Skip to main content

AI-powered SDLC with AWS AI-DLC and Bedrock

AWS AI-DLC suits organisations that want to change the delivery operating model, not merely add a coding assistant. Its open-source, MIT-0-licensed implementation, awslabs/aidlc-workflows, is worth reading directly rather than by reputation: verified against its v2 branch, it is a near-total rewrite of the original 3-phase (Inception/Construction/Operations) release — 5 phases and 32 stages (Initialization, Ideation, Inception, Construction, Operation), driven by a deterministic TypeScript state machine rather than LLM judgment, executed by a 14-agent roster (11 domain experts, 2 quality-gate reviewers, 1 adaptive-workflows composer), with an 82-event audit trail for enterprise traceability (AI-DLC, AI-DLC for financial services, AI-DLC v2 branch).

Notably, v2 runs natively on Claude Code as one of seven supported harnesses (alongside Kiro, Codex CLI, Cursor, and others) — the shipped Claude configuration authenticates through Amazon Bedrock rather than the Anthropic API directly, and AWS's own README states it currently performs best with Claude Opus 4.8. This site's Claude and Anthropic implementation guide treats AI-DLC v2 as its structural grounding — the 5-phase/32-stage model and 14-agent roster below is the same one that document builds on.

Use this approach when the team is already in the AWS ecosystem, needs enterprise governance, or wants to combine developer agents (including Claude Code, via Bedrock) with managed runtime, knowledge, guardrails, and operational infrastructure.

Reference architecture​

LayerAWS capabilitySDLC role
Delivery methodologyAI-DLC v2 workflowsReframes work into 5 phases and 32 deterministic, auditable stages, executed by a 14-agent roster, runnable natively on Claude Code via Bedrock (AI-DLC v2 workflows).
Spec-to-code workspaceKiroTurns prompts into specs, code, documentation, and tests (Kiro documentation).
Developer assistanceAmazon Q DeveloperSupports code generation, reviews, documentation, tests, transformations, and GitHub workflows (Amazon Q Developer, Q Developer agent capabilities).
Enterprise knowledgeAmazon Bedrock Knowledge BasesBrings proprietary documentation into agent workflows through retrieval-augmented generation (Knowledge Bases).
Agent executionAmazon Bedrock AgentsDefines instructions, knowledge bases, and action groups for agents that call tools and APIs (Bedrock Agents, action groups).
Multi-agent workBedrock multi-agent collaborationUses supervisor and specialist agents for coordinated workflows (multi-agent collaboration).
Safety and policyBedrock GuardrailsApplies content, grounding, and automated reasoning controls to AI applications (Guardrails).
Agent runtimeBedrock AgentCore and Strands AgentsBuilds, deploys, and operates agents securely with framework choice and AWS integration (AgentCore, Strands Agents).

Implementation map​

SDLC patternImplementation with AWS AI-DLC and Bedrock
1. Requirements and specificationUse AI-DLC Inception or Mob Elaboration to turn intent into a unit of work. Use Kiro for specs, Amazon Q Developer for documentation support, and Bedrock Knowledge Bases for policies, domain docs, runbooks, and regulatory context.
2. Design and architectureUse AI-DLC design artifacts plus Bedrock Agents that can retrieve ADRs, architecture docs, and service catalogs from Knowledge Bases. Add Guardrails for grounding and policy checks before a design becomes an approved ADR.
3. Coding and reviewUse Amazon Q Developer in the IDE and GitHub for implementation, review, documentation, unit-test generation, and code transformation. Keep Construction or Mob Construction sessions human-led: Q proposes and explains; engineers approve and own the diff.
4. Testing and qualityUse Kiro and Amazon Q Developer to generate and explain tests. Use Bedrock Agents to map tests back to requirements, risk classes, and architecture dependencies. Use Guardrails for grounding checks where test rationales rely on retrieved policy or regulated-domain context.
5. Deployment and operationsUse Bedrock Agents action groups with Lambda or APIs for approved release checks, runbook automation, and rollback-plan analysis. For production agents, use AgentCore to operate workloads at scale and keep environment, identity, memory, and observability concerns explicit.
6. Monitoring and feedback loopsFeed incidents, postmortems, operational telemetry summaries, and support trends into Knowledge Bases or governed memory. Use Strands Agents with AgentCore memory when the feedback loop needs durable recall across sessions (AgentCore memory with Strands).

Practical build sequence​

  1. Start with AI-DLC Inception for one high-value product slice; capture the unit of work, acceptance criteria, risks, and required human approvals.
  2. Use Kiro to produce the initial spec, design notes, implementation plan, docs, and tests.
  3. Use Amazon Q Developer for implementation assistance, PR review, unit-test generation, and documentation generation.
  4. Create a Bedrock Knowledge Base for architecture docs, service catalog, standards, regulatory rules, runbooks, and incident records.
  5. Build Bedrock Agents for design impact, test-risk analysis, release readiness, and incident synthesis.
  6. Add Guardrails before any agent output is allowed to influence regulated requirements, customer-facing messaging, security controls, or production automation.
  7. Move durable agent workloads to AgentCore or Strands Agents when the workflow needs enterprise runtime, identity, memory, tool integration, and operations discipline.

Governance controls​

RiskControl
AI-DLC rituals become unbounded meetingsDefine entry criteria, exit artifacts, and decision owners for Inception, Construction, and Operation sessions.
Retrieved enterprise context is staleVersion and refresh Knowledge Base sources; mark deprecated runbooks and superseded ADRs.
Agent calls unsafe APIsPut production actions behind Bedrock Agent action groups, IAM, Lambda/API validation, and human approval gates.
Hallucinated policy interpretationUse Guardrails, grounding checks, and named policy owners for regulated decisions.
Tool outputs bypass change controlKeep all generated code, tests, docs, and runbooks inside normal PR and release-management workflows.

When this works best​

AWS AI-DLC and Bedrock fit regulated enterprises, AWS-heavy platforms, and teams willing to change rituals around AI-assisted work. The combination is strongest when SDLC agents need governed enterprise knowledge, approved tools, IAM boundaries, and production-grade operations.

It is less lightweight than a local coding-agent setup. Use it where the governance and runtime value justify the platform footprint.

Validated citations​