A finance director we placed earlier this year mentioned something that stuck with us: she now spends more time verifying reports from her own team than she did reviewing external submissions two years ago. The outputs look professional, the formatting is clean, the language is confident. What she cannot tell anymore is whether the person who sent it actually understood the underlying problem.
AI generates work that reads well but lacks substance, and the traditional markers of effort and expertise are becoming unreliable. A well-structured memo used to signal that someone had thought carefully about the question. A polished presentation implied preparation and command of the material. Those proxies are breaking down. Colleagues across DACH companies describe the same frustration: they receive work that ticks every surface-level box but collapses under the first serious question.
Interview processes have long relied on written exercises, case studies, and preparatory work to separate candidates who can think from those who cannot. If a candidate can generate a plausible deliverable in minutes without understanding it, the exercise stops measuring what it was designed to measure. Several clients have redesigned their assessment stages to include live, unassisted components precisely because the take-home test no longer tells them anything useful.
Candidates who lean on AI to clear hurdles they do not understand may pass one stage but struggle to perform once in role. Meanwhile, professionals who can explain their reasoning and defend their choices under direct questioning are finding that quality counts for more in final-round conversations. A Munich-based data team lead told us last month that her most successful recent hire was the candidate who asked to walk through his analysis on a whiteboard rather than submit a document.
Companies that treat this as a quality-control nuisance are missing the larger risk. When credibility signals fail, decisions slow down, trust erodes, and organisations lose the ability to identify who actually knows what they are doing.
Prompted by reporting from Diginomica.