Ask Londoners where they learned to use AI and the answer is stark. Three-quarters taught themselves by experimenting. Educational institutions accounted for 8%. Libraries and community groups, 4%.

Incandescent is not a word I use lightly (pardon the pun) but oh my lord, it's how I feel every time I see evidence that the institutions built to give people access to what they can't afford privately are carrying almost none of the load in the most consequential skills shift of our working lives.

Not because they should be running courses. But because somewhere along the way we decided that reading, writing and arithmetic were foundations worth building deliberately in every child - and that a healthy relationship with technology was something people would simply pick up on their own (and now increasingly banished to outside the school gates). One of those decisions was right. We are still, decades in, refusing to notice that the other one wasn't.

Because foundations aren't the same as training. Nobody teaches you to read so you can pass a reading course. They teach you so that for the rest of your life you can walk up to anything written and have a go at it without being invited. That's what's missing here - not curriculum, but the confidence to noodle away at something unfamiliar and assume you're allowed.

Last week I wrote about the on-ramp into work narrowing - fewer junior roles, weaker ladders, and a generation walking into a labour market where the first few rungs are quietly disappearing. This is the second-order effect, and in some ways it troubles me more, because it decides who benefits from all of this even when the jobs are there.

Back when IT was still in short trousers, nothing could happen without “training”. Made a change to the email system? Can’t be used until the people have been trained. New HR system? Worthless without at least a year of classroom sessions and a manual the size of a Toronto telephone directory. Thankfully, as the technology got easier to use, it also crept into our personal lives and we stopped waiting to be told. I used to have this argument with corporate audiences – “when was the last time someone from Google or Amazon popped round your house, with a white lab coat and a clipboard to tell you there was a new feature and we thought you might like some training before you use it?”

Sure it sounds ridiculous now when anyone who can speak can write code or translate a sentence into a different language, or into poetry - you get my point. We stopped waiting to be taught and the technology got so easy that learning just happens naturally and quickly, proportionate to the value it brings you.

Self-teaching is a wonderful thing but it can only happen when the conditions are right. It needs a device you don't have to share. Connectivity that works. A quiet hour. Enough slack in your week that experimenting doesn't cost you something else. Enough confidence to assume that prodding at an unfamiliar tool is a legitimate use of your time rather than skiving. And ideally someone at home who's already worked out what it's for.

Strip any one of those away and self-teaching stops being a democratic route to a better future and becomes a filter.

The digital divide has always moved on us. First it was devices. Then connectivity. Then skills. Each time we half-closed the gap, declared it solved and moved on to the next one. But every previous version had a shape we knew how to attack, because closing it needed capital - cables in the ground, hardware on desks, devices in hands.

This version of digital poverty behaves differently, and it's the difference that worries me because it widens the more the tool gets used. Every hour of practice compounds the advantage of whoever was already ahead, because fluency with these systems isn't a threshold you cross once and then own. It keeps deepening for the people who keep going, in a way that broadband speed never did.

The workplace isn't rescuing anyone either. Among the smallest Canadian firms that had adopted AI, 24% trained their existing staff. Among the largest, 68%. So the smaller your employer, the more likely you are to be entirely on your own - and small employers are where most people work, and crucially where much of the benefit could be felt. Not that a training course would fix it - but somebody in that building has to be the person who says have a go.

Education isn't catching it. Of the students and apprentices the London Taskforce surveyed, 40% said AI hadn't been covered in their training at all. Thirty-nine per cent didn't feel they had the skills they'd need for work. Which isn't an argument for adding an AI module to the timetable. It's evidence that nobody thought this needed building at all.

Which produces the thing I keep circling back to. The Taskforce's warning is that AI capability is becoming a requirement for work without any accessible route to acquire it, and without any recognised way to demonstrate you have it. Not a stated requirement - nobody writes it into the person specification. It just sits there in the gap between two candidates, invisible and decisive.

That's how advantage compounds in practice. Not because anyone is excluded on purpose. Because the requirement is never written down, so it can never be taught, funded, or argued with.

And it will not be fixed by handing out licences. Handing everyone a subscription solves the 2005 problem. The 2026 problem is capability, and capability needs somewhere to be built and something to prove it with.

The scale isn't small. The Taskforce counted around 600,000 Londoners in occupations combining high exposure to AI with the fewest resources to adapt - concentrated in lower-paid clerical and administrative work, disproportionately women. They're careful to say this shows where people have fewest options rather than predicting those jobs will vanish. And London is simply the city that bothered to count. Nobody has run this exercise for Manchester or Montréal or Melbourne, which means the number isn't absent everywhere else - it's just unknown.

But as we’ve seen, are training and courses really the answer? Three-quarters of those Londoners got useful with these tools by prodding at them, which tells you the barrier to entry is astonishingly low - lower than any technology I've worked with in thirty years. No waiting for the network to be built.

AI is the only technology I can think of where the capability could double overnight and every connected citizen would have access to it the moment they woke up and switched on. All this revolution needs is time, permission, and someone credible saying it's worth your while.

But "Every connected citizen" is an optimistic stretch at best. Plenty of people still aren't - not because the cable never reached them, but because they can't afford what runs through it. No connection at home. A phone with no data unless there's free Wi-Fi within walking distance. The engineering problem is largely solved. The affordability one isn't, and it never got the same attention.

Which means the fix has two halves, and we usually only argue about one of them. Yes, provide the access - the device, the connection, the hour. But access has never been sufficient on its own, and this is the first divide where the second half is nearly free.

Curiosity and the confidence to prod at something unfamiliar cost nothing to hand to someone. A teacher can do it in a sentence. So can a manager. So can a parent. What they can't do is give it to someone who's been taught to wait until they're told.

We may well all be in the gutter, but some of us are looking at the stars - and how I wish our education system spent as much time giving us the confidence to look up as it does giving us the cheat codes to get full marks on a six-mark question.

And this remains the cheapest fix we have ever been offered for a divide this consequential - which is precisely why failing to make it would be unforgivable.

It’s also why the most interesting of the Taskforce's recommendations isn't a technology one at all. The London Future Ready Guarantee is an attempt to give young people a way to demonstrate judgement, communication, creativity and adaptability alongside their qualifications. Which is, when you look at it properly, the careers conversation we should have been having all along - sitting alongside the subject choices and the apprenticeship or university options, rather than being left to whoever may happen to have a parent in the industry.

The uncomfortable truth is that this divide won't close by accident, and it's the first one in my career where waiting makes it worse rather than cheaper. Every month we wait, the people already fluent get more fluent - and the gap they're opening gets harder for anyone else to close.

That’s the part I keep having to remind myself of. Nobody is coming to the door in a white coat with a clipboard. Not for me, not for you, and not for the people this piece is actually about. Which means what decides who thrives here isn't the technology, and it isn't the training. It's who gets told this is for them. And any one of us can be the person who tells them.