Team Topologies for Agentic AI

Organising 100+ Engineers to Build Autonomous AI Systems

By The AK Dispatch

View All Frameworks

Conway's Law in Action

The structure of your teams determines what you can build. Design teams to create the architecture you want.

Four Fundamental Team Types

Stream-Aligned Teams

60-70% of headcount

Cross-functional teams owning end-to-end delivery of agentic AI products

4-6 teams• 10-12 people each

Platform Teams

20-25% of headcount

Build internal products that Stream-Aligned teams consume as self-service

4 teams

Complicated Subsystem Teams

10-15% of headcount

Handle parts requiring deep specialized knowledge

2 teams

Enabling Teams

5% of headcount

Help teams overcome obstacles and learn new capabilities

1 team

Three Interaction Modes

Collaboration

Working closely for defined period

X-as-a-Service

One team consumes with minimal interaction

Facilitating

One team helps another learn, then disengages

Evolution Strategy

Months 1-6: Foundation

  • ▸Stand up platform teams first
  • ▸Product teams in collaboration mode
  • ▸High coupling acceptable - learning requirements
  • ▸Prove platform value with 1-2 pilots

Months 6-12: Maturation

  • ▸Platforms harden interfaces
  • ▸Move to X-as-a-Service
  • ▸Product teams gain autonomy
  • ▸Add 1-2 more product teams

Months 12+: Scale

  • ▸Add product teams every 2-3 months
  • ▸Optimize for developer experience
  • ▸Split platforms if needed
  • ▸Extract patterns from research

Metrics That Matter

Product Teams

  • ✓Deployment frequency: Weekly
  • ✓Lead time: <1 day
  • ✓Customer outcome metrics

Platform Teams

  • ✓Adoption rate: Track usage
  • ✓Developer NPS: >40
  • ✓API uptime: 99.9%
  • ✓Time to onboard: <1 week

Organisation

  • ✓Cross-team dependencies: →0
  • ✓Cognitive load: <30% overwhelmed
  • ✓Top performer retention: >90%

Critical Success Factors

Platform Teams = Product Mindset

Treat engineers as customers, measure satisfaction

Clear API Contracts

Versioned APIs with docs and SLAs, full self-service

Inverse Conway Maneuver

Design architecture first, then align teams to it

Minimize Dependencies

Teams only depend on self-service platforms

Fast Feedback Loops

Product teams ship weekly, platforms never bottleneck

Team Stability

Give teams 12-18 months to gel before reorganising

The Payoff

Months 1-6: Productive chaos • Months 6-12: Velocity increases • Months 12-18: You're scaling

Months 18-24: Organisational competitive advantage

While competitors coordinate in meetings, you ship agentic AI products that work

Click on team types and evolution phases to explore details

Based on Team Topologies by Matthew Skelton and Manuel Pais