I have heard it several times in recent meetings: someone argues that a chatbot forces us to redefine efficiency. The first time, I was about to nod along with the room. Then I thought about what that would actually mean, and the idea fell apart.

Efficiency has had one definition since the industrial revolution: more output per unit of time. That has not changed; the speed at which we produce has. That is the whole shift, and it is also where the trouble starts.

When every part of a process speeds up except one, that one part becomes the bottleneck. In this round of speeding up, the slow part is a person. A chatbot already produces faster than a person can; the judgment about whether what came out is any good still takes as long as it ever did.

A colleague showed me a report last week. He had built it with AI: ten pages, cleanly formatted, with charts and recommended actions. Twenty minutes for something that used to take a day. I asked if he had read it. “Of course,” he said. “Skimmed it.”

The skim followed from what the report had become. Something produced in twenty minutes reads like a draft. The person who built it passes it along as a finished piece anyway.

Efficiency rises, judgment does not keep pace, and the gap widens with every round of speeding up. We produce more and check less of it.

The counterargument is real: AI genuinely does in minutes what used to take teams weeks. That is an exaggeration for some tasks, though it feels true regardless for others. Still, those weeks gave time to think, argue, discard a draft and start over. That time was where judgment formed.

Judgment needs time because a good assessment has to mature. Sleeping on a draft changes how you see it the next morning. A conversation with someone else surfaces a gap you missed. None of that compresses. It is the one part of the process that stays human, even as the rest moves to automation.

I know companies that have quintupled their content output since introducing AI: newsletters, blog posts, social media, reports, five times what they produced before, if that is even enough by now. The people checking the work have not quintupled. One person now skims five times as much as before. The predictable result: mistakes, interchangeable copy, claims nobody checked or even read. Inside these companies the consensus stays the same: the output is fine.

Internally, these companies call it a productivity gain, and the numbers back that up: more produced, less time spent, costs down on paper. None of those numbers say whether what gets produced is worth anything.

I expect to wait a long time before anyone describes slowness as a skill. That AI can write a report in twenty minutes is settled. The question left open is who sits down afterwards to check that it is right, and whether what it recommends should actually happen. What risk sits inside a decision built on a report nobody read past the skim? That question stays open for as long as production and judgment move at different speeds.