AI Engineering Is Now in the Fine Print

Gergely Orosz describes the 2026 software labor market as contradictory: classic software hiring is recovering in the United States and the United Kingdom, while demand for AI engineering is rising much more strongly. In Germany, the same shift looks less spectacular.

It does not arrive as a role explosion. It arrives as fine print in old roles.

Bitkom reports around 109,000 missing IT specialists in Germany for 2025. That is below the 2023 peak of 149,000, but it is still a structural gap. 85 percent of surveyed companies currently see a shortage of IT specialists, and 79 percent expect it to worsen further.

At the same time, the market is not simply hot. The Federal Employment Agency reports 1.12 million employees subject to social insurance contributions in ICT occupations for 2024, up 4 percent year over year. But the reported stock of open positions is 16,000 and has fallen because of the weak economy. The Institute of the German Economy even sees a 26.2 percent decline in open IT positions in 2024, down to 46,431. For IT experts, the decline was even stronger at 33.7 percent.

That sounds like relief. It is only superficial relief.

Because even with fewer job postings, bottlenecks remain. According to IW, more than 13,500 open IT positions could mathematically not be filled in 2024. Among computer science experts, 6,920 positions remained open. That was roughly 69.9 percent of open positions in this segment.

So the German market is not saying: we suddenly no longer need anyone.

It is saying: we are hiring more cautiously, but we still cannot find the people we truly need.

AI intensifies this tension. Bitkom reports that 42 percent of companies expect AI to create additional demand for IT specialists. 27 percent expect job reductions through AI, 16 percent expect the disappearance of roles that cannot be filled anyway. Only 8 percent are already using AI deliberately as a tool against the IT skills shortage.

That is Germany in one sentence: everyone talks about relief, but almost no one has built the operating system for it yet.

The visible job portals show the same finding in miniature. Stepstone currently lists 451 results for “AI Engineer”, 122 for “Artificial Intelligence Engineer”. Heise Jobs finds 65 results for “KI Entwickler”. That is relevant, but it is not a role explosion. Above all, the ads rarely seek pure research roles. They look for people who can bring together RAG, agentic AI, LLMOps, Python, cloud, data pipelines, business translation, and production integration.

This is not a new profession in a white coat. It is software development with additional load.

In Germany, the AI Engineer is therefore often still called Software Developer, Data Engineer, System Engineer, Consultant, or Business Informatics Specialist. Only now the fine print says: please also integrate models, build evals, contain hallucinations, think through data protection, understand business processes, and stop departments from turning every chatbot into a product.

The real change is not the title. It is the expectation.

Until now, a company could often read software competence as delivery capability: can someone build, test, deploy, operate? With AI, a second layer is added: can someone judge when a model is good enough, which errors matter, which tasks can be delegated at all, and where an apparently good automation breaks the process?

This is not a pure coding problem. It is a mix of Spec, Evaluation, and Terrain.

Spec: What should the system actually do?

Evaluation: How do we know whether it is good enough?

Terrain: In what operational environment does it land, with which data, rights, dependencies, and political side effects?

That is why the U.S. story cannot be transferred one to one. In the United States, AI roles emerge in a market with more capital, more product companies, more platform proximity, and more VC pressure. Germany has more Mittelstand, more regulated industries, more co-determination, more legacy, and often less internal product discipline. Here, AI engineering will appear less often as a heroic new role. It will run into existing roles.

That makes it less visible, but not less important.

The management mistake would be to treat this as a training detail: “Our developers should look into AI.” That will not be enough. If AI engineering becomes a baseline competence, role profiles, career paths, procurement processes, and operating models have to follow.

Otherwise the usual thing happens: new expectations are hung onto old job descriptions, it is called transformation, and later everyone wonders why people are overloaded.

The better question is not: how many AI Engineers do we need?

The better question is: which of our existing roles are silently receiving AI responsibility, without time, mandate, and evaluation standards for it?

That is probably where the German labor market for AI will emerge first. Not as a new job family with a clean label. As a quiet side contract inside old roles.

And quiet side contracts are usually the most expensive ones in operations.