The frog doesn't notice.

Yes, real frogs jump out. The metaphor is rubbish biology. Stay with me.

The frog story is famous because it's useful, not because it's true. Drop a frog in boiling water and it leaps. Put it in cool water and turn the heat up slowly and — supposedly — it sits there until it's soup. Doesn't actually happen. Frogs are smarter than that.

Humans, on current evidence, may not be.

The study, in plain English

A new paper from a team at CMU, Oxford, MIT and UCLA (Liu, Christian, Dumbalska, Bakker and Dubey) ran three randomised experiments — over a thousand people — to find out what happens to your problem-solving when you've just been using an AI to help.

The headline finding is unkind. People who used AI on a set of fraction problems then did worse on the next set, which they had to do alone. Solve rates dropped by up to sixteen percentage points. Skip rates roughly doubled — that is, the people who'd just been using AI were measurably more likely to give up on the next problem before really trying. The same pattern showed up on reading comprehension. Different task, same effect.

You were faster while you had the help. You were worse, and lazier, the moment it was taken away. The authors call this reduced persistence. My grandmother had a different word for it.

The bit that actually matters

Here's the part the headline buries. Inside the AI group, the damage wasn't even — it was concentrated.

About sixty-one per cent of participants used the AI as a vending machine: they asked it for the answer. Those were the ones who took the hit on the follow-up tests. The rest — the people who asked for hints, clarifications, or a check on their working — looked statistically indistinguishable from the no-AI controls.

The tool isn't the problem. The reach is.

That sentence is the whole post. The rest is just trying to explain why people are reaching the wrong way without realising.

Why it's a boiling-frog problem

The reason this gets called boiling-frog and not people are getting dumber is that no single session does the damage. Every individual interaction with the AI feels fantastic. Quicker. Cleaner. Less friction. You got the answer, the email's drafted, the code compiles. You feel sharper, not duller.

The cost is invisible in the moment because it shows up on the next task — a different one, on a different day, when the AI isn't there. By definition you can't feel it. The only signal is a slow drift in how much friction you'll tolerate before bailing out, and almost nobody self-monitors that. We notice when our jeans get tight. We don't notice when our patience for a hard problem gets short.

You'll know you've been in the pot when you sit down to draft something cold and the cursor blinks for a beat too long, and the first instinct isn't to think — it's to open a tab. We're not getting stupider. We're getting impatient with our own brains.

Two ways to get this wrong

Both common. Both, as ever, the same mistake in opposite trousers.

Frog in denial. I'd never let it do my thinking for me, said by someone who has, for six months, in small ways, every day. The water's been getting warmer the whole time. The denial isn't a lie — it's that none of the individual swims felt like swimming.

Frog jumping out of the pot. The over-correction. Banning the tools. Treating any AI use as cheating. Requiring everything to be hand-typed at a wholesome typewriter. Loses every upside, fixes the wrong problem, and produces the kind of policy document that nobody actually follows. (Same shape as the fence so tight nothing useful happens failure from a couple of weeks ago. Familiar trap.)

Neither is the answer. The answer is paying attention to the temperature. Notice which way you're reaching for the tool. Notice whether your team can still do the thing without it, when they need to. Make the unassisted bit a periodic, ordinary part of the week, not a fire drill the first time it's needed in anger.

That's the manager's job, increasingly. Not to police the kitchen. To watch the temperature gauge — for the systems and for the people running them. That's the kind of work we do, if you want a hand with it.

The frog doesn't notice the heat. The frog notices the cold — when the AI's offline, and the form's still blank, and the thinking has to come from somewhere.