Every company is hiring for AI skills. Few can say what that means in practice. Our latest workforce data reveals where genuine capability demand is heading − and it is not where most job adverts suggest.
A financial services client recently briefed us on a new role. The title was AI Engineer. The job description asked for experience with language models, prompt design and RAG pipelines. Standard enough. But buried in the requirements − almost as an afterthought − was this line: must be able to own reliability and incident response for production AI systems. That line tells you more about the current market than most trend reports. The headline skills get attention. The operational capability is what clients actually need − and what they are struggling to find.We are in a period where the vocabulary around AI engineering is expanding faster than the underlying skill base. Titles proliferate. Expectations are vague. And the gap between what gets written on a CV and what survives a technical interview is widening.
The differentiator is not who has used the most AI tools. It is who can keep AI systems running reliably when things go wrong at 2am.
What the data shows
Our latest analysis of live demand across technology hiring in the UK reflects the expected picture on AI-native skills. But context matters here − these are growth rates from a low base, and the absolute volumes are still concentrated in a relatively small number of organisations.
What the headline numbers do not show is the more important trend sitting underneath them. Alongside AI-native demand, we are seeing sustained and accelerating growth in roles that require classic software engineering depth: observability, testing, DevSecOps and system architecture. These are not separate trends. They are the same trend viewed from two angles.
As AI capability moves from experimentation into live production, the engineering stakes around it rise sharply. A prototype can afford to be unreliable. A customer-facing claims system, a fraud detection pipeline, or a loan decisioning tool cannot. The moment AI ships into production, every surrounding engineering practice becomes load-bearing.
The skills gap no one is advertising for
Ask most hiring managers what they want and they will mention LLMs, vector databases and Python. Ask them what actually causes pain in their teams and a different list emerges:
Testing non-deterministic outputs: most teams have no coherent approach to validating model behaviour across releases
Observability: instrumenting AI components so that silent failures surface before customers notice them
Cost governance: inference costs can scale in ways that surprise finance teams; few engineers are thinking about this at design time
Context management: knowing what goes into a prompt at scale, and why, with appropriate audit trails
Graceful degradation: building systems that fail safely when model performance drops, rather than silently producing worse outputs
These are not cutting-edge AI problems. They are mature software engineering problems applied to a new class of component. Developers who have built reliable distributed systems already have much of the instinct required. The gap is in recognising that the same disciplines apply − and in developing the AI-specific tooling knowledge to implement them.
Where the market is heading
Our view, based on current demand signals and conversations with technology leaders across Financial Services, Insurance, Telecoms and Utilities, is that the next 18 months will see the AI engineer title bifurcate. On one side: generalist AI tool users, for whom prompt skills are table stakes and commoditising quickly. On the other: production AI engineers who can design, integrate, test, monitor and own AI-enabled systems end to end.
The second group is where compensation pressure is building. It is also where supply is shortest.
In 18 months, ‘AI engineer’ will mean as little as ‘digital transformation lead’ did five years ago. What will matter is the boring question: does the system hold up?
For developers, the implication is less about learning new tools and more about deepening existing engineering practice − and learning to apply it to AI components specifically. For organisations building hiring strategies, it means moving away from broad AI skill requirements toward role definitions that specify what reliable AI delivery actually looks like in their context.
Experis works with technology organisations across Europe to identify, attract and develop engineering talent at this intersection of AI capability and production engineering depth. If you are building a team or rethinking your talent strategy for AI-enabled delivery, we can share what we are seeing in the market in more detail. Get in touch here.
Experis Talent Communities connect you with top talent, fast.
Built with qualified professionals, our communities are organised by specialist skillsets and backed by proven performance. Fully qualified, carefully selected and ready to deploy − we help you move quickly and stay ahead in a competitive tech market.
Talent Communities. Better Talent. Better Outcomes.
