AI is thinning the analyst bench direct lending cannot spare

Creditflux's 23 September commentary argues that artificial intelligence has begun to reshape the credit analyst's job around benchmarking, and that change is already slowing recruitment into direct lending, where hiring now concentrates on specialised roles and candidates who work fluently with the tools. The piece, bylined by Claude Risner, David Orbay-Graves and Patrick Costello and filed under direct lending, people, Europe and North America, sits behind the subscriber wall beyond that opening claim: no supporting detail, no headcount numbers, no named desks.

Take the thesis as given and the argument that follows is not really about the analyst but the analyst's replacement, because benchmarking is the most teachable work on a credit desk — stacking a new borrower against the last comparable deals, aligning covenant packages, ranking spreads, settling on what the collateral is worth — and the most teachable work is the work that gets automated first. It is also the work that has been doing the teaching, because the comparable set is where a junior learns which covenant bites before the breach, which sponsor renegotiates before the downturn, and which structure only resembles the last one until it does not.

This publication has argued that the private credit workout wave is already here, with recoveries negotiated amendment by amendment and the manager holding the pen setting the loss for every other lender in the stack, and that position makes a recruitment slowdown sharper rather than softer. A bench rich in fluency and thin on repetitions can run a covenant check faster than any previous generation of analysts, and speed is the wrong currency in an amendment negotiation, where the advantage belongs to the lender who has watched the same sponsor behave badly in a different deal. Funds trimming the analyst intake this cycle are borrowing against their 2032 workout teams, and the loan is unsecured.

There is a duller explanation for the same fact, and it may be the better one: desk budgets follow fee pools, and the pipeline we covered this month — a EUR60bn pipeline, two-thirds of it M&A — pads unitranche volume while compressing what that volume pays, and the compression is the part that lasts. When price falls faster than volume rises, hiring slows without any help from software, and the slowdown is attributed to whatever technology happens to be fashionable in the year it lands. The commentary's own detail, recruitment tilting toward specialised roles, fits either story; specialisation is what a market does when it can no longer afford generalists, and the constraint could be the spread rather than the model.

If specialisation means anything concrete, it likely points at the seats that manage credits rather than the seats that win them: surveillance, amendments, sector coverage, workout. Hiring into those seats is what a desk does when it expects to spend more of the year on loans it already owns, which turns a recruiting tilt into a forecast about credit conditions before it is a statement about software.

The scarce input in private credit is no longer capital but origination desks and warehouse structures that can securitize the collateral, and a hiring slowdown at the bottom of the desk is a bet that one originator will carry more assets than the last generation could, with the visible portion of the commentary carrying no arithmetic for that. If the multiplier is real, the industry buys cheaper origination capacity and the bill arrives in five years in a workout market that prices judgement; if it is not, the cuts are a margin decision with a technology label.

AI also has a second life in this market, on the demand side: our reporting in August on the $500 billion compute financing push covered memorandums of understanding with six firms that could open a channel for AI infrastructure lending, with the hard terms not public. Compute, data centres and the power behind them are becoming underwriting problems in their own right, and they do not look much like a sponsor-backed unitranche. A desk that hires fewer generalists and asks the survivors to be fluent in the tools is pointing specialisation in a sensible direction; the risk is that fluency on the model arrives well before judgement on the collateral, and the first stressed AI-adjacent credit is an expensive place to find out the sequence was wrong.

The testable version of the Creditflux claim is narrow enough to settle: watch how direct lending desks write their job specifications across the next two hiring cycles, because if fluency requirements spread while intakes shrink in a year when spreads are stable or wider, the technology is doing the work the commentary says it is doing. If intakes shrink into a compressing spread, the fee pool is the driver and the model is a passenger. Either way, the number worth asking for is the one the commentary does not print: analysts per dollar of assets before the tools arrived and after.