scalable/scale/team topologies/lesson 07 cph / /
lesson 07 / 07 · 12 min · updated ·

evolving the topology

teams change. how to move a team between types without blowing up delivery — and when not to.

topologies aren't permanent

a team topology is a snapshot. business changes, products mature, capabilities spread or concentrate — and the shape of the org has to move with them. the mistake isn't picking the 'wrong' topology; it's treating any topology as permanent. evolution is the default case, not the exception.

this lesson is about moving teams between types — when, how, and what to watch for.

the five transitions that actually happen

  1. stream-aligned → split into two stream-aligned the team got too big, too loaded, or started owning two coherent flows. the move is along a fracture plane: a boundary where each half can work for weeks without syncing. done well, both halves retain full end-to-end ownership of their slice.
  2. stream-aligned → platform (extraction) some capability the team built — ci tooling, a shared library, an ops dashboard — turns out to be needed by several other teams. the choice: either invest in making it a real platform product, or kill it and let consumers build their own. hybrid 'informal platform' arrangements almost always drift into fake-platform.
  3. complicated-subsystem → platform a specialist team's output stabilises, its interface matures, and consumers grow. at some threshold it becomes a platform team — the work isn't specialist anymore, it's productised. keep the name if it helps; change the operating model.
  4. enabling → end, or rotate enabling teams finish engagements. when they do, they either move to the next team that needs the capability, or they disband. the failure mode is drift into permanent attachment — always from an individual coach deciding it's easier to do the work than teach it one more time.
  5. any type → disbanded teams end. a product sunsets, a capability is no longer strategic, a merger removes a reason to exist. disbanding well — redistributing people, archiving knowledge, owning the handover — is a skill that most orgs are bad at and that pays for itself every time.

changing mode more often than type

most evolution isn't a type change; it's a mode change. the same platform team collaborates with a new consumer, then facilitates their onboarding, then settles into x-as-a-service. the same complicated-subsystem team collaborates with one team on an extension, then returns to x-as-a-service for everyone else. mode is cheaper to change than type; use it first.

read the signal. when someone describes a team as being in 'permanent collaboration' with another team, that's the signal. either the interface hasn't matured (fix the interface) or the two teams are really one team (merge them). either way, perpetual collaboration is a smell, not a state.

pick the right move

five transitions

each scenario is a team whose situation is pushing it to change. what's the cleanest move?

the discipline of writing it down

the thing most orgs don't do — and team topologies quietly insists on — is maintaining a team api: a one-page document per team stating its type, purpose, owned surfaces, interaction modes with neighbouring teams, and runbook for being reached. when a team evolves, the team api is the first thing to change. when a team api and the real behaviour diverge, the divergence is the first thing to fix.

the format doesn't matter. the habit does. a topology you maintain deliberately is a topology you can evolve; a topology you drew once and stopped looking at is already wrong.

that's the course

seven lessons, one framework. four types, three modes, cognitive load as a constraint, Conway's law running underneath everything. the book is longer and has more examples; most of what's there is elaboration on these primitives. the test you'll keep using is the same every time: is this team's type clear, is its interaction mode with its neighbours clear, and does that match its actual behaviour? if yes, you have a topology. if no, you have a reorg coming.

scalable labs·cvr 30091604·github·linkedin·hello@scalable.dk