De-layering announcements all read the same: spans widened, layers collapsed, savings booked. What the announcements never specify is sequence, and sequence is the entire game. The companies that flattened and thrived did the structural work before removing managers. The companies that flattened and broke removed managers and hoped structure would emerge. Both look identical on day one. They diverge around month nine, when the first cross-team conflict finds no broker and the first struggling employee finds no coach.
The pressure to flatten is real and partly justified. Deloitte's 2025 Human Capital Trends finds managers spending nearly 40% of their time firefighting and doing administrative work, against 13% developing people, while US middle-management job postings fell 42% between spring 2022 and end-2024. Gallup finds manager engagement dropping faster than any other group's, from 27% to 22% in a single year. A layer this squeezed, this disengaged, and this automatable invites the cut. The question is not whether many organizations will de-layer; it is why some will get Buurtzorg's outcome and others will get chaos, and the research on both is unusually clear.
What the successes did first
Buurtzorg, the standing proof that scale without middle management is possible, over 900 self-managing nursing teams, two directors, no middle managers, is routinely cited and routinely misread. The longitudinal research on it (Martela and Nandram, 2025) stresses that the structure was designed flat from the founding, around hard constraints: teams capped at twelve, low interdependence between teams, and a support layer of about 21 coaches for roughly 950 teams, coaches who hold no formal authority and no productivity responsibility. The researchers' warning is explicit: copying the structures without the underlying philosophy tends to fail, and firms like GitHub and Valve added hierarchy as they grew rather than sustaining flatness. Buurtzorg is not evidence that removing managers works; it is evidence that a system engineered so that managers are unnecessary works. The engineering is the point.
The cautionary tale on the other side is Spotify, whose famous squads-and-tribes model became the most copied org design of its decade. By the account of insiders, including agile coach Joakim Sundén, the model was aspirational even internally: "even when we wrote the whitepaper, we weren't doing it." The company fixated on team autonomy but never completed the alignment and accountability mechanisms the model promised, producing organizational chaos as it scaled, and its co-author acknowledged the awkwardness of companies copying something that did not really exist. Autonomy shipped first; alignment never shipped. That is the failed sequence in one sentence.
The order of operations
Read the successes against the failures and a sequence emerges in three steps that do not compress.
First, move decision rights and information down, while the layer still exists. People cannot absorb authority they have never practiced. This is where transparency does its quiet work: a team cannot own decisions without seeing the numbers those decisions move. There is a vivid precedent, as Ricardo Semler has told it, in Semco's 1985 factory-scheduling experiment: management predicted the assembly line would collapse when workers set their own hours, and instead the workers coordinated themselves, parents taking mornings, single workers evenings, under one stated rule: talk to the person next to you. The scheduling authority moved before anything else changed; adults, given real information and a real decision, behaved like adults. Meta-analytic evidence backs the general bet: Ryu and colleagues' 2022 synthesis finds team autonomy positively related to team effectiveness, with the payoff conditional on task design and organizational support, conditions, again, that must be built in advance.
Second, build peer accountability to replace vertical accountability. The manager's least visible function is being the place where conflict, coordination, and performance problems land. Flat systems that survive route these to explicit peer mechanisms while the safety net still exists, so the new muscles develop under supervision. Skip this and the old accountability disappears the day the layer does, with nothing underneath.
Third, and only third, remove the layer, and convert rather than discard the people in it. Deloitte's 13% figure is an indictment of the managerial role's design, not of the humans in it; the coaching and development work they were prevented from doing is precisely what the flattened organization needs more of. Buurtzorg's coaches, influence without authority, show what the post-layer version of the job looks like.
The test before the announcement
The sequence also yields a simple audit for any de-layering plan: for each thing the middle layer currently does, name its destination, a team, a mechanism, or a coach, and show that the destination existed before the removal date. If the plan is mostly a savings number and a new spans-and-layers chart, it is Spotify's sequence with worse branding. Flattening is not subtraction. It is a construction project that ends with a subtraction, and the ending, done in the right order, is the easy part.
Sources
- Deloitte (2025). Global Human Capital Trends. Managers spend ~40% of time firefighting/admin vs 13% developing people; US middle-management postings down 42% (spring 2022 to end-2024). https://www.deloitte.com/us/en/insights/topics/talent/human-capital-trends/2025.html
- Gallup (2025). State of the Global Workplace. Manager engagement fell from 27% to 22% in one year, the steepest drop of tracked groups. https://www.gallup.com/workplace/659610/state-global-workplace.aspx
- Martela & Nandram (2025). Buurtzorg: scaling up with hundreds of self-managing teams but no middle managers. Journal of Organization Design. Flat by design: capped teams, low interdependence, 21 coaches without authority for ~950 teams; copying structure without philosophy tends to fail. https://link.springer.com/article/10.1007/s41469-024-00184-y
- Lee, J. (2020). Spotify's Failed #SquadGoals. Insider account: the squad model was aspirational and never fully implemented; autonomy without alignment mechanisms produced chaos at scale. https://www.jeremiahlee.com/posts/failed-squad-goals/
- Ryu, Neubert & Gonzalez-Mulé (2022). Putting the team in the driver's seat. Personnel Psychology. Meta-analysis: team autonomy positively relates to team effectiveness, conditional on task and support factors. https://onlinelibrary.wiley.com/doi/10.1111/peps.12468