The condition and the side that carries it
UBS requires candidates for 2027 graduate and intern places in Global Banking and Markets to demonstrate how they use AI to improve outcomes and efficiency, and it puts that question into the interview. The bank says the requirement extends to other newly posted roles, and that AI skills complement academic and social ability without replacing it. Santander is looking for advanced AI use in applicants to parts of its trainee programme.[1]
The condition lands on people who have not yet been hired. An applicant must arrive with tools, practice and a demonstrable result, and the bank has published no measurement of what those tools return inside the firm. Two further numbers sit around that entry point, and UBS set neither of them: Morgan Stanley analysts project more than 200,000 European banking job losses within five years, and JPMorgan's Europe head warned in December that junior staff cannot let go of fundamental skills. UBS is also testing analyst avatars for client presentations, which is the same junior work seen from the employer's side.[1]
What the buying side can show
On the buying side the measurement runs behind the spending. Uber deployed Claude Code in December 2025 and had used its entire 2026 AI coding budget by April; it now caps spending at 1,500 dollars per employee each month. Microsoft cancelled Claude Code licences across its Experiences and Devices division. Duolingo reversed a plan to factor AI use into performance reviews. Gartner expects spending on AI agent software to reach 207 billion dollars in 2026, against 86.4 billion dollars in 2025.[2]
Where somebody counted it, the gain becomes visible. Everlaw said 3,500 dollars of tokens cut a Java infrastructure implementation from 9.5 engineer-months to 2.5, and between 27,000 and 40,000 dollars of tokens cut an unlaunched product from 90 to 100 engineer-months down to 19. Those are one company's figures for two projects it chose to report. Databricks product director David Nasi said the routing decision stays transparent at runtime and the company avoids making the router opaque. That statement describes how the tool works; it gives no figure for whose work is redistributed.[2]
The measurement that is missing
The two developments meet at one point: the firms that cannot show what AI spending returns are setting the same proficiency as a hiring gate. UBS asks a candidate for evidence of improved outcomes and efficiency, while Uber's per-employee monthly cap, Microsoft's cancelled licences and Duolingo's reversal are what the buying side does when that evidence does not arrive. Asking a candidate for the requirement costs the employer nothing, and producing the proof costs something; the asymmetry at the door comes from there.[1], [2]
One number settles the distribution question. If UBS applies the announced interview questions to the 2027 Global Banking and Markets graduate and intern round, the share of offers going to applicants who demonstrate AI use rises above the preceding round's share, and the bank's own reporting can show it by the end of 2027. Until then the entry condition asks the applicant to carry a productivity claim the employer has not yet published.[1]