CJ OliveYoung Builds an AI-Collaborative Development Process and Applies AI-DLC in Practice

CJ OliveYoung held a 3-day workshop at the AWS Korea office in March 2026 and applied AI-DLC (AI-Driven Development Life Cycle) in practice. From 24 candidate projects, it selected 4 Greenfield (new development) projects and 1 Brownfield (existing system improvement) project, and about 30 members of the technology organization, including the CTO, verified an approach of collaborating with AI across the entire cycle—from requirements definition to design, implementation, and quality verification—centered on Kiro.

The core focus was turning individual productivity into a team process. In the Inception Phase, requirements analysis and workflow planning were always carried out, and reverse engineering was added for the Brownfield project. In the Construction Phase, feature design, non-functional requirements, infrastructure design, code generation, and build/testing were connected, and a quality gate was created where each stage went through a team review before moving to the next stage.

On Day 1, rules were fixed using Kiro Steering documents and requirements and development units were organized, while Day 2 focused on code generation and verification. Issues were resolved by feeding screen captures or test status back in as markdown, and defining OpenAPI early improved integration quality. In the Brownfield project, Subagents were used to separate analysis and refactoring roles.

The 5 selected projects are as follows.

• Conversational Search Enhancement: Combined Amazon Bedrock and Amazon OpenSearch Service's hybrid search with RAG, using Intent Extraction to identify purchase intent and attaching recommendation reasons and review summaries.
• Automated Partner Channel Verification: Since audit trails and Exactly-once execution guarantees were needed, multimodal Bedrock and AWS Step Functions Standard Workflow were chosen, sequentially verifying screenshots and SNS metadata and delivering AI judgment results and confidence scores to the admin back office (BO).
• Logistics Center Simulation Platform: Predicted worker placement and bottleneck sections using Monte Carlo simulation and route finding.
• Incident Response and Compensation Automation: Used the Strands Agents SDK to connect incident status summarization, affected customer classification, and compensation simulation result interpretation via tool calling, integrating all functions into a disaster recovery Agent interface while leaving the final decision to Human-in-the-loop.
• Enterprise Modernization PoC: Verified whether a legacy UI solution with difficult component reuse and open-source integration could be migrated to a modern React-based stack.

After the workshop, a demo day was held to share the results, and each team continued to further develop their projects, spreading the AI-DLC experience across the entire organization.