1. Product 2. Agents 3. Governance 4. CloudOps 5. FinOps 6. Security

AI + Data + Cloud · Pillar 2

🤖

Talent Bench

37AI Agents

37 AI agents available instantly — 4 decision-makers (opus), 15 execution specialists (sonnet), 7 orchestrators, 4 operators (haiku), 4 digital transformation, 3 XOps. No hiring pipeline. No ramp time.

⚡ 15 agents available in <5 minutes via npx install

AI agents build governed & Humans ship trusted. 80% autonomy & 100% accountability.

Section Two (Ch.8-12)

Building Your Talent Bench

“No company can outsource its way to digital excellence. Being digital means having your own bench of digital talent — product owners, experience designers, data engineers, data scientists, software developers, etc.”

The goal should be to have 70-80% of your digital talent be in-house. ADLC addresses this with 37 AI agents available instantly — no hiring pipeline, no lead time. Decision Layer (4 opus), Execution Layer (15 sonnet), Orchestrators (7 sonnet), Operations (4 haiku), Digital Transformation (4 sonnet), XOps (3 sonnet).

Platform Evolution

Agent frontmatter is model-version-agnostic. Upgrade Claude Opus/Sonnet/Haiku — agents get smarter automatically. No code changes needed.

Talent Bench Golden Path

Each phase answers: Who does it, Why it matters, What if you skip it

1Assess

Evaluate current talent gaps against the 15-agent model

bash scripts/governance-score.sh

Who: HITL reviews gaps, developer-experience-engineer maps roles to agents

Why: You cannot build a team without knowing what you lack. The 3-tier model (opus/sonnet/haiku) ensures the right cost-performance tradeoff per role.

Skip? Build without specialists; product-owner decisions go unchallenged; security gaps undetected

→ Governance score + capability gap report

2Compose

Select agents for the project and map Decision/Execution/Operations tiers

/speckit.plan (agent delegation matrix)

Who: product-owner selects scope, cloud-architect maps architecture needs to agent capabilities

Why: Agent selection must match project scope. Under-staffing wastes HITL time on tasks agents can handle; over-staffing wastes tokens on unnecessary coordination.

Skip? Wrong agents assigned, COORDINATION_WITHOUT_DELEGATION violations, specialist domain mismatches

→ Agent roster with tier assignments + delegation matrix

3Coordinate

Establish PO+CA authority chain and wire enforcement hooks

enforce-coordination.sh + enforce-specialist-delegation.sh

Who: product-owner + cloud-architect approve before specialists execute

Why: Authority chain prevents STANDALONE_EXECUTION (anti-pattern #1). Hooks are deterministic — advisory rules degrade under context load, hooks do not.

Skip? STANDALONE_EXECUTION violations; agents work without business validation; ungoverned autonomous decisions

→ Coordination hooks active + authority chain enforced

4Execute

Specialist agents deliver within their domains via governed delegation

/speckit.implement

Who: AI specialist agents build autonomously, HITL reviews evidence at Phase 3+ gates

Why: Governed execution: agents prepare, humans decide, humans commit. Each agent works in its domain with full context.

Skip? Uncoordinated work, agents stepping on each other, duplicate effort, NATO violations

→ Working software + test results + evidence in tmp/

5Score

Sequential PO+CA+QA scoring — 3 agents, 1 foreground round

/speckit.retrospective

Who: 3 scoring agents run sequentially (PO then CA then QA), HITL reviews consensus

Why: Fix-then-score prevents SCORING_THEATER. Sequential execution prevents RACE_CONDITION_SCORING. 3 agents produce sufficient signal.

Skip? SCORING_THEATER — manufactured deltas; RACE_CONDITION_SCORING — contradictory evidence from parallel reads

→ Consensus score + gap analysis + corrective actions

6Evolve

Upgrade agent frontmatter, retire underperformers, add new specialists

/speckit.improve

Who: developer-experience-engineer proposes changes, HITL decides what ships

Why: Agent frontmatter is model-version-agnostic. Upgrade the underlying model, agents get smarter automatically. No code changes needed.

Skip? Stale agent definitions, no team improvement, capabilities plateau while models advance

→ Updated agent definitions + improvement backlog items

Start Here

Spec-Driven workflow and product skills — copy/paste to start

Solo Developer

You need a full team from day one. 15 agents, zero hiring, zero ramp time.

1.ln -s .adlc/.claude .claude

2./speckit.specify

3./speckit.implement

⚡ Full agent team in <5 minutes

Platform Team Lead

You need agent governance across your engineering team. Visibility into who does what.

1.git submodule add adlc-framework .adlc

2.bash scripts/governance-score.sh

3./metrics:daily-standup

⚡ Team governance visibility in 1 day

Enterprise Architect

You need to map AI agents to your org capability model. 3-tier talent bench.

1.Review .claude/agents/*.md

2./speckit.plan

3./speckit.improve

⚡ Agent-capability mapping in 1 day

Component Map

15 components implementing this pillar

Type	Name	Why	Business Value
Agent	`product-owner (opus)`	Business validation and sprint governance — ALWAYS FIRST	Prevents misaligned delivery; enforces business-led roadmap
Agent	`cloud-architect (opus)`	Technical design and deployment strategy — ALWAYS SECOND	Architecture approval before any specialist execution
Agent	`security-compliance-engineer (opus)`	SOC2, APRA CPS 234, ISO 27001 compliance enforcement	Regulatory risk eliminated at design time
Agent	`python-engineer (sonnet)`	CloudOps Runbooks PyPI package, CLI commands, boto3	Production-grade Python with TDD and battle tests
Agent	`infrastructure-engineer (sonnet)`	CDK + Terraform IaC for AWS multi-account landing zones	Reproducible, auditable infrastructure as code
Agent	`kubernetes-engineer (sonnet)`	K3s, ArgoCD, Helm chart lifecycle management	GitOps delivery pipeline for containerised workloads
Agent	`qa-engineer (sonnet)`	3-tier test strategy: snapshot / LocalStack / AWS	90-100% bug detection before production
Agent	`observability-engineer (sonnet)`	DORA metrics, daily standup, sprint ceremonies	Data-driven velocity and quality measurement
Agent	`fullstack-engineer (sonnet)`	Docusaurus docs, React components, marketplace UI	Terminal-inspired, WCAG 2.1 AA compliant interfaces
Agent	`devops-security-engineer (sonnet)`	CI/CD pipelines, supply chain hardening, SBOM	SLSA Level 2+ provenance for every release
Agent	`sre-engineer (haiku)`	Incident response, runbook execution, on-call automation	MTTR reduction through codified operational procedures
Agent	`finops-engineer (haiku)`	FinOps cost attribution, rightsizing, waste detection	Autonomous cost reduction within READONLY guardrails
Agent	`qa-automation-engineer (haiku)`	Playwright E2E, BDD step definitions, smoke tests	Regression prevention at every sprint boundary
Agent	`technical-writer (haiku)`	CLI docs sync, ADR generation, hand-curated content protection	Docs always match code — HAND_CURATED_CONTENT_DESTRUCTION prevented
Agent	`developer-experience-engineer (sonnet)`	Framework architecture, agent scoring, PDCA facilitation	Continuous improvement of the ADLC framework itself

Risk & Scalability

What happens without this pillar, and why ADLC scales from 1 person to enterprise

What if you skip?

Industry insight: no company can outsource its way to digital excellence. Companies that rely on external contractors lack business context — technologists need to gain precious understanding of the business context. Context matters to developing great digital solutions. Without an in-house bench, you are perpetually dependent.

Scalability

ADLC provides the full Digital Factory pod pattern with AI agents: 3 decision-layer agents (opus), 8 execution-layer specialists (sonnet), 4 operations-tier workers (haiku). Agent frontmatter is model-agnostic — upgrade the underlying model, agents improve automatically.