As critics point out that artificial intelligence (AI) has fallen short of expectations for work efficiency in development settings, Korean companies that adopted Amazon Web Services' (AWS) AI-driven development methodology have achieved notable improvements in their operations. The approach earns praise for capturing both speed and safety by giving AI the initiative to verify tasks directly, rather than using it merely as an auxiliary tool.
AWS Korea held a press conference at its Seoul headquarters on the 16th, unveiling its AI-driven development methodology "AI-DLC" and its spec-based development tool "Kiro," while introducing actual results. Park Hye-young, a senior solutions architect (SA) at AWS Korea, noted that "94% of organizations that have adopted AI in development settings are not achieving the results they expected." According to CircleCI, a provider of continuous integration and deployment (CI/CD) solutions, only the top 5% of organizations that already had systems in place produced results when adopting AI coding tools, and while feature development speed increased 85%, final deployment speed improved by just 26%.
Increased security risk was also cited as a problem. While the proportion of vulnerability-free code written directly by humans was 75%, AI-generated code stood at around 55%. "This phenomenon occurs because current AI coding tools only shorten the coding area, which accounts for just 20% of the entire development stage," Park said. "The reality of AI functionality is that building has become faster, but fixing takes longer."
The methodology AWS has presented to solve this is AI-DLC. Whereas existing AI tools only produced results according to commands, AI-DLC is characterized by understanding the entire process and going through verification. The development process proceeds in three stages—inception, construction, and operations—with "context accumulation," in which decisions from previous stages carry forward, "approval gates," in which humans directly approve, and "audit trails," which document the entire process, working in organic connection. Through this, not only code but also requirements definitions, design documents, and operations manuals are generated simultaneously.
Clear adoption results are already being achieved in Korea as well. LG Electronics' MS Business Division doubled its productivity by adopting AI-DLC, and CJ Olive Young rapidly completed MVPs (minimum viable products) for five projects in a three-day workshop. In the case of a business-to-business (B2B) server platform project with a scale of more than 800 lambdas and containers, the sprint period was shortened from two weeks to one week, and overall productivity doubled.
AWS also unveiled "Kiro," a spec-based development tool for running this methodology in practice. Unlike the existing method in which code emerges immediately upon entering a prompt, it first organizes requirements and design into documents before writing code. Kiro is equipped with four core functions: "spec-based development," in which humans approve at each stage; "agent hooks," which automate testing and security checks; "steering," which teaches a team's development rules using just three documents; and "MCP (Model Context Protocol)," which directly links to external systems. According to AWS's own open-source benchmark results, when the same task was performed with the same model, using Kiro could sharply lower costs to between one-sixth and one-tenth of those of other companies' tools.
Cases of work results from Korean companies that earlier participated in the "AI-DLC Challenge" at "AWS Summit Seoul 2026" were also introduced. Jang Jin-woo, an associate in the Data Science Part at Hyundai Marine & Fire Insurance, explained that "in the past, software development took a lot of time going through cycles starting from requirements definition to planning, development, and verification." He added, "Using AI-DLC, the entire process can be carried out together with AI, greatly reducing the time required for projects."
Kwon Young-woo, a manager on the Mobility AX development team at LG Uplus, introduced a multi-speaker agent for vehicles and said, "Vehicle code must pass security certification standards the length of a book, so development costs were enormous during post-verification. But we handed AI a guideline from the early stages of the project and directed it to conduct continuous verification." He said, "It informed us of violations immediately at each stage, dramatically reducing the costs associated with security verification."
Yoo Hae-sik, a manager on the AI Architect Team at SK AX, assessed that "the most difficult thing in IT projects is maintaining the consistency of the deliverables created by multiple role players." He added, "Because AI-DLC verifies and guides in a consistent flow from requirements design to final deployment, we achieved a significant effect in terms of quality."
