Key points
- AI agents are already doing work, but most leadership teams haven't agreed who owns them.
- Every agent needs a named human owner, a supervisor and a clear escalation path.
- Redesign the workflow before adding the agent, or you'll automate the dysfunction.
- Accountability for agents is an organization design question, and belongs with HR as much as IT.
Something quietly changed on most organization charts this year. Alongside the boxes for people, there's now work being done by AI agents: drafting responses, reconciling accounts, triaging requests, running parts of a process from start to finish.
The problem isn't the agents. It's that almost nobody has decided who is responsible for them. In a July 2026 survey of CHROs, Gartner found that only 24% of leadership teams are aligned on who should be accountable for AI agents. At the same time, Gartner reports that 63% of CEOs and senior executives are radically redesigning their operating models to get returns from AI. That is a lot of redesign with very little clarity about ownership.
Why this is an HR question
When a person does work, we know how to handle accountability. There is a role description, a manager, performance expectations and a way to escalate when something goes wrong. When an agent does the work, none of that exists by default. Gartner also notes that agents shift ambiguity, judgment and recovery work back onto employees, which means people end up carrying responsibility for outcomes they didn't design.
Deciding who owns what, who supervises whom and who makes which decisions is organizational design. It is the same discipline HR applies to people, extended to a new kind of teammate.
Four moves to make now
1. Name an owner for every agent. One accountable person, by name, not a team or a department. They own the agent's purpose, performance and retirement.
2. Define supervision and escalation. Who reviews the agent's work, how often, and what happens when it gets something wrong? Write it down like you would for a new hire.
3. Redesign the work first. Microsoft's Frontier Firm research puts it simply: lean the process before adding agents. Automating a broken workflow only makes it break faster.
4. Update the people side. Roles change when agents take on tasks. Update role descriptions, performance expectations and the skills you hire and develop for, especially judgment, quality control and the ability to delegate work well.
Keep the important decisions human
The organizations getting this right are clear about one line: decisions about people stay with people. Hiring, promotion, discipline and termination may be informed by data, but they are made by accountable human leaders.
If your leadership team can't answer "who owns this agent?" for every agent in production, that's the place to start.