AI Productivity Burnout
What happens when the tools are so good you forget to stop
This is probably the most personal post I’ve written.
Last Tuesday I was putting Noa to bed and I caught myself checking my phone waiting for an agent to finish a deployment. Not something urgent. A side project. A watch face for my Garmin. My daughter asked me something and I answered on autopilot, and when I realized what I was doing I felt a kind of shame that’s hard to describe because the thing making me absent wasn’t work stress or a deadline. It was excitement. I wanted to see if the thing I’d built actually worked.
I need to back up and explain how I got here.
What happened
Over the past few weeks, while doing my actual job, I also built: Spilo (with Javi), a product now used by thousands of users. Drophere.cc, a server application for agents to host websites. A Garmin watch face. A bunch of video games Luzia themed. A real-time companion app for live events (private prototype for now). And probably a few things I'm forgetting.
I did this because I wanted to learn. I wanted to see how far the tools could go, how much one person could actually ship when the constraints that used to exist suddenly didn’t. The answer turns out to be: very far. And also: too far, if you’re not careful.
Six months ago, none of this was possible for me. The coding agents weren’t reliable enough, the workflows didn’t exist, the whole thing was still in “nice demo but doesn’t really work” territory. Then something flipped, and in a matter of weeks I went from running occasional experiments to running five parallel projects with tens of agents handling development, testing, documentation, and maintenance. It felt so cool.
And then, I noticed I got tired. Not the normal tired of a long week. Tired to exhaustion. And the confusing part is that I enjoyed almost every minute of it.
Where the exhaustion comes from
The pipeline never stops. Once you have a working setup with coding agents, you can run multiple projects at the same time. Agents are constantly producing: this PR is ready, that test failed, this deployment needs a decision. You’re context switching between projects, re-reading what you wrote two hours ago, figuring out the next step for each thread. It’s like managing a team that never sleeps, never takes a break, and ships faster than you can review. The bottleneck moved from production to attention, and attention is the one thing you can’t multiply.
One thing that makes it both incredible and exhausting: you get user feedback and you can implement it within minutes. Someone says “this button doesn’t work right” and fifteen minutes later it’s fixed and deployed. That loop is addictive. It’s also relentless, because there’s always another piece of feedback, always another thing to improve, and the cost of acting on it is so low that not acting feels like a choice.
The guilt loop. Every moment you’re not running agents, there’s a pull: I could be building something right now. And when you finish a project, you tell yourself you’ll clear the list and then stop. But software needs are infinite. There’s always another idea, another thing that would only take a couple of hours, and a couple of hours is now enough to build something real. So you start the next one. The friction that used to protect you from overcommitting (this would take weeks, I’d need a team, it’s too expensive) is gone. It turns out that friction was doing more work than I realized.
This is the part that affects my family. It’s not that I’m working 18-hour days. It’s that even when I’m not working, part of my brain is running a background process on what I could be building. That’s a different kind of tired than just long hours.
The slot machine effect. You write a prompt, kick off an agent, and wait. While it runs, you’re anticipating the result. Will it work? Will the avatar generator I just built actually look cool? (I actually wrote that one while writing the post xD) Then the notification comes, you check it, and there’s a hit of excitement mixed with “almost, but these two things need fixing.” So you go back, tweak, run again. It’s a variable-reward dopamine loop, the same mechanism that makes actual slot machines addictive, except you built the machine yourself and the rewards are real products.
Why I’m hesitant about this post
I’ve written repeatedly that the biggest risk with AI is people not using it. I wrote “The Year to Push.” I still believe that. A post about AI exhaustion feels like handing ammunition to the people looking for reasons to stay on the sidelines, and there are a lot of those people.
The number of people operating at this intensity is tiny, probably 0.0001% of the population. Most people haven’t sent their first serious prompt. The biggest risk is still people not using AI at all.
But it probably will become more people’s problem over time. Six months ago I couldn’t work like this even if I wanted to. The tools weren’t there. Now they are, and they’ll only get better and easier to use. What took me weeks of experimenting to figure out will eventually be the default workflow, and the questions I’m asking myself now (how do you disconnect, how do you protect your attention, how do you keep this from eating into the parts of your life that matter) might be worth thinking about before the answers become urgent.
What I know and what I don’t
I know it’s really fun. That sounds contradictory after everything I just wrote, but it’s true. Seeing how much you can build, how fast things move from idea to working product, how constraints you took for granted just disappeared, it’s exhilarating. I needed to see how far I could push both the technology and myself, and the answer is: further than I expected, in both directions.
What I don’t know is how to manage it. I don’t have advice. I haven’t figured out the off switch. I haven’t learned to disconnect. I’m just continuing, hoping that either I get used to the intensity or I crack, and then I’ll know where the limit is. That’s not a recommendation. It’s a confession.
This exhaustion is a luxury problem. It comes from having too much capability, not too little. If you’re reading this and you haven’t started using AI seriously, please don’t take this post as your excuse. The overwhelming majority of people need to push harder, not pull back.
This is a note from slightly ahead on the trail, where the air is thinner than I expected. The view is incredible. I just need to figure out how to breathe.
Hit close. I shipped 16 products in 2 months with an AI agent running autonomously. Sounds great on paper. In practice I had 24 overdue review tasks piling up because I physically couldn't consume what the system produced.
The dopamine loop you describe is real. I kept saying yes to the next build because building was the easy part. Reviewing, editing, approving, that's where it broke down. Turns out production speed was never the bottleneck. I was.
I wrote about the whole thing (the numbers and what I actually changed to survive it): https://thoughts.jock.pl/p/ai-productivity-paradox-wellbeing-agent-age-2026
Damn. This hit.
Saw someone on X arguing that vibe coding is the fastest-growing game mechanic in history.
Hadn’t thought about it that way but the dopamine cycle you bring up shows why.