Return-to-office edicts, brittle budgets, and sliding morale set the backdrop. Global employee engagement fell to 20 percent in 2025, the lowest since 2020, with Gallup estimating about 10 trillion dollars in lost productivity, or 9 percent of global GDP (Gallup, 2025). Manager engagement has dropped sharply too, from 27 percent in 2024 to 22 percent in 2025 (Gallup, 2025). Now leaders plan to scale capacity with AI agents. In a global survey of knowledge workers in 31 countries, 82 percent of leaders said they will use digital work agents to expand workforce capacity in the next 12 to 18 months, and 46 percent say they already automate entire workflows with agents (Microsoft WorkLab, 2025). The middle of the organization is strained just as the work is changing.
What frontier firms are signaling
Microsoft’s Work Trend Index labels the early movers “Frontier Firms,” defined by organization-wide deployment of AI and active use of agents. In those firms, 71 percent of workers say their company is thriving, compared with 39 percent globally (Microsoft WorkLab, 2025). The report argues for replacing the static org chart with a dynamic “Work Chart,” a map of outcomes, accountabilities, and the human and agent capabilities doing the work. It also anticipates new hybrid managerial roles, including the provocative idea of an “agent boss,” where orchestrating autonomous software becomes as central as managing people (Microsoft WorkLab, 2025). These are vendor-reported findings and concepts derived from surveys and product telemetry, so they are directional rather than definitive. Yet they capture a practical truth many operators feel: once work fragments into dozens of software agents, coordinating by title and tier is too slow.
From hierarchy to human–agent networks
Consultants see the same fault line. McKinsey estimates that 89 percent of organizations still run on industrial-age hierarchy, 9 percent on agile or product-and-platform models, and only 1 percent as decentralized networks (McKinsey, 2025). In its client work, McKinsey reports that a human team of two to five people can supervise an “agent factory” of 50 to 100 specialized agents executing an end-to-end process such as onboarding or closing the books (McKinsey, 2025). These are not peer-reviewed data, and they reflect one firm’s experience. They do, however, sketch the operating implication of agents: a flatter network of outcome-aligned units coordinated by work, not managerial rungs.
Management research offers a useful compass for that shift. Self-managing organizations decentralize authority radically and systematically, eliminating the formal manager–subordinate relationship, which is their defining feature (Lee and Edmondson, 2017). The literature reviewed by the authors argues that managerial hierarchy is well suited to reliable execution of known tasks in stable settings, but it hinders complex, cross-functional problem solving and can reinforce status differences that suppress development (Lee and Edmondson, 2017). Human–agent organizations do not need to copy every feature of self-management, but the direction of travel is similar: less reliance on positional authority, more on explicit accountabilities, transparent information, and peer coordination.
The critical human rule: safety before scale
Technology does not erase the human conditions for learning. In a seminal field study of 51 work teams, Amy Edmondson showed that team psychological safety, the shared belief that the team is safe for interpersonal risk taking, was associated with learning behaviors, which in turn mediated team performance (Edmondson, 1999). That causal chain matters in a world of agents. When autonomous systems act, people need to question outputs, surface defects, and report near misses without fear. Otherwise, errors compound quietly as agents speed them up.
Taken together, the evidence suggests a working thesis. Use agents to move routine work faster, but organize the human system for discovery, escalation, and redesign. This reframes the manager’s job from gatekeeping to stewardship. Managers become stewards of outcomes and standards, not approvers of every step, and they host the learning loops that keep agentic workflows safe to improve. Gallup’s data show that organizations with best-practice designs see 79 percent of managers engaged, far above the global average of 22 percent, which underscores that engagement is a design variable, not fate (Gallup, 2025).
How to start a human–agent organization
Do not begin with a wholesale reorg. Treat agents as the trigger to redesign one process end to end and make the design visible.
- Draw the work chart for a single business process. Name the outcome, handoffs, data sources, decision rights, and the specific human skills and agent capabilities involved (Microsoft WorkLab, 2025).
- Stand up a small supervision cell. Two to five people can monitor 50 to 100 agents in McKinsey’s client experience, but treat this as a planning heuristic, not a target (McKinsey, 2025).
- Make psychological safety explicit. Publish escalation paths, log reversals, and hold regular retrospectives. The point is to promote learning behavior, which Edmondson found links safety to performance through the way teams actually work (Edmondson, 1999).
- Set outcome metrics, not activity metrics. If the process touches customers or risk, add kill switches and audit trails. Frontier firms report using agents to fully automate workflows, which raises the bar for controls (Microsoft WorkLab, 2025).
- Redesign the manager role. Shift from span-of-control to span-of-outcomes. Define responsibilities like exception handling, standard setting, talent development, and cross-boundary coordination, consistent with the self-managing insight that hierarchy impedes complex problem solving when it hoards authority (Lee and Edmondson, 2017).
- Watch engagement and capacity together. Use a simple dashboard that pairs throughput and quality with pulse measures for the humans supervising agents. The global slump in engagement and the drop among managers are real constraints you need to manage, not wish away (Gallup, 2025).
The near-term prize is pragmatic. Microsoft’s data suggest that firms leaning into agents are reporting healthier performance climates, though those findings come from a vendor-run survey and telemetry, so they should be read with caution (Microsoft WorkLab, 2025). The longer-term prize is structural. As more work becomes computable, advantage will come from how cleanly you map outcomes to capabilities and how quickly your teams learn when the map is wrong. That is not a software rollout. It is an operating choice: retire the org chart as the primary coordination artifact and replace it with a living chart of work, then build the human conditions that let the chart improve.
In a year defined by RTO battles and efficiency drives, this is the constructive provocation. Stop trying to make yesterday’s hierarchy supervise tomorrow’s agents. Start teaching your organization to manage work directly. The firms that do will buy capacity without burning out their managers and will turn agent speed into compounding learning rather than compounding risk.
Sources
- Microsoft WorkLab (2025). Work Trend Index 2025 — The Year the Frontier Firm Is Born. Reports 82% of leaders plan to expand capacity with digital work agents, 46% already automate workflows, and introduces Frontier Firms and the Work Chart. https://www.microsoft.com/en-us/worklab/work-trend-index/2025-the-year-the-frontier-firm-is-born
- McKinsey (2025). The agentic organization: contours of the next paradigm for the AI era. Estimates 89% of firms still run on industrial-age hierarchy and observes, in client work, 2–5 humans supervising 50–100 agents across end-to-end processes. https://www.mckinsey.com/capabilities/people-and-organizational-performance/our-insights/the-agentic-organization-contours-of-the-next-paradigm-for-the-ai-era
- Lee and Edmondson (2017). Self-managing organizations: Exploring the limits of less-hierarchical organizing. Defines SMOs and synthesizes evidence that hierarchy suits stable execution but impedes complex cross-boundary problem solving. https://www.sciencedirect.com/science/article/abs/pii/S0191308517300059
- Edmondson (1999). Psychological Safety and Learning Behavior in Work Teams. Shows team psychological safety is associated with learning behaviors that mediate performance, a foundation for safe improvement in human–agent teams. https://journals.sagepub.com/doi/10.2307/2666999
- Gallup (2025). State of the Global Workplace 2025. Finds global engagement at 20% with an estimated $10T productivity cost, manager engagement down to 22%, and 79% manager engagement in best-practice organizations, underscoring design effects. https://www.gallup.com/workplace/659610/state-global-workplace.aspx