New roles, or just new names?

Exactly the question Stefan Kiendl put to me here a few weeks ago: is forward deployed engineering genuinely new, or consulting with a fresh label? Banker gives both sides — Palantir has grown its commercial business to 46% of revenue on the model, while Anaplan’s CEO dismisses it as sales theatre: a polished POC, a nice dashboard, and then engineers you pay for in perpetuity, plus lock-in. My reading: both are true. The role is old — complex B2B always had solution engineering. What is new is the proportion. When a model carries 46% of revenue, it has stopped being a job title and become a business model.

What boards are actually doing, from 2,000 CEO interviews across 33 countries: three quarters of organisations have a Chief AI Officer in 2026 — a year ago it was around a quarter. 85% say every functional leader must become technology-literate. So titles are being created fast; the question is whether accountability sits behind them or only a signal. The most honest finding comes last: 83% say success depends more on people’s willingness to adopt than on the technology. That matches everything I see in mandates.

And here is the function that genuinely did not exist before. Keller’s line is the best of the week: "Generation is a commodity. Verification is a moat." When AI writes the code, the bottleneck moves from producing to checking — he would put 30% of engineering resources into generation workflows today and 70% into verification infrastructure. Wolfgang Strunck told me last week that his firm is piloting exactly that as a role: someone who defines and verifies what counts as an outcome. When the same function appears simultaneously as a product feature and as a job description, it is structural.

The academic finding behind the intuition: generative AI does not displace uniformly, it reorganises demand. The authors decompose employment change into shifts between occupations and shifts within them — and find both. For the roles debate that means most "new roles" are old roles with recomposed task bundles. That is not reassuring, quite the opposite. Rebuilding tasks is organisationally much harder than posting a vacancy.

The European reality check to close, drawing on the European Working Conditions Survey: there is a considerable gap between workers being exposed to generative AI and actually using it in daily work — and it varies sharply by country and sector. So before we argue about new org charts: across much of European industry the technology has not arrived in daily work yet. The roles debate is running ahead of reality. That is not an argument for waiting, but for settling usage first and structure second.