A New Development Process for Collaborating with AI, How Did Olive Young Start (feat. AI-DLC)

The Olive Young development organization ran a 3-day Unicorn Gym workshop applying the AWS AI-DLC (AI-Driven Development Lifecycle) methodology in March 2026, in order to find ways to embed AI into the entire SDLC rather than treating it as just an individual productivity tool.

AI-DLC structures the process from requirements analysis through design, code generation, and testing into defined stages, with clear role division between people and AI at each stage. In particular, because outputs from one stage feed into the next as inputs, the process of refining and documenting ambiguous requirements itself became the starting point for development quality.

About 30 developers, data engineers, staff engineers, and QA engineers participated in the workshop, experimenting on 5 projects. The projects consisted of 4 Greenfield and 1 Brownfield, including search engine enhancement, multimodal partner channel automated validation, logistics simulation, incident response automation, and an enterprise management system modernization PoC.

The workshop proceeded as follows:

• Day 1: Understanding the overall AI-DLC structure, with a focus on the Inception Phase
• Day 2: Construction Phase and hands-on coding
• Day 3: Build & Test and results retrospective

Early on, AI jointly analyzed requirements.md, asking questions and helping to flesh out User Stories and Application Design. This continued into Functional Design, NFR (Non-Functional Requirements) analysis, Infrastructure Design, and Code Generation, with design and implementation proceeding simultaneously.

However, teams whose requirements weren't sufficiently refined on the first day had to return to Inception at the next stage. Through this experience, participants realized that good design starts with a good understanding of requirements, reinforcing the perception of AI not as a tool for quickly generating code, but as a thinking partner for jointly working through design.

Another key outcome was the value of spec coding. By first structuring the requirements document, domain entities, business rules, and NFR documentation, and then collaborating with AI based on that spec, a solid structure emerged where code and documentation explained each other's context.

In the Brownfield project, the Reverse Engineering stage was especially effective. As AI automatically analyzed and documented the legacy codebase, initial understanding costs were reduced, showing that transitions leveraging an AI pipeline had meaningful potential for effort savings compared to manual migration.

At the same time, clear challenges emerged:

• How to accumulate and manage the large volume of design documents generated by AI
• How to determine the appropriate depth of design needed for each project
• What context to convey first, in what order, and how to compress it

As a follow-up, Olive Young established an AI Sandbox, and an internal AI Frontier Program is also in preparation. As a next step, they plan to organize a workshop that includes PMs as well as developers, expanding AI internalization across the organization with a focus on the Inception Phase.