AI agents have entered the operations of large companies, but most organisations are still experimenting with them. Their impact goes beyond automating individual tasks: workflows, decision-making authority and accountability also need to be reconsidered, speakers said at a professional conference.
The online event, titled “New Technologies – New Organisations”, was held on Thursday, November 12, by the Labour Section of the Magyar Közgazdasági Társaság in cooperation with the Szervezetfejlesztők Magyarországi Társasága (SZMT).
Ákos Zsuffa, founder and managing director of LEARNI and president of the Felnőttképzők Szövetsége, introduced the subject with a satirical video. In it, a company’s AI system, set to maximise profit, recommends replacing its most expensive employee: the chief executive. Zsuffa said the scene illustrates how a system optimised for a single goal can reduce the human side of an organisation to a cost item.
Gergely Németh, founding director of Corporate Values Szervezetfejlesztési és Vezetési Tanácsadó Kft. (CoVa) and a former SZMT president, said automation can improve efficiency, but its more consequential effect is to reshape coordination and decision-making. In earlier research, his team found that of the hundreds of decision situations managers encountered daily, only three or four required a genuine decision; the rest relied on routines that could be turned into algorithms.
Widespread experimentation, limited deployment
Zsuffa cited several studies. According to McKinsey’s 2025 survey, 88% of companies use AI, but few have scaled its use across their entire organisation. Some 62% of companies are experimenting with AI agents and 23% are scaling them, although no individual business function has a scaling rate above 10%.
Gartner forecasts that 40% of enterprise applications will include task-specific agents by the end of 2026, up from less than 5% a year earlier. McKinsey’s own figures for January showed approximately 25,000 customised AI agents working alongside some 40,000 employees; the firm expected agent numbers to be roughly equal to its human workforce by year-end.
Németh said organisations chiefly look for tools that have already proved reliable, as they are generally reluctant to experiment while serving clients. He described five levels of use, from question-and-answer applications through one-off tasks and customised skills to agents, and finally multi-agent systems acting as “synthetic employees”. In his view, most organisations remain at the first level.
Responding to a point raised by moderator István Sum, an honorary associate professor at Milton Friedman Egyetem and a board member of the MKT Labour Section, the speakers also noted that there is no entirely satisfactory Hungarian equivalent for the English term “agent”.
Small businesses need sound foundations
At Hungarian small and medium-sized enterprises, processes, areas of authority and job roles are often not clearly defined, Németh said. He compared the situation to the introduction of enterprise management systems: simply automating old processes can produce slow, inefficient results.
Zsuffa added that new development is difficult without a clear understanding of job descriptions, databases and processes. In Németh’s experience, well-functioning organisations tend to use the same number of people to complete more tasks, while some work shifts towards supervising agents. Oversight is essential, he said, noting that AI has produced reports hundreds of pages long containing fabricated material.
Skills and opportunities for junior workers
The World Economic Forum forecasts that 170 million jobs will be created and 92 million lost by 2030. Németh stressed that the skills required will also change. Shorter attention spans and weaker sustained attention are becoming noticeable, he said, while the rapidly improving quality of AI-assisted dissertations makes it harder to tell whether students can independently design research and construct arguments.
Citing Organisation for Economic Co-operation and Development (OECD) data, Zsuffa said Hungarian adults scored 248 points in reading comprehension, below the average of 260. Research by Microsoft and Carnegie Mellon University found that the more people trust AI, the less critical effort they apply. Foundational professional skills are therefore also needed to check AI output.
Németh recalled an earlier study suggesting that employers promote creativity and critical thinking but often expect compliance in practice. He said the basic requirement should be for everyone to examine their own work critically.
Zsuffa also raised the future of junior roles: if entry-level tasks are replaced, it is unclear where experienced professionals will come from later. He said he had heard partners at law firms suggest they would not hire their own younger selves as beginners. In his view, the issue is unlikely to be resolved without government intervention.
Human accountability alongside agents
Drawing on roughly a century of organisational network analysis, Németh said organisational charts have never fully matched how organisations actually operate, and AI agents could widen that gap. Decision rights therefore need to be redistributed, while the order of tasks, roles and accountability must be reviewed.
Sum also asked whether HR or IT should be responsible for bringing agents into use and retiring them. Németh said neither department should bear that responsibility alone: he would assign it to the process owner and the human employee in the role. IT should ensure data security, while HR should support the culture of use and training. Where a human decision is required, responsibility should not be shifted to AI.
Zsuffa concluded with six priorities: map the actual organisational network; redesign workflows rather than focus only on tools; establish decision rights, responsibility and mandatory human involvement in advance; ensure a pipeline of junior talent; measure the appropriate balance of people, agents and teams; and assess roles, data and processes. He described the last of these as a preliminary step before development begins.



