Every recommendation I made last week rests on an assumption that I conveniently left out. That the people in your organisation arrived with judgement in the first place.

Deliberate practice without using a machine, spotting and learning from error signals, forming your own answer before you ask - all of it assumes there's something there to maintain. Judgement is a capability built earlier, somewhere else, by someone else. And for anyone under about twenty-five, that somewhere is a school system currently in the middle of the most difficult few years it has faced in my lifetime.

Because here is the problem, and it's genuinely hard rather than rhetorically hard.

Educational researchers have a name for it - the safety gap. The more helpful the tool, the more it risks eroding the very processes deep learning depends on. Which sounds like an argument for keeping AI out of classrooms, and isn't, because the thing being eroded is not knowledge. It's difficulty.

Productive struggle - hitting obstacles, making mistakes, sitting with discomfort while still working towards a solution - isn't an obstacle to learning. The difficulty, when it's the right difficulty, is the mechanism. You don't learn despite the struggle. You learn because of it.

Which bizarrely leaves education trying to protect the one thing every business is trying to eliminate. Friction.

When I set out to write this series of articles on skills and education, I had no idea that I would uncover the paradox that lies at the heart of everything I’ve been trying to get underneath.  Why, in my view, education struggles to prepare people with the right kind of digital skills and relationship that their future will require.  Everything I do professionally involves helping organisations remove friction - from processes, from customer journeys, from the working day. Friction is waste. Friction is what the elephant paths are cutting across. And then you walk into a classroom and discover that the friction is the product, and removing it is the one thing you must not do.

Both are true. They're just true about different things. Friction in a process serves nobody. Friction in learning is the only reason anything sticks.

So what does a school system do about it? Not what I expected, when I started looking.

The practical response emerging isn't about banning or detecting. It's about building the friction back in deliberately - asking students to argue against their own conclusion, to name the part of their work they're least confident in, to justify a decision rather than simply present it. Live discussion instead of submitted essays. Oral defence. Journals that track how you got somewhere rather than where you arrived.

The principle fits in one line, and it was drummed into me by Mrs Matthews, over and over, in fifth-form maths: "show your working".

Which is what my maths teachers were reaching for all those years ago, and mostly missing. They made me grind through the arithmetic when what they actually wanted was for me to understand the route - and it's the route, not the answer, that lets you look at somebody else's number and know it's wrong.

All of this means I owe an earlier argument of mine a correction.

A fortnight ago I complained about the six-mark question - about being taught how to collect six marks rather than to understand the concept. But look again at why it's worth six and not one. Someone decided that how you got there mattered more than where you arrived. The marks were always for the reasoning. We just stopped noticing, and started teaching the scheme instead of the thinking.

The instrument was right all along. We hollowed it out ourselves, over decades, because the scheme was easier to teach and easier to measure than the thinking it was supposed to reveal. And now the answer is precisely what these tools produce best - which means a system that marks answers has become a system for manufacturing surrender. The machine has read every mark scheme ever written and never has to show its working.

There's a version of this that ought to worry any employer, incidentally, and it's considerably wider than the graduate intake. If assessment rewards the answer, then the people arriving in your organisation have spent fourteen years being trained that the answer is what matters. So did everyone already sitting there. So did I. Nobody assessed any of us on judgement either. The difference is only that the rest of us got to build it ourselves afterwards - on small, cheap, low-stakes work where being wrong cost almost nothing and somebody found out quickly. Not because anyone designed it that way. Because there was enough unglamorous work to go round, and enough time to be bad at it.

Which is the thing that's actually changed. We are asking for a capability we chose not to build, at exactly the moment we're removing the rungs where people used to build it for themselves. And the real question behind all four weeks of this is what we put in their place - because if judgement no longer gets built by accident, somebody has to start building it on purpose.

And notice what all of this does to the model I've been complaining about for a decade. We built an education system to turn out people who could reliably produce correct outputs - for a world of Victorian production lines, where reliability was the whole job. The outputs are now free. The only thing left worth teaching is the judgement to know whether they're any good.

Epilogue:

I certainly didn't plan this, but across the last four weeks I've ended up asking the same question in four different places.

The first was about opportunity - can people get in at all, when the bottom rung of the ladder is quietly being removed? The second was about capability - once they're in, does anyone give them the confidence and curiosity to make these tools their own, or do we leave them waiting to be told? The third was about judgement - whether, having got in and got good, there's still somebody at home doing the thinking. And this one is about where judgement comes from in the first place - and whether anything left in the pipeline is still permitted to do the difficult part.

Which, four weeks in, turns out to be the first question again. The rungs I was worried about at the start weren't just where people got in. They were where people got good.

But notice what else they have in common. Not one of them is a technology problem. The technology is extraordinary. It is cheaper to reach than anything I've worked with in thirty years, it makes weaker performers better, and it hands back time that no previous generation ever got. Every single failure across all four is a human one. A hiring decision nobody examined. A foundation nobody thought worth laying. A supervisor who stopped turning up. An assessment system nobody thought to redesign.

Which is either the most depressing thing I could tell you or the most hopeful, and I've spent four weeks concluding it's the second. Because problems we created are problems we can choose differently about. There's no law of physics here. No inevitability. Just a very large number of small decisions, most of them being made by people who haven't yet noticed they're deciding anything at all.

I'm not giving up my agents. I have no intention of going back to doing long division by hand, or to filing my own tax return unaided. But there's a difference between handing over the what and handing over the why and the how - and the boundary moved while I was busy being delighted.

Noticing when it moves is the whole job. That's what the supervisor is for - and in this arrangement, the supervisor is you.

The rise of the humans was never a prediction about what the machines would do to us. It's a description of what's left when we stop waiting to be told, and start deciding for ourselves what all of this is actually for.

It starts with us, and it starts right now.