'/%3E%3C/svg%3E){kind=small}
The European analysis of Artificial Intelligence skills needs says it plainly: the prompt engineer is an emerging role that requires urgent attention, and the training on offer is almost non-existent. Yet few crafts are so misunderstood: it is not about chatting with machines, but about designing, validating and maintaining the instructions, context and controls of generative AI systems. The method, from the official guides of the model makers.
Let us start with the European figure, because it is the sharpest. The skills-needs analysis carried out by the EU co-funded Artificial Intelligence Skills Alliance (ARISA) warns: “An emerging role that requires urgent attention is prompt engineer”; and it adds, on the training-supply side, that “there is hardly any supply related to emerging demand like prompt engineering”. A demand declared urgent and a supply almost absent: this is the snapshot of a craft born faster than the paths to learn it. All the more reason to say precisely what an AI Prompt Engineer actually does.
What it really does, and what it does not
No small number of misunderstandings have gathered around this craft. The prompt engineer does not train models: they design, validate and maintain the instructions, request templates and tool chains through which generative AI systems are deployed, safeguarding the accuracy, safety, traceability and fairness of the outputs, in collaboration with the technical profiles and with the legal and compliance functions. The object of the work is not the model, but the boundary between the model and the organisation: what goes in, what comes out, and the rules for both.
The method, from the makers’ official guides
The documentary peculiarity of this craft is that its primary sources are the official technical guides of the makers of Large Language Models (LLMs). The definition is offered by Google’s documentation: “Prompt design is the process of creating prompts, or natural language requests, that elicit accurate, high quality responses from a language model”. On method, OpenAI’s guide recommends: “Be specific, descriptive and as detailed as possible about the desired context, outcome, length, format, style, etc”. And on the technique of examples, Anthropic’s documentation observes that “a few well-crafted examples (known as few-shot or multishot prompting) improve accuracy and consistency”.
Behind these indications lies a consolidated repertoire: structured instructions with delimiters, breaking tasks into steps, chaining requests, control of the generation parameters, constrained output formats. Nothing esoteric: the engineering of instructions, with quality criteria and test methods.
Where professionalism is decided
Three practices separate the professional from the improviser. The first is to treat prompts as engineering artefacts: versioned, documented, gathered into reusable libraries; improvisation, in corporate settings, is the premise of unreliability. The second is empirical evaluation: sets of test cases and metrics to measure accuracy, consistency and robustness before going into production and at every change of the model or the instructions; without evaluation, any claim of quality is anecdotal. The third is the oversight of the risks specific to generation: reducing so-called hallucinations by grounding in data and Retrieval-Augmented Generation (RAG) techniques, guardrails against misuse and malicious instructions, and attention to distortions in the generated content.
To this is added compliance: logging the relevant interactions, transparency towards users about the use of generative AI in line with the obligations of Regulation (EU) 2024/1689 (the AI Act), and respect for Regulation (EU) 2016/679 (the GDPR) when personal data pass through the requests or the context.
Italy has written the profile
To the urgent demand flagged in Europe, Italy has responded with a verifiable perimeter: the UNI 11621-8:2026 standard includes the AI Prompt Engineer among the professional role profiles of AI, with competences set out according to the methodology of the European e-Competence Framework, the UNI EN 16234-1 standard. Anyone wishing to have those competences attested may undergo assessment by an accredited body under the UNI CEI EN ISO/IEC 17024 standard, within the framework of Accredia Information Circular DC No. 21/2026: a systemic answer to a market in which the title has, until now, been self-assigned with remarkable ease.
Conclusions
Prompt engineering sits between the model’s capabilities and the reliability of the result: to underestimate it is to hand users systems that are brilliant and unreliable. Europe has certified its urgency, the model makers have written its method, the Italian standard has fixed its competences: the ingredients of professionalisation are all there.
In the light of the above, one wonders whether training and business will close the gap flagged by the European analysis before self-certifications close it, badly; the difference, for those who will use these systems, will be one not of label but of reliability.
