HR AGENTIC AI: WHAT IMPACTS FOR LABOUR AND MANAGEMENT HR?

 On 4 February, the second workshop of the 2025-26 edition of theHR Innovation Observatoryentitled 'Agentic AI: What Impacts for Business and Management HR?" in which some results of a survey by the Observatory to understand how artificial intelligence (AI) is changing HR organisations and processes in Italy were presented. Following the presentation of the data, it is spoke by Filippo Virdis, HR & Organisation Application Manager of SEA Milan who reported on an ongoing initiative in his organisation to use Agentic AI solutions to support administrative activities.

The workshop then continued with a workshop activity in working groups, in which participants shared experiences of using AI in their organisations' HR processes and reasoned about how Agentic AI solutions can be implemented to address the needs of talent attraction, training and development, performance management and administration processes.

 The Observatory's research

Artificial Intelligence in organisations: where do we stand?

In 20 years, will it be as difficult to imagine a world without Artificial Intelligence as it is today to imagine a world without the internet?

However plausible, it is difficult to answer this question now. However, there are other three relevant issues and present in the current public debate, which were the subject of the workshop's analysis: How well do AI solutions currently perform? Is the scale such that we are afraid they will steal our jobs? Are we heading towards a future where the workforce will be made up of people and machines?

Let us start with the first question. First of all, according to some research, the outputs of AI systems are increasingly difficult to distinguish from those produced by humans. An experiment by two researchers at the University of California (Jones and Bergen, 2025) tried to find out how effectively people are able to distinguish an output produced by humans from one produced by artificial intelligence. The result was surprising. When instructed to adopt a specific personality (young, introvert, internet culture connoisseur), AI models have, in fact, proved particularly effective. The most successful AI system was judged to be human in 73% of cases. Also with regard to specific skills, according to Kiela et al. (2023), AI solutions already seem to have surpassed average levels of some people's skills, including image recognition and language interpretation.

The increase in effectiveness is going hand in hand with the complexity of these solutions. In the past year, there has been a marked increase in interest in the 'new'.Agentic AI', i.e. artificial intelligence tools 'endowed with their own initiative', as defined by the Artificial Intelligence Observatory of the Politecnico di Milano. These solutions show greater complexity than classical chatbots, because they are able to perform certain actions autonomously and, above all, they present a hierarchical structure that may provide for a subdivision of agents on several levels (one speaks in this regard of a 'multi-agent structure'). An example of this technological architecture can be found in the research by Hosseini and Sailani (2025), which states three hierarchical levels among 'AI AgentsAt the top of this structure is the so-called master agentthat is, the main agent that supervises and manages the operation of the entire system, providing instructions for the subsequent actions of each of the sub-agents; a step further down are the orchestrator agentagents who coordinate activities and allocate tasks to the various subordinate agents; finally, the last level is represented by the micro agentor agents specialising in individual activities.

Is the increase in effectiveness and complexity of these tools translating into improved business performance? In 2025, the HR Innovation Observatory calculated an average saving of about 30 minutes per day for those who use AI to support their individual productivity. Aoday, we therefore have some evidence of increases in the individual's time to perform specific tasks, but this does not necessarily translate into increased productivity of the individual and, above all, that there are measurable effects on the organisation's performance. For this to happen, organisational maturity is required in reviewing activities and processes to integrate them with Artificial Intelligence.

For instance, if we look at the news about the impact of AI on job opportunities. the thoughts are still open. Certainly some newspaper articles about organisations announcing major staff cuts, justifying them on the basis of the efficiency gained from the new AI solutions adopted, have made noise. Similarly, just as much resonance was gained by news reports on how, a few months later, the companies themselves started hiring people again because the efficiency gained with AI tools did not compensate for the contribution of humans.

Beyond the media hype of some announcementsIn fact, if we look at data on the US labour market, which is more susceptible to these fluctuations because it is more flexible and less rich in worker protection than the Italian market, the number of dismissals by companies in 2025 is in line with the annual average since the early 2000s.

If, in quantitative terms, there seems to be no change in the number of job opportunities, one effect AI is having is on the demand for certain profiles. The negative impact of IV has been discussed in the public debate on these issues. for two categories of people in particular: younger profiles, because they have less experience and perform more operational tasks, which are considered easy to automate with AI, and specialist profiles with IT skills, given the effectiveness of certain AI tools in software programming activities for instance. In the Italian context, however, there is no such evidence. According to data from the HR Innovation Observatory, organisations report an increase in the demand for junior profiles and a sharp rise in figures with IT skills. These data indicate that the impact of these technologies should not be taken for granted and that it may vary according to the context of reference.

A substantial change, however, across job profiles, concerns the relevance of AI-related skills in the world of work. As proof of this, in some countries, such as the US and China, the importance of these skills in education is beginning to be concretely considered. For example, at Purdue University in the US, the possession of certain AI-related skills is an entry requirement to enter university, while even in China they are introducing training on these tools from the age of 6.

The level of possession of skills related to the use of AI solutions is also relevant for answering the third question presented at the beginning, related to the relationship between humans and machines in work contexts. If we will indeed have colleagues who are not human, it will be crucial to know how best to interact with them. But how close is this scenario to the current reality? Going back to China, humanoid robots capable of autonomously replacing batteries are operating in some factories to be operational 7 days a week, 24 hours a day. However, the introduction of AI solutions will not only be limited to production contexts. Even in more office-based jobs there will be, and in some realities already are, digital colleagues. So much so that, in some contexts, the ability to interact and collaborate with new 'colleagues' is already being assessed in candidates during the selection phase.

So if some companies are starting to use AI tools as colleagues, other organisations are starting to declare themselves 'AI-first', thus explicitly putting not only the human at the centre, but also AI solutions. And this means rethinking processes by putting artificial intelligence, and consequently the collaborative relationship with the human being, at the centre.

However, how far are companies moving towards these models? From the Observatory's data, the impact is still limited on business models, rethinking of organisational models and accountability of decision-making processes. The responses of the research sample thus show that AI is still perceived as a useful tool to support specific and limited activities, and not as a digital colleague with which to collaborate more broadly and which implies a rethinking of the organisational structure.

Research data on AI in HR Directorates in Italy by the HR Innovation Observatory

For more information how artificial intelligence is influencing the HR processes of organisations in Italy, the Observatory's research also investigated the level of deployment of these solutions, the investments in AI and the role played by the HR Directorate in this change.

The results show that AI solutions are now most present in the Talent Attraction processa figure in continuity with last years' surveys. More surprising, however, is the growth recorded in the area of performance management, where 33% of organisations claim to use AI-based solutions: a significant increase compared to last year, when the percentage was around 20%.

However, looking at the decisions to adopt these instruments one figure above all is critical: only 36% of the HR directorates, among those that claim to have AI solutions, have an active role in defining policies for the use of these tools, in identifying its impacts and actions in terms of competencies, roles and organisational models, together with the company's top management. This is relevant because the absence of HR management from the decision-making tables limits its ability to govern transformation processes, exposing it to the risk of simply managing their effects.

The limited influence of Directorate HR on strategic choices is also evident from the investment figures. Compared to the previous year, the number of organisations investing in AI in HR processes has grown only marginally, reaching just over half. Future prospects appear similarly dim: less than one in two organisations plans to increase investment in the next 1-2 years.

Two more figures highlight the poor strategic input of the HR Directorate into business decisions. On the one hand, in fact, in most cases these solutions are adopted: or by inertia, as existing digital solution providers automatically introduce updates with AI functionality, without an active choice on the part of the organisation; or reactively, where, compared to the case of inertia, following updates by technology suppliers, there is an active choice of the AI features to be introduced among those proposed by the supplier company. On the other hand, the main reason stated by the HR departments for introducing these tools is related to increasing operational efficiency, while more strategic aspects, such as the strengthening of the HR's decision-making role through the increased use of objective analysis and data, are much less mentioned.

So what is preventing the HR Directorates from taking on a more strategic role? The research shows that underlying this dynamic is the lack of specific skillsand within the HR Directorate, which consequently also results in a limited awareness of potential AI use cases. This deficiency manifests itself particularly in two specific competences: the ability to assess the appropriateness of solutions and the ability to design or adapt roles and activities according to the integration of AI into work processes.

It is therefore clear that, although these solutions are often adopted with the support of other business areas, such as the IT Department, a HR Department lacking these skills will hardly be able to participate in strategic decisions on the use of these technologies and will increasingly risk being affected by them, without the ability to govern and direct them.

The testimony of SEA Milan

The workshop continued with the testimony of Filippo Virdis, HR & Organisation Application Manager of SEA Milan, who shared with the participants the experience of introducing Agentic AI solutions in HR processes that he followed in SEA Milan.

SEA Milan's HR team, assisted by ICT colleagues, is in fact introducing an AI agent, represented by a chatbot equipped with Artificial Intelligence, within the activities involved in supporting employees in retrieving information related to administrative aspects. The tool was inserted within a process that already provided two other contact channels for requests for assistance: a phonic tree, i.e. a telephone system that directs the user's call to the correct office or information voice through a sequence of voice or numeric commands; and a web form, i.e. an online form that employees can fill in to forward requests or reports to the most appropriate area. With the inclusion of the chatbot, the company has therefore added a new contact channel.

In terms of user experience, the user can ask the chatbot any questions about administrative aspects. To answer, the AI agent acting as chatbot retrieves information from a knowledge base compiled by SEA Milano's HR team.

The introduction of this tool therefore broadens the channels to which employees can turn in the event of a need for assistance, but at the same time it has shown three main limitations. These included the need to maintain an up-to-date knowledge base, to always provide answers in line with the regulations in force at the time of the interaction. A second limitation is that the tool answers questions in a standard way, not customised on the basis of the user with whom it interacts. Finally, the tool shows difficulties in tapping into the source of data, often confusing information from related, albeit different, areas, thus sometimes providing information beyond the users' original requests.

To overcome these difficulties, the HR team therefore initiated an evolution of the system by introducing a multi-agent structure, where the original chatbot, in addition to maintaining the frontend role towards end users, is enriched with an 'orchestration' role to query specialised agents and obtain information. On the one hand, agents dedicated to specific subject areas of the knowledge base, managed and updated by internal contact persons, will be inserted. On the other, an AI agent connected to the company's HCM system is being developed, capable of accessing employee information and providing customised answers based on the user's profile.

According to this evolutionary scenario, therefore, in the course of interactions, thehe main chatbot employs specialised agents, retrieving through triage the one most relevant to the employee's request. According to Virdis, this architecture allows the system to enrich the chatbot offering through a scalable model that can be integrated with further employee services (e.g. maintenance, ICT helpdesk) and third-party AI agents.

THE EXPERT'S OPINION

“The evolution towards theAgentic AI marks a paradigm shift for HR Directorates. AI is not just automation: just as the steam engine replaced horses, today it redefines the concept of value in businesses. Traditional HR structures and standardised roles are becoming obsolete, while the HR becomes a knowledge architect, designing algorithms and learning ecosystems. To fully exploit the potential of AI, organisations need to review processes and roles, and HR departments need to acquire specific skills to actively participate in strategic decisions and drive digital transformation, instead of passively undergoing it.” Filippo Cannavò, Head BU Service & Training of Alveria.

Source: Observer HR Innovation