Everyone talks about AI saving time. In my experience so far, it hasn't. What it has done is take away a kind of tiredness I'd stopped noticing because it was just part of the job.
You'll know the kind I mean. Checking a subscription list against an invoice line by line. Matching timesheets to bills. Reading something for the third time to make sure there are no typos. Being obsessive about not missing anything, because missing one thing is the whole problem.
None of it is hard. All of it is exhausting.
Why checking work wears you out
There's decades of research on this. Psychologists call it the vigilance decrement: when you have to keep watching for something rare, your accuracy drops the longer you go.
It kicks in faster than most of us would guess. Under most conditions the drop is noticeable within about 15 minutes, and one recent study found attention starting to slide within 10.
The part that rang true for me is why. It used to be assumed this kind of work was boring but easy. Researchers at the University of Cincinnati tested that and found the opposite: monitoring tasks are tedious and they carry a high mental workload and real stress, including worry about whether you're doing it well enough. The simple, repetitive nature is exactly what makes it demanding, because you have to keep forcing your attention back onto something uninteresting.
So when you finish a morning of reconciling and feel wrung out, that's not you being soft. That's how brains work.
What changes when AI does the checking
When I hand a checking task to AI now, it works through the whole thing and comes back with the exceptions. This doesn't match. That rate looks wrong. This client is on last month's price.
My job shifts from hunting for problems to deciding what to do about them. That's a completely different kind of thinking. It's the part I'm actually good at and the part that needs my judgement.
It's like having an excellent assistant who has been through everything and brings you only what needs a decision.
And more than once it's found things I had missed. Not because I'm careless, but because of exactly what the research describes. By line 80 of anything, nobody is as sharp as they were at line 1. AI doesn't get tired at line 80.
The research on AI at work is starting to show the same shift. A 2025 study by Microsoft and Carnegie Mellon University surveyed 319 knowledge workers and found AI reduced the mental effort they felt across most thinking tasks. Their effort moved from gathering information to verifying it, and from doing the task to overseeing it. The researchers call that task stewardship, which is a good description of how it feels.
The honest bit: it's not quicker (yet)
I'll be straight with you. So far it isn't faster. Sometimes it's slower than I'd be doing it myself.
And I don't go off and do something else while it works. I sit and watch it. I'm not trusting it yet, and I think that's right at this stage.
But I can see where this goes. As I get to know where it's reliable, I'll start it off and get on with something else. Then it won't matter that it's slower, because I won't be the one sitting there.
Even now, the real win isn't time. It's that I finish the task with energy left for the work that needs a clear head.
The catch: don't hand over your judgement
There's a flip side, and the research is clear on it too.
Once people trust an automated system, they tend to stop checking it. It's called automation complacency, and it's been studied for years in aviation, medicine and control rooms. It's worst when you're busy with other things, which is exactly when you'd most want to rely on it. The same Microsoft study found that the more confident people were in the AI, the less critical thinking they applied.
So the goal isn't to stop thinking. It's to move your thinking to where it counts. For me that means:
- Watching it closely on a new task until I know where it's reliable
- Spot-checking results now and then, even once I trust it
- Keeping the decisions with me. AI flags the anomaly; I decide what it means for the client
What this means for your business
If you run a small business, you probably do a fair bit of this checking work yourself, or pay someone to. Reconciliations, invoice runs, timesheets, price changes.
The case for AI here isn't that it'll halve the hours. It's that it takes on the part that drains you, and hands you back the decisions. You still need to know the work well enough to judge what it finds. But you get to spend your concentration where it matters.