tan, who runs Y Combinator, posted that he “stayed up 19 hours” and was “so addicted to Claude Code.” To his credit, he came back a few days later and added, “this is unhealthy, by the way.” Vercel’s CTO said the rush was as addictive as playing slot machines in Vegas. A veteran engineer named Steve Yegge wrote that he now runs a nightly escape plan: slam the laptop shut, bolt out of the room, and sprint away before he starts one more agent run at 2 a.m.
These are not interns. These are some of the most experienced engineers alive, and they’re describing themselves like they’re in recovery.
Six Months Ago It Got Me Too
I started using Claude Code about six months ago, over Thanksgiving. I was supposed to be on vacation. Instead I sat there prompting, getting a great result, then prompting again. I could not put it down.
It has only gotten worse since. Or better. Depends how you want to look at it.
Over the last week and a half I’ve put close to a hundred hours into building what I can only call a content machine. Here’s why I went so hard at it. Our website traffic fell about 80% in April after Google changed its algorithm. Our content was never that good, and we knew it. It was one of those things we kept ignoring.
I figured if anyone could fix it, I could. Real, trustworthy content is the one thing AI still can’t fake at scale, and I’ve made plenty of it over the years between a book, the newsletters, and hundreds of LinkedIn posts. So I fed all of it to Claude: who I am, what Full Scale does, everything I’ve published. Then I used that to train an engine that writes in my actual voice. In the last week alone I rebuilt our entire website and rewrote a big chunk of our content.
It has felt productive in a way that’s genuinely hard to describe. Magical, even. And that’s the trap. When it feels that good, stopping feels like a mistake. I get a real hit of adrenaline keeping it running and a weird stab of guilt when I let it sit idle. The laptop in my passenger seat isn’t dedication. It’s a problem.
The ADHD Part Nobody Talks About
Here’s the piece I haven’t seen written about much, and it’s the one I actually want to get into.
I have ADHD. When I lock onto something, I can’t let go. Hyperfocus is the polite word for it. My whole life it’s been a double-edged thing: incredible when it’s aimed at the right problem, and a disaster when it grabs the wrong one, because I’ll disappear into that single thing and let everything else I’m supposed to be doing fall apart.
AI is the most powerful hyperfocus fuel I have ever touched.
Think about why. I write one sentence and get back something that would have taken me a full day. Then I tweak it and get another one. The reward shows up instantly, every single time, and it’s a little different each run, so I never quite know what I’m going to get. Psychologists have a name for that pattern. It’s called variable reward, and it’s the same wiring that makes slot machines and social feeds so hard to put down. Researchers at Boston Consulting Group and UC Riverside even named the burnout version of it “brain fry,” the mental fatigue you get from running more AI than your head can keep up with.
For a builder with an ADHD brain, that loop is the most dangerous thing in the room.
My Friend Disappeared for Two Months
I have a good friend who started using Claude Code and basically vanished.
For two months he was a ghost, and everything he was actually supposed to be doing just stopped. Every time we talked he was lit up about what he was building. Old side projects he’d shelved for years were suddenly alive again, and things he always wished he’d finished were finally getting finished.
He was building and building and building.
Then he did the most engineer thing imaginable. He got so deep running five Claude terminals at once that he decided he needed to build his own elaborate system on top of it, just so he could hand even more work to the agents. He built a framework to feed the framework. If you know developers, you know this is where we always end up. We don’t just use the tool, we build a tool to manage the tool.
He really did turn into a vampire for a couple of months.
The whole time, I kept asking him one annoying question. How much money have any of these projects actually made you?
It’s six months later. The answer is still none.
Output Is Not the Same as Value
That question matters a lot more than the highlight reel, and the industry is starting to learn it the hard way.
Bloomberg ran a piece this year on what they called the great productivity panic. Managers started tracking how many times a day people pinged their coding agents, as if that number meant anything on its own. At Amazon, some people reportedly spun up agents to do pointless busywork just to keep their token stats high. Uber burned through its entire 2026 AI budget in the first four months of the year, a lot of it on Claude Code, and when its own COO got asked whether all that usage made the product better, he admitted the link “is not there yet.”
That’s the same story as my friend’s empty revenue line, just with a bigger logo on it.
The high feels like progress because something is always happening on the screen. But motion is not the same as outcome. You can generate a hundred thousand lines of code, or fifty blog posts, or fifteen revived side projects, and still not have built one thing a single person will pay for. Andreessen was right. The output is outrunning our ability to tell whether any of it was worth doing.
Touch Grass
Here’s the good news about my friend.
I got him out on the golf course last week. He left the terminals at home. For four hours he didn’t run a single agent, and somewhere around the back nine he started sounding like himself again instead of like a man doing math on the money he was losing by not prompting.
He finally touched grass.
I’m not telling you to quit. I’m an AI vampire writing this with my laptop within arm’s reach, and I have zero plans to put it down. The tool really is incredible, and the work I’ve gotten done in the last week is work I’ve wanted to do for years.
But the dopamine will lie to you. It will tell you that keeping the agents busy is the win. It isn’t. Building the right thing and getting it in front of a real customer is the win, and that takes the one resource a vampire refuses to spend: time away from the screen to think about what’s actually worth building.
This is the whole reason I wrote Product Driven and why we run Full Scale the way we do. The engineers our clients love most have never been the ones who crank out the most code.
They’re the ones who ask what we’re building and why before they write a line of it. AI makes that kind of thinking more valuable than ever. Anyone can generate code now. Knowing what’s actually worth building is the rare part.
So use the tool. Build the thing you’ve been putting off for years.
I’ll leave you with the same thing I tell my 11-year-old when he feels guilty about not having watched every single episode in the Star Wars universe.
There will always be more to watch, and more to build. Stop feeling guilty. There’s no finish line.
Close the laptop, go touch some grass, and get some sunlight. It’s summertime.

