By Chris Morrow.
About this piece
This was written after a live interview on CNN International in August 2025, with news anchor Ben Hunte, on how artificial intelligence is changing the outlook for graduates. It is left as a record of that conversation rather than quietly updated, so the dates and examples are as they were then.
1. We are in the dial-up phase
The useful comparison is not the smartphone, it is early consumer internet: obviously important, visibly unfinished, and impossible to opt out of. The difference is speed. What the internet reshaped over two decades, AI is reshaping in a small fraction of that.
For anyone at the start of a career, that cuts both ways. The roles being created are genuinely new rather than relabelled, and the people who move early get to define what they are. The cost of waiting to see how it settles is that it will have settled without you.
2. AI literacy is the baseline, not a specialism
The point that got the most reaction was that this is not a computer science question. A doctor, a lawyer or an architect without working AI fluency will be at a real disadvantage well inside a decade, and none of those people are going to pick it up from a module in somebody else’s department.
Embedding it across every discipline is slow, institutionally awkward and mostly not happening at the pace the labour market is moving. The result is a mismatch between what graduates can do and what employers are already asking for.
3. Education is arguing about the wrong thing
A great deal of institutional energy has gone into whether students are using AI to cheat. That is a real question and it is a small one. It is also the easier conversation to have, which is part of why it dominates.
The larger question is whether anyone is being taught to use these tools well: where they fail, how to check them, what they are actually good for. A graduate who can direct a model competently and knows when to distrust it is worth considerably more than one who has been kept away from them for three years.
4. AI is now the gatekeeper in hiring, and not always for the better
Most large graduate schemes now screen with automated tools before a person reads anything. That is a straightforward efficiency for the employer and a much worse experience for the candidate, particularly anyone whose background does not pattern-match to whatever the system was trained on.
It is worth being honest that our own industry drove a lot of this. Screening technology got adopted faster than the judgement needed to use it well, and plenty of good people are being filtered out by a process nobody has audited.
5. Adaptation is urgent, and it is not only the graduates’ job
The version of this conversation that puts the burden entirely on twenty-two-year-olds is the wrong one. Employers who have not worked out what an AI-literate junior is worth, and educators who have not changed a curriculum in five years, are as much of the problem as any individual’s choices.
What this means if you are building an AI team
Two things follow from the above that have held up in the year since.
The bar for a junior has moved, and most job specifications have not caught up. Fluency with AI tooling is now closer to a baseline expectation than a differentiator, in the same way that using version control stopped being worth mentioning. Specifications still written as though it were exceptional are selecting on the wrong thing.
Screening technology cannot assess judgement. Everything above about automated graduate screening applies with more force at the senior end, where the question is not whether someone has the keywords but whether they have shipped something that survived contact with production. That is the part no tool does for you, and it is the reason we run first-stage interviews ourselves rather than sending a shortlist of CVs.
If you are hiring senior AI or machine learning people, the practical version of all this is on our services page. If you want the market detail, the US and UK pages have it by geography.