Skip to content

04.02 · Team Topologies Deep Dive

Level: Practitioner — Organisational · Reading Time: ~9 mins · Track: Org & Leadership Prerequisites: 04 · Team Effectiveness

Team Topologies (Matthew Skelton & Manuel Pais) is a widely used pattern language for fast flow, not a universal org-chart recipe. It gives Principal Engineers and Engineering Managers vocabulary for diagnosing dependencies, cognitive load, team boundaries, and interaction modes.

TL;DR — The 30-Second Mental Model

  • 4 Fundamental Team Types: Stream-Aligned, Platform, Enabling, and Complicated Subsystem.
  • 3 Interaction Modes: Collaboration for discovery, X-as-a-Service for clear service consumption, and Facilitating for temporary capability building.
  • Platforms Are Products: Start with the Thinnest Viable Platform that reduces user friction; measure adoption, reliability, and flow rather than platform feature count.

The Four Fundamental Team Topologies

Most delivery capacity normally sits in stream-aligned teams; the other types help those teams deliver and operate with lower cognitive load. Avoid imposing a fixed percentage or mapping teams by label without examining real work.

graph TD
    SA["Stream-Aligned Teams
Own end-to-end value flow and operations"] PT["Platform Team
Compelling internal product and self-service capabilities"] ET["Enabling Team
Temporary coaching and capability transfer"] CS["Complicated Subsystem Team
Deep specialist subsystem with a clear interface"] PT -->|"reduces extraneous load for"| SA ET -->|"transfers capability to"| SA CS -->|"encapsulates specialist complexity for"| SA style SA fill:#1976D2,color:#fff style PT fill:#388E3C,color:#fff style ET fill:#F57C00,color:#fff style CS fill:#7B1FA2,color:#fff

Detailed Breakdown of the Four Team Types

Aligned to a flow of customer or business value—a product, journey, domain, or other coherent stream.

  • Ownership: Design, build, test, deploy, operate, and improve its services, with explicit support and escalation.
  • Sizing: Small enough for shared context and effective communication; use observed cognitive load rather than a universal headcount rule.
  • Design Goal: Minimal blocking handoffs for routine value delivery.
  • Anti-Pattern: A feature factory that throws code to separate QA, operations, or release teams and owns no production outcome.

Builds and operates a compelling internal product that accelerates stream-aligned teams. A platform may itself contain multiple team types.

  • Ownership: Reliable self-service capabilities such as delivery workflows, compute, secrets, observability, and data provisioning.
  • Thinnest Viable Platform: Start with the smallest thing that solves user friction—even a curated standard or documented service catalog may be enough before building a large IDP.
  • Product Signals: Voluntary adoption, time to first safe deploy, task success, support burden, reliability, unit cost, and perceptual DevEx—not NPS alone.
  • Anti-Pattern: A manual shared-services queue rebranded as a platform, or a large platform built without validated internal user needs.

Helps stream-aligned teams overcome obstacles and acquire missing capabilities, then moves on.

  • Ownership: Capability transfer in areas such as OpenTelemetry, contract testing, threat modelling, accessibility, or AI evaluation.
  • Engagement Horizon: Defined by explicit independence/exit criteria rather than a universal number of weeks.
  • Success Metric: The receiving team can perform the capability sustainably without the enabling team.
  • Anti-Pattern: Permanently implementing or operating the stream team's services.

Owns a subsystem whose specialist mathematics, algorithms, or technical depth would create excessive cognitive load for stream-aligned teams.

  • Justification Bar: Demonstrate genuine specialist complexity and a clean consumable interface; ordinary legacy CRUD complexity does not qualify.
  • Examples: Real-time media codecs, specialized pricing engines, cryptographic infrastructure, or foundational model training.
  • Anti-Pattern: Recreating a component silo for ordinary code and forcing every consumer through synchronous coordination.

The Three Team Interaction Modes

Team interaction modes should be explicit and revisited as boundaries mature:

graph LR
    A["Team A"] <-->|"1. Collaboration
High-bandwidth discovery
Explicit exit criteria"| B["Team B"] C["Stream Team"] -->|"2. X-as-a-Service
Clear service contract
Self-service consumption"| D["Platform / Subsystem"] E["Enabling Team"] -.->|"3. Facilitating
Temporary coaching
Capability transfer"| F["Stream Team"]

|| Interaction Mode | Coordination Profile | Expected Evolution | Danger Signal | |:---|:---|:---|:---| || Collaboration | High-bandwidth joint discovery; valuable but expensive | End when the API, boundary, or capability is understood and exit criteria are met | Permanent collaboration hides unresolved ownership or a poor boundary | || X-as-a-Service | Lower coordination through documented APIs, support, SLOs, and predictable changes | Mature consumption with measured reliability and low user friction | Consumers still need routine tickets or absorb undocumented breaking changes | || Facilitating | Temporary coaching, workshops, pairing, and feedback | End when the receiving team demonstrates independence | Enabling team becomes the permanent owner of production work |

“X-as-a-Service” does not mean zero communication. Providers still need user research, support, deprecation policy, incident response, and contract evolution.


Evolution of Team Topologies Over Time

A Principal Engineer uses interaction modes dynamically:

  1. Discovery — Collaboration: Platform engineers work with a representative stream team to solve a real delivery problem and discover the smallest useful interface.
  2. Stabilization — Move Toward X-as-a-Service: Harden the workflow, docs, SLOs, security controls, migration path, and support model into a reliable paved road.
  3. Scale — X-as-a-Service + Facilitating: Other teams self-serve; an enabling team briefly helps with capability gaps while adoption and operational evidence guide product improvements.
  4. Reassess: Reopen short collaboration when assumptions change, rather than treating the initial team boundary as permanent.

Primary Team Topologies References


Active Recall & Self-Check

Self-Check 1: Eight product teams maintain separate CI/Terraform setups. How do you charter a platform team without creating a ticket queue?

Answer Rubric: 1. Start with user pain and collaboration: Build the first paved road with one representative team; do not standardize eight workflows by committee before understanding them. 2. Make routine fulfillment self-service: Provide tested modules, templates, documentation, policy checks, and observable operations. Keep an expert support path for exceptions and incidents. 3. Treat the platform as a product: Measure task completion, adoption, support burden, reliability, unit cost, and change lead time; deprecate unused capabilities. 4. Separate guardrails from preference: Teams may deviate for documented needs, but security, privacy, and regulatory controls remain explicit and enforceable. “Opt-in” is not a valid model for mandatory controls.

Self-Check 2: How do you diagnose excessive cognitive load, and which three levers can reduce it?

Answer Rubric: - Diagnose with evidence: Onboarding friction, excessive context switching, unclear ownership, high unplanned work, operational anxiety, and repeated mistakes across too many domains. - Reduce extraneous load: Move undifferentiated infrastructure/workflow friction into a reliable platform or automation. - Repartition germane load: Split an overly broad business stream along a coherent domain and ownership boundary; do not split merely to hit a team-size target. - Build missing capability: Use an enabling team or focused coaching when unfamiliar technology or practices create intrinsic load; define independence as the exit criterion.