When generative artificial intelligence (AI) arrived in the workplace, the first thing to change was sheer output. A single click now turns out pages of code, reports, or proposals. Yet a veteran software founder recently offered a diagnosis that points somewhere else: the resource that has truly become scarce is not the output, but the human attention needed to read and judge it.
The reason is simple. The volume AI produces has exploded, while the amount a person can absorb in a day has not. So the bottleneck is no longer how much gets made, but how much of it a human can actually review.
What makes this worse is what we might call AI slop. AI is fluent at producing long-winded explanations and unnecessary tangents. Harmless as they look, the next AI reads those tangents, learns from them, and serves up something just as bloated. Slop breeds slop, and quality erodes by degrees.
The striking part is the claim that you cannot stop this drift through human review. When AI dumps thousands of lines at once, a person hits the approve button without reading to the end. It gets marked as reviewed, but it was never actually read. So the work gets cut into pieces small enough for a person to handle, and human attention is saved for the decisions that are hard to undo.
With that, the human role shifts. Producing in bulk goes to the machine, while people take on how the overall structure should be designed and what to refuse. As the cost of making something falls toward zero, the value of deciding what to make rises instead.
In practice, the answer is to split the territory. There is a safe playground where AI can experiment freely, and a core skeleton where one bad design choice will hobble you for a long time. Whatever comes out of the playground can be thrown away and rebuilt if you do not like it, because rebuilding now costs a fraction of what it once did.
One piece of conventional wisdom gets inverted, too. In the past, when you spotted a flawed design midway through a project, you tended to leave it and tell yourself you would fix it later, because rewriting was expensive. Now it is the reverse. Rewriting has become cheap, but if you leave a flawed design in place, the AI agents working on top of it copy the mistake, and the problem compounds fast. Putting it off has become the more expensive choice.
This shift reprices people as well. The forecast is that the ability to juggle several tasks at once without losing the thread matters more than deep expertise in any single field. Someone who handles AI well multiplies their output, while someone who cannot becomes a fast generator of problems. The same tool acts as an amplifier for one person and a hazard for another.
It is a story from the world of coding, but the same shape — AI flooding you with filler while you lack the time to filter it — turns up in nearly every office where reports, meeting notes, and email change hands. The more a tool produces for you, the more your own judgment about what to keep and what to discard decides the quality of the work.