The New AI Superpowers: Focus and Followthrough

Rick Manelius starts from the assumption that burnout comes from overwork, which would mean that anyone getting tasks done 2-100× faster with AI should now be underworked and fine. He wasn't. The piece is short, and it is a confession rather than an argument.

His backlog going in was over 100 article titles with outlines and a someday-maybe list of 50 projects, accumulated across a decade of founding startups and becoming a father. Once agents made the shelved work feasible, he started running side projects between meetings, queuing Claude in five-minute increments and letting it work while he took the next call. It worked, and working is what caused the problem: success multiplied his appetite for starting things rather than satisfying it. He ended up with roughly 40 proof-of-concept projects running at once, each of them a new open loop, a new responsibility, a new thing to tend. The line he lands on is the one worth keeping: "I used AI to reduce my required work, so I invented an endless amount of make-work."

The correction he proposes is Greg McKeown's Essentialism — less but better — applied to the choice between spreading horizontally across more projects and going vertically deeper on fewer. He borrows Garry Tan's eclipse comparison for why the depth is worth it: the gap between a 99% and a 100% eclipse sounds like a rounding error, but a partial eclipse reads as a cloud passing while a total one turns day into an eerie night. The products and books people love pay for that last 1%, which can eat 50-90% of the total time, which is why most work ships at "good enough". His claim is that once AI has compressed the other 99%, there's no longer an excuse to skip it.

He wrote the piece as an intervention on himself: he had another article he was pushing out the door, judged it a B/B- on a topic that deserved better, and stopped to give it more revisions instead. The intent going forward is fewer articles, deeper.

The closing sentence overshoots — "no one should be experiencing burnout in the age of AI" is the same optimism the opening paragraph just spent the essay disproving, and it doesn't survive the 40 open loops that came before it. The mechanism he describes is better documented elsewhere in the vault. disposable-code explains the supply side: when code is cheap enough to write, fail at, and discard, the ROI threshold for starting something drops and maintenance dread stops deterring marginal projects. agentic-coding-fatigue covers what the resulting workload feels like from inside, with decision fatigue and gacha-loop pacing rather than volume as the thing that wears you down. never-enough is the same appetite viewed as a cultural condition instead of a personal failure, and know-thine-enemy is Amy Ko finding that three months of agent use cost her focus and joy while still producing results.