Skip to main content

Hybrid AI-powered SDLC implementation

The default hybrid model on this site pairs Claude Code as the primary agentic harness with AWS AI-DLC v2 as the lifecycle discipline and Amazon Bedrock as the governed enterprise runtime. AI-DLC v2 already runs Claude Code natively as one of its supported harnesses (AI-DLC v2 branch), so this isn't a forced combination of unrelated ecosystems — it's the vendor's own documented integration path, plus this site's own opinion about where Claude Code should lead.

Other ecosystems earn a place only where they cover a real gap: Codex and the OpenAI Agents SDK for teams that specifically need AGENTS.md-driven CI/CD automation and OpenAI's tracing/trace-grading tooling (Codex and OpenAI implementation) — treat that as a scoped exception, not a co-equal default.

The point is to give each ecosystem a defined job and a clear governance boundary, not to deploy every available tool.

Tool choice by pattern​

SDLC patternBest primary choiceBest supporting choices
1. Requirements and specificationAI-DLC v2 Ideation/Inception stages (intent-capture, requirements-analysis, user-stories) run through Claude CodeClaude Cowork for product/compliance stakeholders who review the enriched spec without a terminal; Bedrock Knowledge Bases for regulated-domain context.
2. Design and architectureClaude Code subagents (or AI-DLC's architect, architecture-reviewer, design agents) for architecture explorationClaude Design for prompt-to-prototype artifacts handed off to Claude Code; Bedrock Knowledge Bases for enterprise policies and historical ADRs.
3. Coding and reviewClaude Code for implementation, skills, and background subagent reviewCodex GitHub Action where a team specifically needs CI-native repo automation outside Claude's own tooling.
4. Testing and qualityClaude Code test subagents and skills, mapped to AI-DLC's build-and-test and ci-pipeline stagesAmazon Q Developer and Kiro for AWS-native generated tests tied to specs.
5. Deployment and operationsAI-DLC's deployment-pipeline/deployment-execution stages via Bedrock Agents, Guardrails, and AgentCoreClaude Code self-hosted runners for internal-network release checks; Claude Cowork for release-note and runbook review by non-developers.
6. Monitoring and feedback loopsAI-DLC's observability-setup/feedback-optimization stages plus Claude Code memory and skills for team-level lessonsBedrock Knowledge Bases and AgentCore memory for governed enterprise recall; Claude Cowork for on-call/support triage.

Shared artifact model​

Hybrid only works if every tool reads the same operating reality. Do not create parallel, unreviewed memories.

ArtifactPurposeUsed by
CLAUDE.mdClaude-specific project behavior, conventions, role guidance, recurring correctionsClaude Code, Claude subagents, AI-DLC's Claude harness
AI-DLC state (aidlc-state.md)Deterministic phase/stage progress, versioned workflow stateAI-DLC's engine, all 14 roster agents
docs/adr/Architecture decisions and trade-offsAll ecosystems
docs/specs/ or issue templatesRequirements, acceptance criteria, and rollback criteriaAll ecosystems
docs/runbooks/Deployment, incident, and rollback instructionsAll ecosystems
Skills and plugins directoriesRepeatable procedures and scriptsClaude Code skills/plugins
Bedrock Knowledge Base sourcesEnterprise policy, regulated-domain context, service docs, incident historyBedrock Agents, AI-DLC's aws-platform and compliance agents
AGENTS.md (scoped exception)Repository rules for the Codex-covered slice of CI/CD, where in useCodex, OpenAI workflows

Hybrid operating model​

  1. Start every change with one source of intent. The issue or spec is the canonical input. Claude Code, AI-DLC's roster, Q Developer, Kiro, and Bedrock agents may enrich it, but accepted changes go back to the same artifact.
  2. Let Claude Code carry the default agentic work. Use Claude Code and its subagents for exploratory reasoning, implementation, review, and background work across all six SDLC patterns, whether or not AI-DLC's phase discipline is layered on top.
  3. Use AI-DLC v2 for lifecycle structure where the team wants it. Its deterministic engine, 32-stage model, and 82-event audit trail give a defined, auditable path through requirements to operations — adopt it when the team wants that discipline, not as a mandatory layer.
  4. Use AWS for enterprise execution boundaries. Put regulated knowledge, production automation, guardrails, IAM, action groups, and long-running agent operations in Bedrock and AgentCore where enterprise controls are strongest.
  5. Extend past the terminal deliberately. Route non-developer participation (product review, design handoff, on-call triage) through Claude Design and Claude Cowork rather than asking every stakeholder to learn a CLI.
  6. Reach for Codex only for its specific strength. If a team needs OpenAI's tracing/trace-grading or AGENTS.md-driven CI automation for reasons Claude Code's own tooling doesn't cover, scope it narrowly rather than running two parallel default harnesses.
  7. Promote lessons deliberately. A lesson starts as a PR comment, incident note, or rejected suggestion. It becomes durable memory only after a human owner promotes it into CLAUDE.md, an ADR, a runbook, a skill, or a Knowledge Base source.

Governance matrix​

Decision or actionDefault ownerRequired controls
Enriched requirementProduct owner and engineering leadLogged accepted/rejected AI suggestions; acceptance criteria mapped to tests.
Architecture decisionEngineering lead or architecture boardADR, blast-radius review, policy check, test impact, rollback criteria.
Code changeDeveloper and reviewerPR review, deterministic tests, background-subagent summary for high-risk changes.
Generated test suiteDeveloper and testerRequirement linkage, edge-case coverage, failure-mode relevance, no shallow coverage padding.
Release decisionRelease owner or on-call engineerCanary evidence, rollback plan, production health, human approval for risky automation.
Non-developer agentic action (Cowork)Task ownerSame write/send approval gates as a Claude Code session — Cowork shares Claude Code's permission model.
Durable memoryNamed artifact ownerProvenance, expiry, supersession, source link, and review date.

Minimal hybrid stack​

Start small before adding orchestration:

  1. Root CLAUDE.md for Claude Code and subagent behavior.
  2. Two shared skills: requirements-enrichment and test-generation.
  3. Claude Code hooks for local lint, format, secret scan, and test checks.
  4. If lifecycle discipline is wanted: install AI-DLC v2's Claude harness (/aidlc) for the 14-agent roster and 32-stage model, authenticated through Bedrock.
  5. Bedrock Knowledge Base for architecture, policy, and incident context.
  6. Bedrock Guardrails and action groups before production automation.
  7. Claude Cowork for the parts of the team that need agentic help but never open a terminal.
  8. Codex GitHub Action, only if a team specifically needs its CI-native automation on top of the above.

What not to do​

  • Do not let each tool maintain its own unreviewed memory.
  • Do not allow production action groups, shell commands, or deploy scripts without explicit ownership and approval.
  • Do not measure success by generated lines of code. Measure lead time, change failure rate, MTTR, test value, review latency, and rollback quality.
  • Do not run two parallel "default" harnesses. Pick Claude Code as the default and scope anything else to what it specifically covers.
  • Do not treat vendor citations as proof of effectiveness. Validate with your own DORA metrics and incident outcomes.
  • Do not treat Claude Design or Claude Cowork as settled, stable products — both are Anthropic "Labs"/preview offerings as of mid-2026; verify current rollout status before committing a team workflow to them.

Validated citations​