Team structures
Who works with whom, who decides what, and who looks after each person: from flat companies with no managers to layered hierarchies.
How it works
Every organization answers three questions, on purpose or by accident: how people are grouped into teams, who decides what each team works on, and who handles each person’s growth, pay and problems. A managed (hierarchical) organization answers with reporting lines: each person has one manager, and managers report to managers. The number of people one manager looks after is the span of control; in the US, Gallup found the average rose to 12.1 in 2025, while the median stayed at five or six. A flat organization removes most of those lines. At Valve, which published its handbook in 2012, nobody “reports to” anybody and people pick their own projects; under Holacracy, adopted by Zappos and Medium, people hold written roles inside self-governing circles instead of job titles. In practice structure appears either way: Valve’s own handbook says temporary structure forms inside every project, and a lead emerges as a clearinghouse of information. Most software companies sit in between: small, cross-functional teams that own a product area, with engineering managers for people and tech leads for technical direction. Team Topologies (2019) adds a vocabulary for how teams relate, and Conway’s law (1968) warns that whatever you choose, your software will tend to copy it.
A jazz band and an orchestra. The jazz band has no conductor and works because a handful of players listen closely to each other; try it with ninety musicians and you need someone at the front with a baton, and sections with principals.
Reach for it when
- Flat: a small group of senior people who know each other and can coordinate by talking
- Managed: past a few dozen people, when someone must own hiring, mentoring, pay and hard conversations
- Any size: deciding team boundaries on purpose, since the system’s architecture will tend to follow them
Think twice when
- Calling a company flat while decisions are really made by an informal inner circle
- Adding layers of managers to a team of five, or giving one manager thirty direct reports
- Copying another company’s model (Spotify’s squads, Zappos’s Holacracy) without the problems it was built for
A real example
A 60-person engineering group has eight teams of six to eight. Six are stream-aligned, each owning a part of the product end to end (checkout, search, accounts); one platform team runs the deployment pipeline and databases as an internal product; one engineer acts as an enabling team, helping others adopt tracing. Each team has a tech lead who guides technical decisions, and four engineering managers each look after about 14 people across two teams, handling one-on-ones, hiring and pay. Search and checkout talk through a documented API because their teams do.
Flat and managed organizations compared#
| Question | Flat (self-managed) | Managed (hierarchical) |
|---|---|---|
| Who decides what to work on | Each person or self-formed team (Valve’s “cabals”, Holacracy circles) | A manager or product owner sets priorities with the team |
| Who looks after you | Peers: reviews, ranking and feedback come from colleagues | Your manager: one-on-ones, growth, pay, performance |
| Where it works | Small groups of experienced people who can coordinate face to face | Large groups, many new hires, work that needs long-range planning |
| Typical failure | Hidden hierarchy and slow cross-team coordination; weak mentoring | Layers, hand-offs and managers too far from the work |
| What its users admit | Valve: not good at mentoring, spreading information, or predicting beyond a few months | Spotify’s line-managed “chapters” cut across teams and, by one former engineer’s account, caused more problems than they solved |
- Most companies mix the two: autonomous teams for daily work, managers for people. The real choices are team size, span of control and how decisions travel.
Team Topologies: four team types, three ways to interact#
Matthew Skelton and Manuel Pais (2019) argue that every team should be one of four types, and that teams should interact in only three ways, chosen on purpose.
Stream-aligned
most teamsOwns a flow of work, usually a slice of the business domain, end to end: “you build it, you run it”, with no hand-offs.
Platform
internal productProvides an internal product, such as deployment or data infrastructure, that speeds up stream-aligned teams.
Enabling
temporary helpHelps stream-aligned teams overcome obstacles and pick up new skills, then moves on.
Complicated subsystem
specialistsOwns a part that needs significant specialist expertise, such as heavy mathematics or calculation.
Interaction modes
three onlyCollaboration (working closely for a set time to discover something), X-as-a-Service (one team provides, another consumes with little contact) and facilitation (one team helps and mentors another).
Often confused with
Where it came from1968–2020
In April 1968 Melvin Conway argued in Datamation that “organizations which design systems … are constrained to produce designs which are copies of the communication structures of these organizations.” Fred Brooks later named it Conway’s law in The Mythical Man-Month. Two years after Conway, Jo Freeman’s talk “The Tyranny of Structurelessness” (published 1972) warned from the women’s movement that “there is no such thing as a structureless group”: removing formal structure leaves informal structure, which “becomes a way of masking power.”
In 2012 two documents made software companies rethink management. In March, Valve published its Handbook for New Employees: Valve is flat, “we don’t have any management, and nobody ‘reports to’ anybody else,” desks have wheels, and people join whatever project they think matters most. In October, Henrik Kniberg and Anders Ivarsson described Spotify’s squads (small teams like mini-startups), tribes of up to about 100 people, and chapters whose leads were line managers. Many companies copied both. Zappos began rolling out Holacracy in 2013, and in March 2015 Tony Hsieh announced that from 30 April there would be no more people managers, offering at least three months’ severance to anyone who preferred to leave; about 210 of 1,500 employees, 14%, took it.
The reversals followed. Medium dropped Holacracy in March 2016, saying it was “difficult to coordinate efforts at scale” and that writing down every responsibility “hindered a proactive attitude and sense of communal ownership.” GitHub, which like Valve had let people choose their own projects, abandoned that structure for something much more traditional, a reversal organization scientists later studied as a case. In 2020 a former Spotify engineer wrote that the squad model had been “only ever aspirational and never fully implemented.” Meanwhile Skelton and Pais’s Team Topologies (2019) moved the debate from “managers or not” to how teams are shaped and how they interact.
Why it’s built this way
- Remove managers so people choose their own work
- Valve argued that hierarchy suits predictability, but telling talented people what to do “obliterates 99 percent of their value.” Its handbook also lists what this costs: mentoring, spreading information and long-range planning.
- Replace job titles with written roles and governance
- Holacracy tried to make authority explicit rather than abolish it: roles with accountabilities inside circles, changed through a formal governance process. Medium found the record-keeping slow and the press coverage harmful to recruiting.
- Split people management from daily teamwork
- Spotify’s 2012 model let squads run themselves while chapter leads, working in squads too, handled people’s growth and salaries. A former Spotify engineer later wrote that this matrix management introduced more problems than it solved.
- Design teams to get the architecture you want
- If systems copy communication structures, you can shape the system by shaping the teams first: the “inverse Conway maneuver”, named by Jonny LeRoy and Matt Simons in 2010 and central to Team Topologies.
Timeline
- 1968
Melvin Conway’s “How Do Committees Invent?” appears in Datamation
- 1970
Jo Freeman’s talk “The Tyranny of Structurelessness”
- 2012
Valve’s Handbook for New Employees (March); Kniberg and Ivarsson’s Spotify paper (October)
- 2013
Zappos begins rolling out Holacracy
- 2015
Zappos ends people managers and offers severance; about 14% leave
- 2016
Medium moves beyond Holacracy
- 2019
Skelton and Pais publish Team Topologies (second edition 2025)
- 2020
“Spotify doesn’t use ‘the Spotify model’”, a former engineer’s account
- Melvin Conway: How Do Committees Invent? (Datamation, April 1968)
- Martin Fowler: Conway’s Law (named by Fred Brooks; inverse Conway maneuver, 2010)
- Jo Freeman: The Tyranny of Structurelessness (talk 1970, published 1972)
- Valve: Handbook for New Employees (first edition, March 2012)
- Henrik Kniberg and Anders Ivarsson: Scaling Agile @ Spotify with Tribes, Squads, Chapters & Guilds (Oct 2012)
- Quartz: Internal memo, Zappos is offering severance to employees who aren’t all in with Holacracy (Mar 2015)
- Fortune: 14% of Zappos’ staff left after being offered exit pay (May 2015)
- Andy Doyle, Medium: Management and Organization at Medium (Mar 2016) · archived copy, original
- Burton et al.: GitHub, exploring the space between boss-less and hierarchical forms of organizing (Journal of Organization Design, 2017), abstract
- Jeremiah Lee: Spotify’s Failed #SquadGoals (Apr 2020)
- Team Topologies: the book (2019; second edition September 2025)
Sources8
- Gallup: Span of control, what’s the optimal team size for managers? (average 12.1 in 2025, median five to six)
- Valve: Handbook for New Employees (flat structure, team leads, “structure happens”, what Valve is not good at)
- Holacracy Constitution 5.0: roles, role leads, circles and governance
- Team Topologies: Key concepts (four team types, three interaction modes)
- Kniberg and Ivarsson: Scaling Agile @ Spotify (squads, product owners, chapter leads as line managers)
- Jeremiah Lee: Spotify’s Failed #SquadGoals (matrix management, quotes from Spotify coaches)
- Charity Majors: The Engineer/Manager Pendulum (May 2017), on managers and tech leads
- Martin Fowler: Conway’s Law (microservices as a tool to structure a development organization)