For a century the manager's job had a stable core: assign work to people, monitor how they do it, and answer for the result. AI quietly rewrites the first two parts. When a model can draft, summarize, analyze, and increasingly act on its own, the manager is no longer allocating tasks among people alone. They are designing a workflow in which some of the work is done by humans, some by AI, and some by the two together, and getting that design right is a different skill from supervising effort.
The shift is already visible in what managers say about their own jobs. Deloitte's 2025 research found managers spending nearly 40% of their time firefighting and on administrative work, and only 13% developing their people, with 36% feeling unprepared for the human side of the role. Into that already strained job arrives a new responsibility: orchestrating machines as well as people. Microsoft's 2025 index reports 82% of leaders expecting to use AI agents to expand their teams' capacity within twelve to eighteen months, and 46% of leaders already using agents to fully automate some workflows. The manager who only knows how to supervise humans is about to be responsible for a team that is part human and part software.
From supervising tasks to designing workflows
Ricardo Semler has told how Semco let factory workers choose their own shifts in 1985. Management expected the assembly line to fall apart; instead the workers coordinated it themselves, and the entire policy reduced to one instruction: talk to the person next to you. The lesson was that the manager's job was not to direct each task but to design the conditions under which good work organizes itself. Human-AI workflow design is that lesson made literal. The manager stops issuing task-by-task instructions and starts deciding the architecture: which steps the AI handles, which a human owns, where the handoffs are, and where a human must sign off before anything ships.
This is genuine design work, and it has a failure mode. A Stanford-led audit of 1,500 workers across 844 tasks found people want automation on low-value work and shared control on high-value work, where a human keeps the judgment. A manager who automates the wrong steps, the ones people find meaningful or the ones that carry consequences, builds a workflow that is efficient and wrong. The skill is knowing which is which.
Onboard the agent like an employee
The most useful practical frame comes from the World Economic Forum's 2025 governance work: onboard an AI agent the way you would onboard a new employee, with a defined role, defined permissions, and defined supervision, and scale the oversight to the risk. McKinsey's 2025 work sketches the scale this can reach, describing, from client experience, small human teams of two to five people supervising far larger sets of specialized agents. Whether or not those ratios generalize, the management principle holds: an agent acting on the company's behalf needs a job description, boundaries, and a supervisor, exactly like a person does.
Accountability is the part that cannot be delegated to the tool. The peer-reviewed work by Chan and colleagues on visibility into AI agents argues that once tasks are handed to autonomous systems, someone must still be able to see what was done and answer for it, through identifiers, monitoring, and logs. A workflow where no human owns the agent's output is not a workflow. It is a liability with no name attached.
The human skills the job now needs
The capabilities this new job rewards are not technical in the narrow sense. They are the ability to decompose work into steps, to judge which steps are safe to automate, to design checkpoints that catch errors without strangling speed, and to hold people accountable for output they produced with a machine. The meta-analytic evidence on empowering leadership, drawn by Lee and colleagues from 105 samples, is a useful reminder here: leaders lift performance and creativity most when they build trust and give people real autonomy, effects that run through psychological empowerment. Designing a good human-AI workflow is empowerment by another name, deciding what to hand off so people can do the work only they can do.
The managers who struggle will be the ones who treat AI as a faster way to do the old job of directing tasks. The ones who thrive will treat it as a reason to do a new job: architect of a system where humans and machines each do what they are best at, and a human always owns the result. That job did not exist five years ago. It is now the center of the role.
Sources
- Deloitte (2025). 2025 Global Human Capital Trends. Managers spend ~40% of time firefighting and only 13% developing people; 36% feel unprepared for people management. https://www.deloitte.com/us/en/insights/topics/talent/human-capital-trends/2025.html
- Microsoft (2025). Work Trend Index 2025. 82% of leaders expect to use AI agents to expand capacity within 12–18 months; 46% already use agents to fully automate some workflows. https://www.microsoft.com/en-us/worklab/work-trend-index/2025-the-year-the-frontier-firm-is-born
- World Economic Forum (2025). AI Agents in Action: Foundations for Evaluation and Governance. Onboard agents like new employees, with defined role, permissions, and supervision proportional to risk. https://www.weforum.org/publications/ai-agents-in-action-foundations-for-evaluation-and-governance/
- Chan et al. (2024). Visibility into AI Agents (FAccT '24). Delegating to autonomous agents requires traceability so a human can still answer for the output. https://doi.org/10.1145/3630106.3658948
- Lee, Willis & Tian (2018). Empowering leadership: a meta-analytic examination (105 samples). Autonomy and trust lift performance and creativity through psychological empowerment. https://doi.org/10.1002/job.2220