Every argument in favour of AI converges on a single benefit: saving time. That's true, and it's measurable. But two questions almost always go unanswered. Where does that time go once it's saved? And the effort you've removed — did it only produce the result, or did it produce something else too?
This week we published a guide to practical uses whose central argument was time saved. It's only honest to turn the question around.
First paradox: saved time doesn't become free time
This is an old observation, well documented by the history of technology. Household appliances were supposed to free up hours of domestic chores. Time spent on housework did indeed fall, but cleanliness standards rose, and the freed-up time was largely absorbed by other obligations.
The same mechanism plays out at work. Email was supposed to save time compared with postal mail. It did, request by request, and produced such a volume of exchanges that no one feels they've gained anything at all.
The reason is simple: when a task becomes cheaper, you do more of it. If writing a report takes ten minutes instead of an hour, the organisation doesn't give everyone fifty minutes back — it asks for six reports. The gain converts into volume, not available time.
That's exactly what the reorganisations we've documented illustrate, from Meta to Allianz: productivity gains don't translate into comfort for those who remain — they translate into revised expectations.
Second question: what the effort produced on top
This is the trickier part, and we need to be precise so as not to fall into nostalgic praise of difficulty.
Some efforts produce only their result. Copying out an address, calculating VAT, reformatting a document: the toil produces nothing other than the completed task. Removing it is a pure gain, and no one should mourn it.
Other efforts produce two things: the result, and a transformation in the person doing it. Writing a difficult text clarifies your thinking at the same time as it produces the text. Searching long and hard for a solution builds an understanding that will serve you elsewhere. Making mistakes and starting over builds judgement.
When you automate the second type of effort, you get the result without the transformation. That's precisely what the research we examined in our article on cognitive delegation suggests, with that notion of cognitive debt: an immediate benefit, an invisible deficit.
One question is enough to distinguish the two cases: do I still need to know how to do this? If the answer is no, delegate without a second thought. No one knows how to compute a square root by hand anymore, and the world is none the worse for it. If the answer is yes — because you'll need to judge a result, supervise someone, or decide in a case the machine doesn't cover — then the effort wasn't waste. It was the investment that lets you decide today.
The shift in fatigue
There's a third effect, less often mentioned. Working with AI doesn't remove fatigue; it changes its nature.
Producing something yourself is slow and tiring in a familiar way. Checking a production that arrives quickly, looks correct on the surface, with possible but invisible errors, is tiring in a different way. It demands continuous vigilance, without the natural rhythm that execution imposed. People in supervisory roles know this phenomenon well: monitoring is more exhausting than doing, because you can never let your guard down.
This isn't an argument against the tool. It's an argument for not confusing faster with less costly for the person.
What to take away
The time saving is real, and it would be absurd to forgo it. But it is neither automatically free time, nor free of charge.
What remains in our hands is the decision about what we delegate. Not all or nothing, but a sorting: automate without hesitation what teaches you nothing, protect what builds something, and stay clear-eyed about the fact that freed-up time will likely be reclaimed by something else if no one defends it.