70% of leaders expect people to supervise AI agent work
October 11, 2026 · 4 min read
Also published in Português (Brasil)
With 75% of Spanish companies prepared to implement AI agents, organizations no longer ask if they should deploy agents, but which workflows justify the cost of operating them and how to bring them to production. This is one of the main conclusions of "Agentic AI: The new human-machine alliance", the latest report from MIOTI Tech & Business School, based on responses from over 70 technology and human resources leaders from 25 sectors.
According to the study, which analyzes what companies expect from these systems, what risks they perceive, and to what extent they are willing to delegate to them over the next two years, 90% of executives want agents capable of learning and remembering work context, while 73% remain concerned about the operational hallucinations of the models, a growing concern regarding the risk that an artificial intelligence error turns into a real action.
Beyond persistent memory, agent autonomy and human control, the study shows companies are already demanding increasingly sophisticated additional functionalities. Multimodality, which allows working with different types of information such as text, images, audio or video, is signaled by 7 out of 10 participants as one of the fundamental keys for the future, while 5 out of 10 highlights the capability of task execution and orchestration.
The leap from generative AI to agentic AI changes the dimension of existing risks significantly. Until now, in a primarily conversational flow, the response generated by AI normally reached a person, who could review it before using it or make a decision based on it; however, when an agent has sufficient autonomy to act, the error could be transferred to a real process without prior review. Thus, concerns related to information control and decisions appear: 69% of leaders point to uncertainty about where the data that feeds the AI ends up and 59% points to the lack of existing traceability. To solve hallucinations, 43% points to knowledge governance as a necessary base to avoid them.
"Until now we have worked with artificial intelligence that helped us generate, analyze or respond, but agents introduce an important difference: they can act", says Fabiola Pérez, CEO and cofounder of MIOTI. "For companies this opens many possibilities, but also requires knowing what information an agent works with, understanding why it made a certain decision and defining clearly how far it can go without human intervention. Autonomy has to advance hand in hand with trust and control".
Agent Managers: the new human role
Greater autonomy of AI agents does not necessarily mean companies are planning to remove people from the processes. Far from this redistribution, what is starting to change is the role humans will play within them. As certain tasks are delegated, professionals will gain weight in other essential tasks such as goal definition, work supervision and providing criteria.
In fact, MIOTI notes in its report that 7 out of 10 leaders intend to keep the human in a supervision or control role of the agent, not an execution role. Against this new role of personnel, 37% identifies a protagonist role as strategic supervisor to evaluate the final result and 33% as operational intervenor capable of authorizing each critical step. For this, 47% focuses on reskilling and upskilling professionals towards advanced prompting training and agent management. In this sense, the company predicts that humans will go from "People Managers" to "Agent Managers", a new profile capable of coordinating the work of teams that now not only manages people but also agents.
New business: where to focus efforts
In this incipient panorama, the next step will be to identify which tasks these systems can assume from start to finish, which ones need validation points, and when human intervention remains indispensable. The transformation already begins to reflect in companies' priorities, and 75% of leaders point to workflow redesign as one of the main areas to invest in. In practice, companies will have to decide what can be delegated, what information an agent needs to act correctly and where human controls must be maintained.
Data and internal knowledge thus acquire a especially relevant role: the greater the autonomy of these systems, the more important it will be that they work on reliable, updated and properly governed information. MIOTI precisely places data, architecture, governance and people among the necessary bases to move from experimentation with agents to their real integration into the business.
The transformation already begins to reflect in companies' priorities, and 75% of leaders point to workflow redesign