Leap Trends · Structure · Organisation design
Also touches on the condition Technology · How we choose the condition
Briefing
Who is the agent's boss? Ownership, mandate and control of AI agents
- Governance of AI agents
- Mandate and ownership
- Agent autonomy
In more and more organisations AI agents carry out real steps in the work and will gain more autonomy in the coming years, while almost nobody has set up the governance: who owns the agent, what it may do without a check, who switches it off. That is a structural question for line management, not an IT question.
This briefing has not been updated for more than three months. Check the figures against the source before you use them.
Compiled with AI from public sources, with the source for every figure. Key figures labelled Checked have been compared with the source by a person. How that works
Key figures
- 5%of companies currently have the full set of controls for AI agents in place; 42% want to give agents autonomy by 2030.Boston Consulting GroupNot yet checked
- 22%of total AI value comes from agents in 2026, against 17% a year earlier; expected by 2030: about two fifths.Boston Consulting GroupNot yet checked
- 6%of leaders say they are making progress in deliberately designing how people and AI work together.DeloitteNot yet checked
- 48%of banks and insurers are creating new roles to oversee AI agents; 10% already use agents widely.Capgemini Research InstituteNot yet checked
What the sources say
BCG surveyed more than 1,300 executives and sees agents’ share of AI value growing fast, from 17 to 22 per cent in a year, with two fifths expected by 2030. More than four in ten companies want to give their agents autonomy by then. Five per cent currently have the full set of controls in place. According to BCG, companies that do get three times as much value out of them.
Deloitte hears from six per cent of leaders that they are making progress in deliberately designing how people and AI work together, and names clear decision rights as a missing building block. In HBR, Srinivasan and Wei describe a new role, the agent manager: someone responsible for how a fleet of agents learns, performs and works safely alongside people, and who steers mainly from dashboards. Whether that becomes a separate job is open; that it is a mandate which has to sit somewhere is clear from all three sources.
Capgemini Research Institute already sees the role emerging. In a survey of 1,100 leaders at banks and insurers in fourteen markets, conducted in the summer of 2025, 48 per cent say they are creating new positions for staff who oversee agents, while 10 per cent use agents widely. A year later the same institute surveyed 800 executives in the automotive industry: 68 per cent use or are testing agentic AI in product development and 5 per cent describe their own AI governance as mature. Trust in agents also falls the closer the work gets to product safety, from 59 per cent in early-stage research to 40 per cent in development.
In October 2026 Josh Bersin described, for the HR function, a set-up with collaborating agents and six decisions that determine whether it works. These include how the agents are governed and monitored, where people must stay in charge and who is accountable when an agent makes a mistake. According to Bersin, those decisions belong with the leadership of the function and explicitly not with IT.
How solid is the evidence? It comes mainly from large American companies, the roles are new and there is no second year of measurement yet. The direction is clear; the practice is thin.
For the leadership team
This is the same question leadership teams have asked about people for years: who may decide what, and who answers for it. Except that an agent does not volunteer that it doesn't know something. Every agent that already takes a step on its own belongs on a list with an owner, a mandate and an off switch; wherever one of the three is blank, there is a question for the board.
Sources
- Capgemini Research Institute68% of automotive organizations use or pilot agentic AI for product development, but only 5% report mature AI governance (Rewiring automotive product development)
very few (5%) organizations report fully mature AI governance.
- The Josh Bersin Company · Josh BersinIntroducing The Leader's Guide To Multi-Agent AI Architecture
Importantly, these are not configuration questions for IT.
- Boston Consulting Group · Jessica Apotheker, Vinciane Beauchene, Nicolas de Bellefonds et al.The Formula for Agentic AI Value
only 5% of companies have the full set of critical controls in place today.
- DeloitteDeloitte Report: Winning Organizations Will Build the Human Advantage (2026 Global Human Capital Trends)
- Harvard Business Review · Suraj Srinivasan and Vivienne WeiTo Thrive in the AI Era, Companies Need Agent Managers
- Capgemini Research InstituteBanks and insurers deploy AI agents to fight fraud and process applications, with plans for new roles to supervise the AI (World Cloud Report in Financial Services 2026)
48% of financial institutions are creating new roles for employees to supervise agents.
We summarise in our own words and quote no more than one sentence per source. Read the original: there is more in it. How we use sources.
Academic research
- UniNettuno University · English and Van den BergAre Organisations Ready to Hand Over the Keys to Agentic AI?
- Université Paris Nanterre · Gluszak and GluszakDelegated Agentic Governance: A Delegation-Centred Framework for Managing Autonomous AI in Organisations
- Kozminski University · Korzyński et al.When Agentic AI Participates in Organizational Change: Rethinking Agency, Leadership, and Accountability
- University of Manchester · Beulen and DansArtificial Intelligence Governance Mechanisms—The Chief Data Officer Perspective with a Focus on Agentic AI Governance
Academic articles on the same phenomenon, no more than a year old at the last search, from universities in Europe and North America, found through the open database OpenAlex. These articles count towards the publications per month; the text and key figures above are based on the sources. How we choose the articles.
How this briefing has been updated
Title now describes the topics; previously “Who is the agent's boss?” (AI)
Added 4 articles from universities in Europe and North America (via OpenAlex). (AI)
First draft based on BCG, HBR and Deloitte. (AI)