They decide which artificial-intelligence capabilities can go into a product, how to constrain them, and at what point their value justifies its uncertainty, its cost and its risk.
The AI Engineer sits at the boundary of software engineering, data and product. The job is not simply to connect an application to an off-the-shelf model. It is to design the system around that capability: selecting and preparing the context, orchestrating the calls, integrating with business processes, adding deterministic controls, handling failure, keeping a trace, and evaluating and supervising the whole thing in production.
The role is distinct from that of a Machine Learning Engineer. Where the latter works mainly on training data and the model lifecycle, the AI Engineer often works with models they have not trained themselves. Their challenge is to turn a probabilistic capability, sometimes supplied by a third party, into application behaviour predictable enough to be put in front of customers or other systems.
A convincing answer on a handful of examples proves neither the reliability of the system nor its ability to cope with real data. The assessment therefore looks at the ability to bound uncertainty, to organise verification, and to own the trade-offs between quality, latency, cost, system autonomy and human control.
Market benchmarks
The title AI Engineer still covers very different scopes on the French market. It can mean a developer specialised in generative systems, someone who came up through the backend, a product-oriented Machine Learning Engineer, or an engineer whose job is to integrate models supplied by others. That variability makes any hiring process built on a list of tools particularly fragile.
Demand is sustained, but profiles capable of getting past the prototype remain scarce. The gaps are less about calling a model than about evaluation, software integration, observability, data security and trade-offs with the product.
Context
In an IT services firm, the assessment focuses on the ability to operate inside a constrained client environment: heterogeneous data, legacy systems, variable security requirements, vendor choices already committed, and contractual promises sometimes made before any technical scoping. The AI Engineer has to spell out what is genuinely guaranteed and keep decisions reversible.
Level 1
Junior
A Junior can integrate a model into a bounded application flow, prepare the inputs, work with a structured output and apply the controls the team has defined. They contribute to an existing setup and can deal with the most visible failures. Faced with a plausible but incorrect result, they still need help to tell a model problem from a context problem or a usage problem.
Click to read
Level 2
Mid-level
A Mid-level engineer is autonomous on a common use case. They build an evaluation protocol, compare several architectural options, organise abstention mechanisms and plan fallbacks. Their usual limit is local: they improve the AI component without always measuring the effect on the business process, or the dependency being created on a vendor.
Click to read
Level 3
Senior
A Senior engineer reasons about the whole system. They distinguish what can stay probabilistic from what must be checked against a rule, a reference source or a human decision. They evaluate separately the quality of the context, of the generation, of the actions triggered and of the user experience.
Click to read
Level 4
Expert
An Expert is not the person who knows the most models or orchestration techniques. It is the person who settles a structural trade-off while owning what their decision gives up: reducing an agent's autonomy to keep control of its actions, keeping a human check despite an automation target, or refusing a deployment while reliability cannot be measured.
Click to read
An AI Engineer has to be able to define what counts as an acceptable output, build an evaluation baseline, measure regressions and decide whether a system can be exposed to real use.
The value lies in the assembly: context, model, tools, application, controls and interfaces. This weighting checks that the candidate can draw clear boundaries between probabilistic components and deterministic behaviour.
An AI system has to be observable, versioned, reproducible, reversible and operable despite its dependencies moving underneath it. Production is not the last step of the prototype: it changes the design itself.
Inputs may be untrusted, data may be sensitive, and outputs may be used well beyond their original intent. The assessment checks that data protection and the right to contest a result are built into the design.
A good AI Engineer knows how to reframe a request and organise real usage. A system that is technically convincing but ignored, worked around, or used without controls produces no lasting value.
The method in action
The scorecard at a glance — hover an axis
Data & artificial intelligence