The return-to-office fight is really about control. Yet while leaders argue about where people sit, a more basic shift is underway: who or what does the work. Engagement is sliding, particularly among managers. Only one in five employees worldwide report being engaged, and manager engagement fell from 27 percent to 22 percent in the last year, a sharp drop that signals strain in the old supervisory model (Gallup, 2025). At the same time, most executives say they will expand their workforce with AI agents. In a global survey across 31 countries, 82 percent of leaders said they plan to use digital labor to boost capacity in the next 12 to 18 months (Microsoft WorkLab, 2025). The result is a simple management test. Can you organize around outcomes, not headcount?
From org chart to work chart
Self-managing organizations have long argued that the manager–subordinate relationship is not a law of nature. Lee and Edmondson define self-managing organizations as those that decentralize authority radically and systematically across the whole firm and eliminate the formal manager–subordinate tie (Lee and Edmondson, 2017). Their review notes that managerial hierarchy is good at reliable execution in stable settings but tends to inhibit cross-boundary problem solving and reinforce status differences that stunt development. That diagnosis has new urgency when work is split between humans and software.
Two recent playbooks converge on the same structural answer. Both argue for replacing the static org chart with a dynamic work chart that maps outcomes, accountabilities, and the human and agent resources attached to them. Microsoft reports that 46 percent of leaders already use agents to fully automate workflows or processes, and it frames the frontier firm as one that deploys agents at scale and uses a work chart to coordinate them (Microsoft WorkLab, 2025). McKinsey describes an agentic operating model as a flat network of outcome-aligned teams that exchange tasks and outputs through a work chart rather than through traditional delegation chains (McKinsey, 2025). These are not peer-reviewed studies and they come from organizations with a commercial interest in AI adoption, so the claims should be read as directional rather than definitive. Still, the architectural move is consistent with what the SMO literature predicts will help in complex, nonroutine work that crosses functions (Lee and Edmondson, 2017).
What early adopters are learning
The headline numbers are blunt. In organizations Microsoft labels as frontier firms, where AI has been deployed organization-wide with active agent use, 71 percent of employees say the company is thriving, versus 39 percent in the global worker average reported by the same survey (Microsoft WorkLab, 2025). The comparison is not causal and it comes from a vendor report, but it signals that confidence is clustering where agent adoption is not piecemeal.
McKinsey adds a concrete operating ratio from client work. In its experience, a human team of two to five people can supervise an agent factory of 50 to 100 specialized AI agents that run an end-to-end process such as customer onboarding or closing the books (McKinsey, 2025). Treat this as a consultancy estimate, not a universal law. The managerial implication is the point. The locus of control shifts from supervising individual contributors to engineering and auditing the flow of work. That is close in spirit to self-management: elevate decision rights to the level of the process and make accountabilities explicit in the work system, not the hierarchy (Lee and Edmondson, 2017).
The promise is not just speed. It is fit for the problems modern firms face. When outcomes are defined clearly and the handoffs are explicit, cross-functional work stops depending on heroic coordination. The risk is also plain. If leaders rush in without the social conditions that support learning, the agent layer becomes brittle and blame gets pushed down the stack.
The human variable: psychological safety and the manager’s new job
The people side is not a soft add-on. Edmondson’s seminal field study with 51 teams found that team psychological safety was associated with learning behavior, and that learning mediated the link to performance (Edmondson, 1999). The study was cross-sectional in a single firm, so it does not prove causation across all settings. It does, however, set a practical bar. Human-agent organizations will generate errors, surprises, and redesigns. Teams that do not feel safe to surface problems will not improve the work chart. They will hide defects and stall.
This matters at a moment when managers themselves are disengaged. Manager engagement has dropped nine percentage points since 2022, and only 22 percent report being engaged in 2025, compared with 79 percent in best-practice organizations, which shows the design gap between common and better management systems (Gallup, 2025). The agent era can widen the gap if it treats managers as hall monitors for software. Or it can close it by redefining the job as steward of outcomes, coach of teams, and owner of data quality. Microsoft’s report points to emerging hybrid managerial roles that mix product thinking, process design, and agent orchestration, a shift aligned with its proposed work chart (Microsoft WorkLab, 2025). Framed this way, the “great flattening” is not the end of management. It is a redesign of what managers manage.
How to start, and what to measure
A practical path begins with structure, not slogans. The following steps are design guidance rather than universal prescriptions. First, publish a live work chart for a single end-to-end process. Name the outcome, list the human roles, list the agent roles, and make the handoffs explicit. This borrows from Microsoft’s concept while keeping the scope tight (Microsoft WorkLab, 2025). Second, staff a small human control room in the McKinsey ratio range to supervise an initial fleet of agents, with a clear cadence for exceptions and improvements (McKinsey, 2025). Treat the ratio as a planning heuristic, then adjust. Third, invest heavily in team practices that raise psychological safety. Make it routine to log and discuss failures, questions, and redesigns, because learning behavior is the pathway to better performance in complex work (Edmondson, 1999).
Measurement should track outcomes, learning, and sentiment. Outcomes belong on the work chart. Learning can be proxied by the rate of documented process improvements. Sentiment, especially among managers, should be watched closely, since engagement is a leading indicator of whether the design is working or simply burning people out (Gallup, 2025).
The close is straightforward. If your operating model still allocates power through a pyramid, you will spend more time arbitrating turf than improving flow. Human-agent organizations invert that logic. They put the work at the center and then ask what combination of humans and agents can best deliver it. The firms that make that move will look flatter, feel safer to speak up, and, if the early reports hold, will feel more like they are thriving.
Sources
- Microsoft WorkLab (2025). Microsoft Work Trend Index 2025 — The Year the Frontier Firm Is Born. Reports that 82% of leaders plan to use digital labor soon, 46% already automate workflows with agents, and frontier firms report higher thriving, supporting the shift to a work-chart model. 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. Offers consultancy estimates that 2–5 humans can supervise 50–100 agents and describes a flat, work-chart operating model, illustrating how human-agent teams can run 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 on when hierarchy helps or hinders, grounding the case for outcome-centric, less-hierarchical designs. https://www.sciencedirect.com/science/article/abs/pii/S0191308517300059
- Edmondson (1999). Psychological Safety and Learning Behavior in Work Teams. Shows that team psychological safety is associated with learning behavior that mediates performance, underscoring the human conditions needed for human-agent work. https://journals.sagepub.com/doi/10.2307/2666999
- Gallup (2025). State of the Global Workplace 2025. Finds global engagement at 20%, manager engagement down to 22%, and a wide gap to best-practice organizations, highlighting the strain on traditional supervision and the need for redesign. https://www.gallup.com/workplace/659610/state-global-workplace.aspx