Teams

Modern XP works with people, agents, or both. The practices do not change. What changes are the roles in team coordination and work execution.

Team Topology

Team Topology is a framework for organising teams (from Team Topologies). Four types exist. Most teams are stream-aligned. The other types are for specific needs.

Every team defines a Team API: services, SLOs, communication channels, change notifications. Without a clear API, teams step on each other.

Human Teams

A human team is stream-aligned and cross-functional. The team is small enough to hold its work in collective working memory. Anyone can talk to a stakeholder or user directly. No gatekeepers.

Key Characteristics

• Stream-aligned. The team owns a complete value stream from start to end. No handoffs.
• Cross-functional. Skills needed to ship live inside the team. Engineering, testing, operations, domain knowledge. The team does not wait for another team to do its job.
• Cognitive load bounded. Nobody is responsible for more than they can hold. If the domain is too large, split the team. Do not split the person.
• Psychological safety. Respect is a performance multiplier. Blameless retros. Sustainable pace. Collective code ownership. When an incident happens, the team asks “what caused this?” not “who caused it?”

Agentic Teams

An agentic team runs the same Modern XP practices with agents as executors. The practices stay the same. The execution changes.

What Stays the Same

• Same values: Communication, Simplicity, Feedback, Courage, Respect. Same practices: TDD, CI, pair programming, retrospectives, delivery cycles.
• Same work items: Features, Stories, Use Cases, Studies, Tasks, Tests.
• Same quality gates: unit tests, integration tests, SAST, SCA.

What Changes

• Roles become specialised agents with system prompts and input/output contracts.
• Communication flows through an Orchestrator. Hub-and-spoke model. Agents do not talk to each other directly, unless specifically designed to do so.
• State lives in structured workspaces. Artifacts follow defined schemas.
• Processes run from specifications. Each step has a role, an action, an input, an output, and a timeout.
• Error handling is automated. Retries, fallbacks, and escalation to human.

Use agentic teams when: scale demands it (>5 active items), speed needs sub-minute responses, consistency is critical, auditability is mandatory, 24/7 coverage is necessary.

Use human teams when: work is exploratory, highly creative, or needs deep contextual understanding that agents do not have yet.

Human-Agent Partnership

The human sets intent and writes the spec. The agent executes autonomously within safety boundaries. No continuous human supervision. The human reviews the output at the gate. Approve, reject, or iterate.

Key rule: do not intervene during agent execution. If intervention is needed, the boundaries or spec are wrong. Fix the spec. Do not fix the execution.

Roles

A role is a responsibility, not an individual. One individual can hold multiple roles. One role can be fulfilled by a person or an agent. The guide describes what each role does. The team decides who fills it.

Not all roles are useful in every situation. The team decides when they are needed, and when to remove them.

Which Model to Use

Teams are not purely human or purely agentic. A human team can use agentic tools for specific activities (test generation, documentation, code review). An agentic team has a human operator for escalation and strategic decisions.