Choosing a career at eighteen: what this site can (and can't) tell you

It arrives in some form almost every week: my kid is picking a course. What should they do? Or, from the kid: is there any point doing law now?

It’s the most important question anyone asks a site like this, so here is the most honest answer we can give. Short version: we can’t tell you what to study. But we can tell you some things that are genuinely useful, and a few popular bits of advice that the data doesn’t back up.

First, what this site can’t do

Nothing here is a forecast. Every number on this site describes what jobs are made of today: which tasks AI software could do, which it could help with, which need a person. It doesn’t predict how many plumbers there’ll be in 2040, what they’ll earn, or who’ll be hiring. Anyone who tells you they know that is guessing, and the guess is usually wrong in the details.

So please don’t use the league table as a careers list, sorted top to bottom. That’s not what it’s for.

Five things the evidence does support

1. Pick a bundle you can keep re-cutting

Jobs aren’t single things. They’re bundles of tasks, and the bundle changes faster than the job title. A bank teller in 1975 and in 2005 did very different jobs with the same name (here’s that story).

So the best choice at eighteen is often not “the safest job” but “a field with lots of different tasks in it”, where you can drift towards the parts machines are bad at as things change. Nursing, engineering, teaching and the trades are all like this. So, as it happens, is law.

2. “Learn a trade” is half right

The most common conclusion people draw from a site like this is “go and be a plumber”. The data supports a narrower version: fine, fiddly handwork, in cramped and unpredictable places, on other people’s premises. That has been the hardest thing for machines for a very long time (here’s why).

That’s a strong case for plumbers (85) and electricians (81). It’s a weaker case for hands-on work in tidy, predictable places like a production line, which robots find much easier. Not all trades are the same trade. The “how fiddly and cramped” figure on each job page tells them apart.

3. Caring work is a social question, not just a technical one

Aged care and nursing score well, and not only because the work is physical. It’s because most people don’t want a machine doing the caring, even if one could. That’s a real protection, but it’s a choice societies make, not a law of physics. It could shift either way.

What we’d say: if caring work appeals, the data is on your side, and the demand from an ageing population isn’t going away.

4. The first rung is getting harder, and that’s worth planning for

Here’s the uncomfortable one. In office jobs, the tasks AI does best are often the junior ones, the work people used to learn on. Early data from the US suggests young workers in the most exposed jobs are already being hired less than their peers.

That doesn’t mean “don’t do law” or “don’t do accounting”. It means: look for courses and first jobs with real, structured training built in (apprenticeships, placements, graduate schemes), rather than assuming you’ll learn by doing the grunt work, because the grunt work may not be there.

5. The skill that transfers is knowing when the tool is wrong

Read down the new jobs page and one thing keeps coming up. The people getting hired into new AI-era roles are mostly experts from an existing field who can tell when the machine’s answer is wrong. A nurse who can check a medical AI. A lawyer who can check a contract tool. An electrician who understands a smart building.

So whatever you pick, go deep enough that you’d notice a confident mistake. That’s the closest thing to advice this data actually earns.

One thing you can do tonight

Pick two or three jobs you’re curious about and look them up. On each page, scroll to “Your week, your score” and slide the tasks around to match what you’d actually enjoy doing all day. It’s surprisingly revealing: two people in the same job can end up with very different scores depending on which parts of it they lean into.

And then go and talk to someone who does the job. They’ll know things no dataset does, including whether they’d do it again.