AI Burnout: When Productivity Tools Become Make-Work Machines

Rick Manelius, a repeat startup founder and new dad, had a problem. He had over 100 article titles in draft, 50 projects on a “someday” list, and an AI assistant ready to help. So he started queueing up Claude in five-minute increments between calls, letting it rip, and watching side projects come to life. It worked. Too well.

In a post published July 26, Manelius describes the crash that followed. He found himself running 40 proof-of-concept projects simultaneously, each one a new open loop, a new responsibility, a new thing to tend to. The burnout pang returned. Not from overwork in the old sense, but from something stranger: AI had made him so efficient at starting things that he had invented an endless amount of make-work.

The conventional wisdom is that burnout comes from doing too much. The AI-era twist is that burnout can also come from starting too much. Manelius used AI to reduce the required work per task, so he filled the freed-up time with more tasks. The machine got faster. The human did not.

This is not a story about AI failing. It is a story about AI succeeding in exactly the way its vendors market it. Anthropic, OpenAI, and Google all sell productivity as throughput. Claude can write a draft in seconds. ChatGPT can outline a book in five minutes. Gemini can summarize a research paper. The implicit promise is that you will do more. The unstated corollary is that you will also start more, juggle more, and close fewer loops.

Manelius calls this the horizontal strategy: using AI to expand the number of things you do. The alternative, which he advocates, is vertical: going much deeper on a smaller set of things that matter. He borrows from Greg McKeown’s “Essentialism” — the idea of doing less, but better. In the AI era, Manelius argues, this is not a luxury. It is a survival skill.

The post includes a concrete example. Manelius had an article titled “Sesame Street Simple” that he was rushing out the door. He stopped himself because he realized he was giving it a B-minus effort. The topic deserved an A-plus. He decided to spend two or three more revisions on it before publishing. That is the vertical move. It requires saying no to the next shiny project.

There is a structural tension here that the AI industry has not acknowledged. The tools are designed to lower the barrier to creation. That is good for getting started. It is bad for finishing. Every new project started with AI is a new open loop, and open loops carry a cognitive cost. The brain does not distinguish between a productive open loop and a wasteful one. It just registers the unfinished thing.

Manelius invokes Garry Tan’s distinction between a partial and a total eclipse. The last one percent of effort, Tan argues, is not a rounding error. It is the difference between a dimming cloud and an eerie nighttime. That last one percent may feel like 50 to 90 percent of the total time. Most people stop at good enough. Manelius argues that AI, by making the first 99 percent faster, makes the last one percent both more achievable and more necessary.

The argument is not new. The framing is. AI productivity discourse has focused on speed, volume, and output. Manelius redirects attention to completion, depth, and follow-through. The hard part is that these are not features the tool vendors optimize for. There is no Claude mode called “finish the thing you already started.” There is no ChatGPT setting that says “do not let me start a new project until I close the last three.”

The post ends with a personal commitment: fewer articles, but deeper ones. It is a small bet against the volume-maximizing logic of the platform era. Whether that bet works for Manelius is his own question. Whether it generalizes to the broader AI-using workforce is the industry’s.

What Manelius describes is not a failure of the technology. It is a failure of the mental model most people bring to it. AI does not solve the problem of prioritization. It amplifies the consequences of bad prioritization. The tool that lets you start 40 projects is also the tool that lets you finish none of them.

The outstanding question is whether the next generation of AI tools will address this asymmetry. Agentic systems that can autonomously execute multi-step tasks may help close loops. But they also risk opening more loops faster. The bottleneck is not the speed of execution. It is the discipline of selection. No model update will fix that.

Manelius’s post is a useful corrective to the “AI will set you free” narrative. The tool can make you faster. It cannot make you focused. That part is still up to the human. And the human, it turns out, is the part that burns out.