AMOREPACIFIC Applies AI to SAP Development with Kiro IDE

The AMOREPACIFIC ERP Platform Planning Team verified whether AI could improve development productivity even in a legacy environment based on SAP ECC 6.0 EHP7 (NetWeaver 7.40), and chose AWS Kiro IDE as the solution. The key was not simple autocompletion, but attaching Spec-Driven Development and AI-DLC, which fit SAP development culture, to the existing workflow.

The legacy SAP environment made AI adoption tricky. Development centered on SAP GUI and NWDS, source code wasn't freely open on the file system, and there was a lot of GUI-based configuration work such as SE11, SPRO, and PFCG, so general-purpose coding tools alone had significant limitations. On top of this, the organization's own ABAP naming conventions, SAPUI5 custom framework, internal Include patterns, and _new file conventions meant a mechanism was needed to make AI understand the company's own way of doing things.

The solution had three parts.

• Kiro Spec mode: For new development, specs are created first in the order requirements.md → design.md → tasks.md, then code is generated
• Kiro Vibe mode: Quickly modifying modules that already have a Spec through natural language conversation
• MCP + Steering: Delivering SAP system context and the organization's coding standards to the AI

In particular, MCP (Model Context Protocol) served as a bridge that let Kiro reference in-house context such as custom RFC functions, table structures, and business logic. Steering was a mechanism for pre-training the AI on ABAP and SAPUI5 coding standards respectively, enabling it to generate code that reflected team rules without a separate prompt each time.

The integration method was also matched to reality. Instead of a direct-write integration with ECC 6.0, they used a Shared Folder to store code generated by Kiro, and developers reflected it via SAP GUI (SE38/SE11) or NWDS. In addition, the existing Jira ticket flow matched well with Spec mode, naturally flowing into requirements.md for requirements, design.md for design, and tasks.md for task breakdown.

Ultimately, this experiment is less about replacing tools and more about a shift in the development approach. The focus was on reducing the time people spent typing code and shifting more time toward requirements verification and design, while building a development system in which AI follows the AMOREPACIFIC way across both ABAP and SAPUI5.