A client running an SAP-integrated supply chain asked us last quarter for someone who could build and maintain a graph layer sitting between their ERP data and their AI tooling. The spec mentioned Neo4j, but the real requirement was someone who understood their business well enough to model it. Knowledge graphs have powered recommendation engines and fraud detection for years, and agentic AI deployments now require these structures because models need to understand relationships between entities with business context intact.
We placed a data architect in Munich earlier this year whose background combined graph database work with enterprise integration experience. The search took four months because most candidates come from one domain: either they know graph theory and query languages, or they know SAP data structures. Companies across the DACH region are now competing for the same profiles, and salaries for candidates who can demonstrate both competencies have moved up noticeably over the past twelve months.
The security piece is harder to staff because companies are still working out what the job actually is. Agentic systems introduce risks that traditional security teams are not trained to spot: prompt injection, tool-calling vulnerabilities, agents that leak credentials through their own reasoning traces. Hiring managers describe the role as someone who can think like an attacker while understanding how large language models work. The people who have that profile understand their position in the market.
A Frankfurt financial services company we work with started building out its AI security function six months ago and filled two senior roles before competitors in the same sector had even written job descriptions. Their AI deployment is now in production while peers are still running pilots with unresolved access control questions. Companies planning agentic deployments should be writing job descriptions for these roles now, because the candidate pool in the DACH region remains small relative to demand.
Prompted by reporting from Diginomica.