Customizing AWS AI-DLC for System Maintenance Environments

LG CNS introduced AWS's AI-DLC (AI-Driven Development Lifecycle) to a System Maintenance project for a large enterprise client, customizing the default workflow—which is centered on new development—to suit an operations and maintenance environment.

Generative AI development methods are evolving from prompt engineering to context engineering, and further to harness engineering, which controls AI tasks through workflows and quality gates. AI-DLC structures the process from requirements analysis to design, code generation, and validation through stage-by-stage approval gates and human review.

Applying the default AI-DLC to operational systems resulted in the following issues:

• Repeatedly asking whether the system was Greenfield/Brownfield and requesting Infrastructure Design, even when dealing with existing systems
• Re-validating extended security rules every time, despite an already established security framework
• The potential for AI to make unnecessary assumptions or expand the scope of work in operations and maintenance tasks

LG CNS added rules, knowledge, and validation frameworks tailored to the System Maintenance environment by utilizing an Override structure, without modifying the original rule files distributed by AWS. This fixed the target operational system as the default and managed recurring lessons and on-site standards as reusable assets.

For execution, LG CNS leveraged Kiro, AWS's agentic AI IDE. By combining Steering rules, Hooks, Skills, and MCP, they applied rules to each session, repeatedly injected core guidelines, and automated artifact publishing and document conversion. Integration with internal systems such as Jira, Confluence, and Bitbucket was also configured via MCP.

The criteria supporting enterprise adoption were token usage optimization, security governance, and cost. Kiro allocated models based on task nature using LLM Routing, applied SSO login, S3 prompt logging, and disabled web features, and presented a cost of approximately $20 per developer per month based on actual operational standards.

The background enabling rapid adoption included prior application modernization, a general-purpose technology stack that is easy for AI to read, and security governance already secured through the introduction of Amazon Q Developer. Additionally, AWS AI-DLC v2 provides 5 phases and 32 stages, 11 domain expert agents, and a harness-neutral single core, expanding these customization elements into official features.