50 personal projects. Nearly none of them shipped, maintained, or useful. That's the ledger one developer tallied when they sat down to assess what their AI tool subscriptions had actually produced - and it's the opening argument in a blog post that makes a case most heavy AI users don't want to sit with: the tools themselves might be the problem.
The author's framing is blunt. They call AI a "thermonuclear ADHD amplifier" - not because it doesn't work, but because it works too frictionlessly. You can spin up a speech recognition system, a video clone pipeline, or a working SaaS scaffold in an afternoon. So you do. Then you do another. The author describes watching a colleague manage "5 rooms where they manage their agents" and noting physical unease at the sight.
The Friction Argument
The core claim isn't that AI produces bad output. It's that removing effort removes commitment, and removing commitment removes focus. The author illustrates this with a speech-to-blog experiment: when writing became as easy as talking, the quality collapsed - not because the transcription was bad, but because they stopped caring enough to think hard before speaking. "Quality writing is not conversational English simply cast through a lens," they write. Handwriting, they argue, remains irreplaceable for serious work precisely because it's slow.
This maps onto research Cal Newport has been doing on knowledge work, which the author cites directly. Newport's findings suggest AI tools are increasing time in shallow tasks - quick, low-concentration work like drafting quick replies or reorganizing notes - while reducing deep work, meaning the extended, focused concentration on a single difficult problem that actually moves important projects forward. The author sees the same pattern in themselves and their peers.
The counterargument is obvious: friction isn't inherently good. Having to hand-code everything didn't produce better software, just less of it. Most people would rather have a working prototype to refine than an idea that never leaves a notebook. But the author isn't arguing for returning to pain - they're arguing that the specific friction of deciding something is worth your time, before the tool makes it trivially easy to start, is the part that matters. Take that decision away and you get 50 projects and no products.
What Actually Ships
The implicit yardstick here is simple: did it ship, and does someone use it? By that measure, a graveyard of 50 AI-assisted experiments is a tool that's excellent at starting things and poor at helping you finish them.
This is worth sitting with if you're a heavy AI user. The question isn't whether ChatGPT or Claude can help you move faster - they clearly can. It's whether "faster" is the right metric for the work you actually care about. A faster pipeline to a project you were never really committed to finishing isn't productivity. It's a more expensive way to procrastinate.
The author's proposed fix - cancelling the subscription outright - is deliberately provocative. The real point is harder to act on: use AI tools for work you've already decided to do, not as a first-pass filter for deciding what's worth doing. Once the tool becomes the reason you start something, you've handed over the judgment call that mattered most.