Skip to main content

Team & Squad Model

ASDD adopts the squad-based structure pioneered by Spotify engineering. Each squad owns a business capability end-to-end and operates autonomously within shared architectural governance. The key evolution: AI agents are first-class squad members, not tools.


Organizational hierarchy

Tribe (Product Area)

A Tribe groups squads working within a related domain. Tribes share architectural standards, tooling, and a global Knowledge Agent that accumulates learning across squads.

Squads

Each squad owns a specific business capability from product intent to production operation. Squads are autonomous — they do not hand off work to other teams for implementation or QA.


Squad composition

RoleHeadcountFTEPrimary Accountability
Product Owner1 shared across 2–3 squads0.5Business intent, spec approval, outcome acceptance
Tech Lead11.0Technical integrity, ASDD lifecycle enforcement, agent governance
Engineers1–31.0 eachSpec fidelity, agent output review, dissent notices
AI Agent System6–10 agentsDiscovery, spec validation, design, implementation, QA, security, CI/CD

A squad of 3 humans + 8 AI agents has the implementation throughput of a much larger traditional team, while retaining formal human authority over every architectural and business decision.


Role definitions

Product Owner

Accountability: Meaning and Business Value

The PO in ASDD is responsible for clarity of intent, not granularity of tasks. The shift from traditional product management is significant:

The PO does:

  • Owns Capability Specs — structured descriptions of what the system should do, not how
  • Defines business rules, constraints, and measurable success criteria
  • Prioritizes the spec backlog and determines slice sequencing
  • Validates and approves intent.md before Phase 1 begins
  • Accepts sprint outcomes based on spec compliance, not demo performance
  • Exercises formal override authority when agent outputs conflict with business intent

The PO no longer does:

  • Write detailed user stories mid-sprint
  • Clarify requirements during implementation
  • Negotiate scope after sprint start

One PO may serve 2–3 squads, depending on domain complexity. The time savings come from the elimination of mid-sprint clarification — specs must be approved before sprint start.


Tech Lead

Accountability: Flow, Quality, and Technical Integrity

The Tech Lead is the most critical role in the ASDD framework. All agent escalations route to the TL. All phase gates require TL sign-off.

Responsibilities:

  • Enforces the ASDD lifecycle and phase gate discipline
  • Facilitates sprint ceremonies and spec readiness reviews
  • Ensures all specs pass the Validation Gate before sprint start
  • Owns technical coherence: architecture, domain model, and steering rules
  • Resolves all agent escalations requiring human judgment
  • Maintains the Agent Failure Log (see Governance)
  • Reviews and acknowledges Reasoning Traces at each phase gate
  • Coaches the squad in ASDD practices

Override authority: Full, at every phase. The TL's decision on agent escalations is final within the squad.


Engineers

Accountability: Execution and Spec Fidelity

Engineers in ASDD are agent orchestrators and quality reviewers, not primary code authors. The work shifts from writing to directing and validating.

Responsibilities:

  • Implement exactly what is defined in the spec — no undocumented behavior
  • Identify spec gaps before sprint start (not during implementation)
  • Write tests mapped to spec behaviors
  • Emit events, logs, and metrics as defined in the spec
  • Reject undocumented behavior — if it isn't in the spec, it doesn't ship
  • File formal Dissent Notices when agent output is technically unsafe or spec-noncompliant

Engineer core principles:

  1. Specs before sprints — No work enters a sprint without an approved spec
  2. Single source of truth — The spec replaces scattered user stories and acceptance criteria
  3. Autonomous squads — Own the capability end-to-end; no cross-team handoffs
  4. Role clarity — Clear accountability eliminates ambiguous handoffs
  5. Flow efficiency — Optimize for fast, stable delivery over resource utilization
For junior engineers

Your primary learning path in ASDD is through spec-fidelity review — understanding why an agent's output does or does not match the spec teaches both domain modeling and system design faster than writing implementation code directly.


The AI Agent System

The AI Agent System is not a single tool — it is a coordinated pipeline of 10 specialized agents, each with a defined role and confidence threshold. See the Agent Catalog for full specifications.

At the squad level, the agent system functions as a highly capable, always-available execution partner that:

  • Works in parallel across multiple slices simultaneously
  • Never blocks on human coordination (within its authority boundaries)
  • Reports confidence scores, not just outputs
  • Escalates automatically when confidence falls below threshold

Decision rights

Clear decision ownership is what makes autonomous squads safe. The following matrix defines who decides what:

DecisionOwner
Business priority and spec sequencingProduct Owner
Spec approval (Phase 0 → 1)PO + TL
Technical approach and architectureTech Lead
Implementation details within specEngineers
Agent output rejection (Dissent Notice)Any team member
Agent escalation resolutionTech Lead
Self-Healing PR approvalTL + at least one Engineer
Security gate bypass (emergency only)TL (logged immutably)
warning

Any decision not on this table defaults to the Tech Lead. Ambiguous decision ownership is the most common cause of governance gaps.


The Knowledge Agent: shared intelligence

One Knowledge Agent operates at the Tribe level, accumulating learning across all squads. This is the system memory:

  • Maintains the State Manifest for each squad's slices
  • Detects conflicts when two squads modify the same shared domain entity
  • Proposes steering rule updates based on cross-squad failure patterns
  • Ensures context-fresh sub-agents receive only relevant domain terms

A single Knowledge Agent across 3–5 squads builds institutional knowledge faster than any human retrospective process.


Next

  • Maturity Model — the L1–L6 progression for phased ASDD adoption
  • Change Management — role transition guidance and resistance patterns
  • Governance — the technical details of confidence scoring and dissent protocols