# What Jobs Will Exist When My Child Grows Up? The Skills to Build Now

**Author:** Dan Fitzpatrick
**Canonical URL:** https://parents.theaieducator.io/posts/what-skills-do-children-need-for-the-future
**Published:** 2026-10-04T07:25:30.000Z
**Updated:** 2026-10-04T07:25:37.918Z
**Category:** The Future

## TL;DR

Nobody can tell you which jobs will exist when your child grows up, and the OECD says so itself: its May 2026 AI exposure measure warns that real outcomes depend on adoption, regulation, organizational change and social choice, three of which nobody has decided yet. What the same work can name is the work AI is worst at: contextual judgment, interpersonal understanding, complex decision-making and responsibility. Build those four at home rather than picking a subject to dodge AI. Do this first: ask a chatbot something your child is an expert in and let them catch the mistake. Entry-level hiring has fallen, but the Economic Innovation Group's January 2026 analysis of 238 million US job postings concludes the cause is interest rates rather than AI. Your school's and college's own careers guidance takes precedence on anything concrete, and if your child is frightened rather than curious, that is the conversation to have first.

## Key takeaways

- Nobody can name the jobs, and the OECD's May 2026 exposure measure says outcomes depend on adoption, regulation, organizational change and social choice, which are decisions not yet made.
- The work least exposed to today's AI needs contextual judgment, interpersonal understanding, complex decision-making and responsibility. Those are capacities, and capacities can be built at any age.
- Build the Hard Four at home: reading a room, carrying a decision you cannot undo, taking responsibility when it goes wrong, and knowing when a confident answer is wrong.
- Entry-level hiring fell, but the Economic Innovation Group's January 2026 study of 238 million US job postings attributes it to interest rate rises from March 2022, not to AI.
- At 16 to 18, do not choose a subject to dodge AI. Choose the one your teenager can reach the top of, because depth is where judgment gets built.

## Article

Nobody can tell you which jobs will exist, and the honest sources say so out loud. What the evidence can name is the work machines are still worst at: judgment in a messy situation, reading other people, carrying a decision, owning the result when it goes wrong. Build those four at home. Any list of AI-proof careers depends on choices employers and governments have not made yet.

## Why nobody can name the jobs, and why that is not a dodge

The organizations whose job it is to forecast this have stopped naming jobs, and they have said why.

In May 2026 the OECD published its [AI exposure measure](https://www.oecd.org/en/publications/2026/05/the-oecd-ai-exposure-measure_489cfd42.html), which maps what AI systems can currently do against what occupations actually require, across nine areas of thinking, social skill and physical work. It does not produce a list of doomed careers. It produces a map of tasks, and it attaches a warning to the whole exercise: the measure gives "a transparent foundation for analysing task-level transformation, changing skill demand and future labour-market effects, whilst recognising that actual impacts will depend on adoption, regulation, organisational change and social choice."

Read that last clause again, because it is the part a parent needs. Adoption, regulation, organizational change and social choice. Three of those four are decisions other people have not made yet. A forecast of your eight-year-old's career has to guess all of them correctly for the next twenty years.

The OECD's own skills paper, [Skills in the AI age](https://www.oecd.org/en/publications/skills-in-the-ai-age_972bd15e-en.html), published on July 8, 2026, lands in the same place: "The effects of AI on employment differ across sectors, regions and cities, as well as across skill levels." Differ. Not fall.

So the question "what jobs will exist" has no answer, and I think chasing one is how parents end up making worse decisions than if they had never worried at all. I have watched families steer a teenager away from a subject she loved on the strength of a newspaper headline about radiographers. That is a real cost paid now for a prediction nobody could stand behind.

## What the evidence can name: the four things machines are worst at

Here is what the same OECD work does say, and it is more useful than a job list.

The occupations least exposed to today's AI are the ones that require contextual judgment, interpersonal understanding, complex decision-making and responsibility. The most exposed are the ones built on routine information processing, administrative work and tasks that can be written down as a procedure. Exposure also splits by the kind of machine: some work is more vulnerable to language and reasoning systems, other work to robotics and machine vision.

That is not a careers list. It is better than one, because it is a list of capacities, and capacities can be given to a child long before anyone knows what job they will be used for.

**The Hard Four are the four capacities AI is measurably worst at, and the four worth building at home: reading a room, carrying a decision you cannot undo, taking responsibility when it goes wrong, and knowing when a confident answer is wrong. None of them is a school subject. All of them are practiced in the ordinary jobs of a family, and every one can be handed to a child years before a career exists for it.**

That is my translation of the OECD's categories into things a parent can actually do on a Tuesday, not the OECD's own wording. The underlying finding is theirs. The kitchen-table version is mine.

> You cannot prepare a child for a job that has not been invented. You can prepare them to be the person in the room who notices the answer is wrong, and that person has never once been out of work.

## What I See in Schools

Schools are being asked to prepare children for this and are mostly being handed the wrong instrument.

I have spent the last few years on stages in more than thirty countries and in advisory rooms with the UK's Department for Education, KHDA in Dubai and Kazakhstan's Ministry of Education, and the pattern is consistent enough to be worth naming. Systems reach first for a new subject. An AI module, a digital skills strand, a unit bolted into computing. It is a reasonable instinct and it is the easiest thing to announce.

The trouble is that the Hard Four are not teachable as content. You cannot assess "carrying a decision you cannot undo" on a worksheet. They are built by giving a young person a real decision with a real consequence and then not rescuing them from it, which is a thing schools have become structurally worse at, not better, over the twenty years I have been near them. Coursework got more scaffolded. Deadlines got more flexible. Risk got managed out.

Meanwhile the thing children are actually learning from AI is thinner than the headlines suggest. Ofcom's [Children's Media Lives 2026](https://revealingreality.co.uk/wp-content/uploads/2026/05/childrens-media-lives-2026-summary-report.pdf), a qualitative study following 17 UK children aged 8 to 17 with fieldwork from November 2025 to March 2026, found that "all the children had seen AI generated content across different platforms and many used LLMs for schoolwork, personal advice, and play." It also found this: "Despite frequent use, understanding of how AI works remained limited. Many of the children did not distinguish between different AI tools." Seventeen children is a small, deliberately deep sample and not a national statistic, so read it as a close look rather than a count. But a close look at children who use these things daily and cannot tell them apart is worth more here than a percentage.

Fluent, and not discerning. That gap is the whole opportunity.

## The entry-level panic, and what the data actually shows

If one story has frightened parents this year, it is that AI is eating the bottom rung of the career ladder. The data does not currently support it, and the paper that looked hardest is worth knowing about.

In January 2026 the Economic Innovation Group published [Looking for the Ladder](https://eig.org/wp-content/uploads/2026/01/TAWP-Iscenko-Millet.pdf), by Zanna Iscenko and Fabien Curto Millet, which examined more than 238 million United States job postings from September 2019 to August 2025. Entry-level hiring did fall. The authors conclude that AI is not why: their reading is that "the data patterns observed are not early warnings of large-scale technological displacement, but rather the predictable consequences of a classic macroeconomic shock", specifically the interest rate rises that began in March 2022. The timing, they argue, lines up with the Federal Reserve and not with the chatbots.

Two things to hold at once, and the authors hold them too. The graduate job market got harder, which is a fact your older teenager may already be living. And the cause looks like the cost of money rather than the arrival of AI, which matters because those two futures are different. The paper is careful to add that "absence of evidence is not the same as evidence of absence." I would put it this way: the ladder is currently bent by something cyclical, and whether AI bends it permanently is not yet knowable. Advise your child on what is in front of them, not on a displacement that has not shown up in the numbers.

## What to build at each age

The Hard Four are the same at every age. What changes is how you hand them over. Whether your child is ready for the tools at all is a separate question, and it has [three signs you can check tonight](https://parents.theaieducator.io/posts/should-i-let-my-child-use-ai) rather than an age.

**Under 8.** Nothing to do with AI, and that is the answer. At this age the Hard Four are built by being allowed to lose a game, settle an argument with a sibling without a referee, and be told the truth when they ask something hard. If AI appears at all it is on your phone, next to you, for a few minutes. What that looks like in practice is in [what to allow before 8 and what to withhold](https://parents.theaieducator.io/posts/ai-for-young-children).

**8 to 12.** Start on the fourth one: knowing when a confident answer is wrong. This is the age for the game where you ask a chatbot something your child happens to be an expert in, their football team, a book they have read six times, and let them find the mistake. Children who have caught a machine out once never fully trust it again, which is the single most protective thing you can install. [Explaining how AI actually works](https://parents.theaieducator.io/posts/how-to-teach-kids-about-ai) belongs here too, and [the five things a parent actually needs to know](https://parents.theaieducator.io/posts/ai-guide-for-parents) is the quicker route if you are starting from nothing. On the platforms themselves, note that OpenAI's help center, read on October 4, 2026, states that [ChatGPT "is not meant for children under 13"](https://help.openai.com/en/articles/8313401-is-chatgpt-safe-for-all-ages) and that users aged 13 to 18 need parental consent, so this is a sit-beside-them activity at this age rather than an account of their own.

**13 to 15.** Hand over a decision with a consequence you will not undo. A budget for something they want. A choice about a club they then have to stick with. The point is not the decision, it is the practice of living downstream of one. This is also the age to be specific about the difference between using AI for the doing and using it for the thinking. Let AI help with the doing; keep the thinking with your child. The research on where that line sits is uncomfortable and clear: students who hand the work over [score lower](https://parents.theaieducator.io/posts/is-ai-making-kids-dumber), and they usually cannot tell.

**16 to 18.** Now the subject-choice question becomes real, and here is the advice I would give, flagged as advice rather than evidence. Do not pick a subject to dodge AI. Pick the subject they are good enough at to reach the top of, because every one of the Hard Four is built by going deep in something rather than hedging across everything. A student who has genuinely mastered history has practiced contextual judgment on contested evidence, which is exactly what the OECD's least-exposed column describes. A student who picked three subjects defensively has practiced nothing. If your teenager is already using these tools heavily, [what teenagers actually use AI for](https://parents.theaieducator.io/posts/what-do-teens-use-ai-for) is worth reading before you advise them.

## What to say to your child

Short, and not in the middle of a conversation about grades.

> You could say: "What did it get wrong this week? There'll have been something."

> You could say: "If you had to decide this one on your own, with no take-backs, what would you pick?"

> You could say: "I don't know what jobs will exist either. Nobody does. Let's work out what you want to be good at."

That last one does more work than it looks like it does. A teenager who thinks their parent has a secret answer and is withholding it stops asking. A teenager who learns their parent is in the same fog, and is still thinking, keeps talking.

## What to ask the school

Three questions, and the third is the one that changes the answer.

1. How are students here being taught to check what AI produces, and in which lessons does that actually happen?
2. Where in the week does a student make a decision with a real consequence that nobody rescues them from?
3. What were staff given by way of training before any of this was switched on?

Ask them at a parents' evening, plainly: the head of year in a UK secondary school, the grade-level lead or counselor in a US middle or high school. When a school cannot answer the third one it is almost never indifference. It is that nobody built the [AI strategy](https://theaieducator.io/ai-strategy-for-schools?utm_source=parents.theaieducator.io&utm_medium=referral&utm_campaign=what-skills-do-children-need-for-the-future) the tools were supposed to sit inside, which is the work I do with schools and districts and a fair thing to forward to a senior leader.

## What none of this can promise

The Hard Four are a reasonable bet, not insurance. The OECD's own caveat applies to everything above: what actually happens to any job depends on adoption, regulation, organizational change and social choice, and nothing on this page changes any of those. A child who has all four can still graduate into a bad year, as plenty did in 2008 and as some are doing now.

Your school's and your college's own careers guidance takes precedence over a blog post for anything concrete: entry requirements, course combinations, apprenticeship routes. If your child is anxious about this rather than curious about it, and some are genuinely frightened by what they read, that is worth treating as the real subject. Your family doctor or pediatrician and the school's counselor or pastoral lead are the right first calls, ahead of any careers conversation.

And if the worry underneath the question is whether your child will be all right, the four capacities above are the ones I would want my own money on. Not because they are AI-proof. Because they were the ones that mattered before any of this, and nothing in the last four years has made them matter less.

If you want this sort of thing explained without the panic, I write a [Sunday newsletter](https://theaieducator.io/?utm_source=parents.theaieducator.io&utm_medium=referral&utm_campaign=what-skills-do-children-need-for-the-future#newsletter) for teachers and parents. Parents are very welcome.

*Dan Fitzpatrick is a former secondary teacher, assistant headteacher and Director of Digital Strategy who now helps schools and families make sense of AI. [More about Dan](https://theaieducator.io/about?utm_source=parents.theaieducator.io&utm_medium=referral&utm_campaign=what-skills-do-children-need-for-the-future).*

## Frequently Asked Questions

### Which jobs are safest from AI?

No list is reliable, and the OECD declines to publish one. What its May 2026 AI exposure measure does say is that the least exposed work requires contextual judgment, interpersonal understanding, complex decision-making and responsibility, while the most exposed is routine information processing, administrative work and anything that can be written down as a procedure. Exposure also varies by machine: some roles are more vulnerable to language systems, others to robotics and machine vision. Treat that as a guide to what to build, not a list of titles to aim at.

### Should my child still learn to code?

Yes, if they enjoy it, and no, not as insurance. Coding is a good way to practice precise thinking and to find out whether a confident answer is wrong, which is one of the four capacities that matter. But choosing it purely as an AI-proof career is the same mistake as choosing any other subject defensively, and software work is among the areas where these tools have moved fastest.

### Is university still worth it for my teenager?

That depends on the course and the alternative, and it is a question for your school's or college's careers guidance rather than a blog. What the current data does suggest is that the harder graduate market of the last few years looks cyclical: the Economic Innovation Group's January 2026 analysis of more than 238 million US job postings from September 2019 to August 2025 attributes the fall in entry-level hiring to interest rate rises rather than to AI, while noting that absence of evidence is not evidence of absence.

### At what age should my child start using AI tools themselves?

Later than most parents assume, and readiness matters more than the number. OpenAI's help center, read on October 4, 2026, states that ChatGPT is not meant for children under 13 and that 13 to 18 year olds need parental consent. Before that, AI is something you open together rather than an account of their own. Age limits are self-declared and weakly enforced, so the platform's number is a floor and not a judgment about your child.

### Is the advice different in the UK?

The four capacities are the same, and the Ofcom evidence cited here is British. The practical difference is the decision points: GCSE option choices at around 14 and A-level or equivalent choices at 16 force the subject question earlier and more narrowly than the US system does. The same rule applies at both: pick depth over defensive breadth, and ask the school's head of year or careers lead rather than guessing from headlines.