Skip to main content

Overview of AI-native Development

What Is AI-native Development?​

When you hand a leave request specification to an AI agent, it generates a business screen that runs on intra-mart Accel Platform (iAP). Instead of writing out code line by line, you hand the specification of what you want to build to an AI agent and let it assemble the artifact itself. The AI-native development covered by this documentation refers to the style of developing iAP user modules with this approach. Accel Orbit is the collective name for the suite of features (tools and assets) that support this development style. For its components, see What Is Accel Orbit?.

The generation is handled by a coding agent. This documentation supports three of them: GitHub Copilot, Claude Code, and OpenAI Codex. Developers work while interacting with an AI agent on Visual Studio Code (VSCode), and the development environment itself is also built on a local PC using VSCode and Accel CLI.

The targets are assets such as business screens that actually run on iAP. Rather than leaving everything entirely to the AI agent, the developer defines the specification and has the artifact generated in line with that specification.

The Flow from Generation to Verification​

Issuing a single instruction does not necessarily return a perfect finished product. Only after the artifact generated by the AI agent is deployed to an iAP verification environment can you tell whether it works as expected.

AI-native development proceeds through the following iteration.

  1. Settle on the specification of the artifact you want to build
  2. State the specification explicitly and instruct the AI agent to generate it according to the skill set
  3. The AI agent generates the artifact and saves it under the project's src/ directory
  4. Deploy the artifact to the verification environment and verify its behavior
  5. If there are defects, ask the AI agent to fix them

Repeat the checking and fixing in steps 3–5 until the artifact works as specified.

Once a working artifact is complete, you package it as a user module and register it in an IM-Juggling project.

The actual operating steps for this flow are covered in the Tutorial.

The AI-native Development Support Feature That Supports Iteration​

To speed up the generation of artifacts that work as specified, it is important to run the checking and fixing work as quickly as possible. This is where the AI-native Development Support Feature comes into its own. Adding it to the iAP development environment lets you improve the productivity of AI-native development using coding agents.

Its main benefits are as follows.

  • Hot deployment: You can instantly apply the assets under development without restarting the application server or redeploying the war file.
  • Deploy via Accel CLI: You can deploy locally created assets directly to the development environment.
  • Use of MCP (Model Context Protocol): The AI agent can retrieve configurations, logs, and annotations on the development server, using them to obtain the information needed for generation and to collect information about errors that occur. The collected information is used to improve the applications the AI agent generates.
  • Development environment management: You can easily perform tasks such as creating, editing, and deleting stagings and deployments, switching stagings, executing tenant environment setup, checking routing, and checking logs.

In particular, because MCP can feed information about errors that occur when the artifact runs back to the AI agent, you do not need to paste exception logs into your prompts, which makes the checking-and-fixing iteration more efficient. The installation steps and usage of the AI-native Development Support Feature are covered in the AI-native Development Support Feature category.

How to Proceed​

Before starting AI-native development, you need to set up your development environment. The following options are available for setting up the development environment.

Once you have set up the development environment by following the documents above, you can experience the flow of actually generating assets in the Tutorial.

For the knowledge required for each task, see Prerequisites. The Glossary also organizes the terms that appear in the text, so refer to it as well.