UBS will require AI proficiency from graduate and intern candidates
The requirement applies to 2027 junior positions. It selects for the skill at exactly the level of the bank where the skill is most likely to be automated.
UBS will require graduate and intern candidates to demonstrate AI proficiency for junior positions starting in 2027, according to the Financial Times.
As a signal it is unambiguous: a global bank has decided that working with AI systems is a baseline competence for entry-level finance rather than a specialisation. As a policy it contains a contradiction the bank has not addressed.
The contradiction
The junior analyst role in investment banking is built out of exactly the tasks that current models do well. Building and formatting models in spreadsheets, assembling comparable company analyses, drafting pitch materials, extracting figures from filings, formatting decks.
If AI proficiency is now required at that level, it is because the bank expects those tasks to be done with AI. Which means fewer hours of that work per analyst, which means fewer analysts needed to do it.
UBS is therefore selecting for skill at operating the thing that reduces demand for the role being hired into. That is not irrational — someone has to run the systems, and the bank would rather it were people who arrive able to — but it sits oddly with an unchanged graduate intake.
The deeper issue is what the analyst years are for. They are an apprenticeship: judgement about what a number means, and what a client will accept, is acquired by producing thousands of them under supervision. If the production is automated, the apprenticeship has to be replaced with something, and no bank has said what.
What "demonstrate proficiency" can even mean
UBS has not published how the requirement will be assessed, and every available method has problems.
A certification tests familiarity with a product that will have changed by the time the candidate starts. A practical exercise tests prompting technique, which is the shallowest and least durable part of the skill. Self-reported experience tests confidence.
The capability that actually matters — knowing when a model's output is wrong in a way that matters, and having the judgement to check — is precisely what a candidate with no professional experience cannot have. It is acquired in the job that is being automated.
The industry is split on which end to train
Caterpillar is spending $100 million training 118,000 existing employees, on the explicit view that deployment depends on experienced operators redesigning workflows rather than on installing models. Its bet is that domain knowledge is the scarce input and AI skill can be taught.
UBS is filtering the intake. Its bet is the reverse.
Meta tried a third approach and abandoned it: the company discontinued AI-adoption dashboards and token-consumption scoring in performance reviews after legal challenges, having found — as reporting on its "AI-native" workforce push indicated — that a 220 percent increase in code changes and 36 percent more delivered features came with 40 percent more incidents.
Measuring adoption produced adoption. It did not produce outcomes.
What UBS has not said
The bank has not disclosed whether graduate intake numbers will change, how the requirement will be assessed, whether existing junior staff face the same standard, or what training it will provide to candidates who lack the skill but have everything else.
That last omission matters most for who gets hired. AI fluency correlates with access to tools, unpaid time to experiment, and university programmes that teach it — which are not evenly distributed. A hiring filter with that property has a composition effect, and the bank has not said whether it has looked for one.
Runs the newsroom. Rename this profile in the studio to your own byline.
Related
Every weekday, the AI stories that moved money or shipped code.
No cross-posting, unsubscribe anytime. See all newsletters