What the announcement asks for
The mechanism itself is simple. Those who participate must declare whether they have used artificial intelligence systems in the preparation of the technical offer, and whether they intend to do so during the execution of the assignment. The announcement associates a condition with this declaration, namely that the prevalence of intellectual work is ensured, together with the control and verification of the results obtained.
It is worth mentioning that this is not an original idea of the announcement. The formula comes from article 13 of Law 132/2025, which establishes a basic principle for intellectual professions: artificial intelligence can be used for instrumental and support activities, but cannot take the place of the intellectual content of the performance. The announcement limits itself to bringing that principle into an assignment procedure, and in this it fits into a broader framework, made up of the European Regulation, the national law and now the ANAC indications, which converge towards the same idea of the centrality of the person.
On the level of the statement it is a fairly clear indication, and it is when we try to translate it into practice that the most interesting questions emerge.
A principle that is difficult to measure
The first concerns the way in which the prevalence of intellectual work should be demonstrated concretely. Neither the law nor the announcement indicates a threshold, a percentage or an objective criterion. This is a qualitative principle, entrusted to a test that no one has really defined yet, and the difficulty becomes evident as soon as you look at the way in which studies work.
When a project is born with the support of artificial intelligence, what remains are mostly the results, some intermediate files, the final outputs. What is often missing is a document capable of reconstructing the decision-making process: how the professional posed the problem, which paths he evaluated, what he discarded and for what reasons.
To understand how concrete the issue is, you can think of a competitor who generates a large number of design simulations and selects one. Is that choice sufficient to speak of human prevalence, or did the machine carry out the work and the professional was only left with the task of indicating the preferred solution?
Where is the human contribution located?
Part of the difficulty probably arises from the fact that we seek prevalence at the wrong time. It is not in having given up the tool that the designer’s contribution is measured, and not only in the final act of choice, but also in what comes before and what comes after.
First there is the necessary competence to decide which data to provide to the tool, how to formulate the request, how to set up the reasoning. Two professionals who turn to the same system with the same question obtain different results, because their ability to query it is different.
Afterwards there is the verification, correction and engineering work that serves to transform a plausible result into an actually achievable and compliant solution. The artificial intelligence returns something that looks like a correct answer, but determining whether it really is remains a task of the professional, and a responsibility of those who sign. In this sense, AI does not reduce the required competence, but shifts it towards the judgment and control of results, which remain far from simple activities.
The practical problem that remains open
There remains the concrete difficulty of documenting a prevalence that lives in judgment and control, activities which by their nature leave few traces. Added to this is the point of view of those who are called upon to evaluate.
At present, tender commissions rarely have the technical tools to ascertain the origin of content or to measure its reliability, and a declaration obligation that relies on a verification that is difficult to carry out remains less than solid.
What remains to be clarified
The ANAC Model Notice n. 2 does not resolve the question of prevalence: it puts it in writing and leaves it open, effectively postponing the task of specifying it to practice and the next calls for tenders. Some questions remain on the table that would be worth addressing before, and not after, jurisprudence begins to produce answers on a case-by-case basis.
The first is what should actually be declared. Document how artificial intelligence has been used – with which tools, in which phases, with which prompts – is a very different burden from simply keeping traces of its use, as is already done with other project documentation.
The second concerns the meaning of that statement. Is declaring the use of AI a neutral act, or can it be penalizing to some extent? As long as there is no shared scale that says whether and how much that use affects the evaluation, it is not obvious that those who declare in more detail will be considered in the same way as those who limit themselves to the essentials. It is a possible distortion that would be worth preventing, and which probably only an explicit indication in the notice – on what to declare and with what effect on the tender – can avoid.
The weekly column “Architectural Prompting” is edited by experts Luciana Mastrolia, Giovanna Panucci and Andrea Tinazzo
>> If you are interested in these topics, also sign up to the free Linkedin Newsletter AI & Design for Technicians, we’ll talk about it here!
