Every vendor deck now shows the same picture: the pyramid dissolving into a network, boxes replaced by nodes, reporting lines replaced by "work charts." Microsoft's 2025 Work Trend Index proposes exactly that, an outcome-oriented Work Chart to replace the org chart, and reports that 82% of surveyed leaders expect to use digital labor to expand workforce capacity within twelve to eighteen months. McKinsey goes further, sketching an "agentic organization" in which small human teams of two to five people supervise fifty to a hundred specialized agents running a process end to end. The imagery is always horizontal. The question a CEO should ask is whether the structure underneath is actually getting flatter, or merely thinner.
The two words get used interchangeably. They describe opposite moves.
Thin is not flat
A thinner organization removes people while leaving the topology of authority intact. The approval still flows to the same place; there are simply fewer stops on the way, and fewer humans employed at each stop. This is the version currently being built. Deloitte's 2025 Human Capital Trends notes that US employers were posting 42% fewer middle-management vacancies at the end of 2024 than in spring 2022, and cites Gartner's prediction that by end of 2026, 20% of organizations will have used AI to flatten their structure, eliminating more than half of their current middle-management positions. Note the verb: eliminate. Nothing in that prediction says the decisions those managers made were pushed downward. In most cases they are pushed upward, or into the software.
A flatter organization is a different animal. It redistributes authority: the people (or teams) closest to the work gain real decision rights, budgets, and accountability, and the remaining hierarchy exists to serve them rather than to check them. Thinness is a cost decision. Flatness is a power decision. AI makes the first one easy and the second one merely possible.
The gravitational pull of thin
There is a reason thin is the default. When an agent can draft the report, reconcile the numbers, and route the exceptions, the executive who owns the agent effectively absorbs the judgment that used to be distributed across a layer of humans. Automation, left to its own incentives, centralizes: it moves discretion toward whoever configures the system. McKinsey's own diagnosis is telling here. The firm estimates that 89% of organizations still run on an industrial-age operating model, and only 1% behave as a decentralized network. If that estimate is even directionally right, then the vast majority of companies bolting agents onto their processes are bolting them onto pyramids. A pyramid with agents is not a network. It is a pyramid with better plumbing and fewer witnesses.
Organizational research has long warned about a related illusion. Foss and Klein, studying self-described flat companies, found that many retain real bosses and real hierarchy, just informal and renamed; the formal chart flattens while an invisible one hardens. The AI version of this will be worse, because the invisible hierarchy will partly live in system prompts, agent permissions, and workflow configurations that no one outside the platform team can read.
What flat would actually require
The honest test is old. Ricardo Semler, as he has told it, took over his father's company at 21 and tore up the org chart; veteran managers predicted destruction, some quit, and the company grew tenfold. The argument that won over his father was not philosophy but a number: the people who set their own hours delivered more. Flatness, in other words, was never about removing boxes from a diagram. It was about moving real decisions, schedules, methods, spending, to the people doing the work, and then holding them accountable for outcomes.
Applied to agents, that test yields concrete questions. Who can deploy an agent: only the center, or any team with a budget? Who reads the agent's logs: only compliance, or the people whose work it touches? When an agent's recommendation conflicts with the judgment of the person on the ground, who wins by default? If the answers all point upward, the company is thinning. If they point outward, it is flattening.
There is a defensible case for either. A firm in a stable, regulated business may rationally choose thin: fewer people, tight central control, agents as instruments of the center. What is not defensible is buying "flat" rhetoric and building "thin" reality, because the failure mode is specific: a shrunken middle, an overloaded top, and a workforce that has lost both its managers and its authority at the same time.
The org chart AI "wants" is neither. Software has no preferences. The chart you get is the one your defaults produce, and the defaults, right now, produce thin. Flat still has to be chosen, and it is chosen the same way it was in 1980s São Paulo: by handing the decision to the person nearest the work and accepting the discomfort that follows.
Sources
- Microsoft WorkLab (2025). Work Trend Index: The Year the Frontier Firm Is Born. 82% of leaders plan to use digital labor to expand capacity in 12-18 months; proposes the "Work Chart" as org-chart successor. https://www.microsoft.com/en-us/worklab/work-trend-index/2025-the-year-the-frontier-firm-is-born
- McKinsey (2025). The agentic organization. Consultancy estimate: 89% of orgs still industrial-age, 1% networked; client cases of 2-5 humans supervising 50-100 agents. https://www.mckinsey.com/capabilities/people-and-organizational-performance/our-insights/the-agentic-organization-contours-of-the-next-paradigm-for-the-ai-era
- Deloitte (2025). Global Human Capital Trends. US middle-management postings down 42% (end-2024 vs spring 2022); cites Gartner prediction of AI-driven flattening in 20% of orgs by 2026. https://www.deloitte.com/us/en/insights/topics/talent/human-capital-trends/2025.html
- Foss & Klein (2022). Who's the boss? The persistence of entrepreneurial hierarchy in flat organizations. Journal of Organization Design. Nominally flat firms often retain informal hierarchy rather than eliminating it. https://link.springer.com/article/10.1007/s41469-022-00132-8