The Problem: AI's Blind Spot in Production IRIS
Modern AI code assistants like GitHub Copilot are powerful, but their effectiveness is severely limited when working with large, production-grade InterSystems IRIS applications. The core issue lies in how these AI tools access and index code. When developers work server-side using VS Code with the isfs:// virtual filesystem, common in many IRIS deployments, AI assistants can only see files that are actively open in the editor. This creates a significant blind spot. For mature IRIS applications, which can contain thousands of classes and complex interdependencies, the AI is effectively working through a keyhole, unable to understand the full context of the codebase. This drastically reduces its ability to provide accurate suggestions, refactor code effectively, or identify potential issues across the entire system.
This limitation was highlighted in discussions around VS Code productivity tools, where comments from users like Pietro Di Leo and Mike.W pointed to the inability of Copilot to index the virtual filesystem. The AI's view is restricted to the immediate, open files, making it oblivious to the vast majority of the production application's logic. This is not a problem of the AI's intelligence, but rather a fundamental limitation of its access to the data it needs to perform its job comprehensively.
Introducing iris-agentic-dev: Bridging the Context Gap
To address this critical gap, a new project, iris-agentic-dev, has emerged. Developed by Thomas Mazur, with contributions and insights from the InterSystems community, this tool aims to give AI assistants a live, comprehensive connection to entire IRIS production codebases. Instead of being limited to open files, iris-agentic-dev allows AI agents to index and understand the full spectrum of classes and code within a running IRIS instance. This fundamentally changes how AI can assist in developing and maintaining large-scale IRIS applications.
The project's goal is to overcome the limitations imposed by the isfs:// workspace. By enabling AI to access and process the entire codebase, developers can leverage their AI assistants for more sophisticated tasks, including code completion, bug detection, refactoring, and even generating new code that adheres to existing patterns and best practices across the entire application. This moves beyond simple text prediction to genuine contextual understanding.
Getting Started with iris-agentic-dev
The iris-agentic-dev project is available on GitHub, providing the necessary components to integrate AI agents with your IRIS environment. The initial setup involves configuring the tool to connect to your IRIS instance and then integrating it with your preferred AI development environment. The project focuses on providing a robust mechanism for AI to interact with the IRIS ObjectScript codebase, treating it as a connected, indexable system rather than a collection of disparate files.
Key to getting started is understanding the project's architecture. It essentially acts as a bridge, exposing the IRIS codebase in a format that AI agents can effectively parse and query. This involves setting up the agentic development environment and ensuring the connection between the AI model and the IRIS instance is stable and performant. The project aims to democratize the use of AI in IRIS development by removing the technical hurdles associated with code access and context limitations.
What This Means for IRIS Development
The implications of iris-agentic-dev are significant for the InterSystems IRIS development community. For years, developers working on large, established IRIS applications have grappled with the challenge of maintaining code quality and developer productivity when the available tooling, particularly AI assistants, could not grasp the full scope of their work. This new tool directly tackles that challenge.
Developers can now expect their AI assistants to provide more relevant and accurate suggestions, identify potential bugs that span multiple classes, and assist in refactoring efforts with a much deeper understanding of the application's architecture. This has the potential to dramatically speed up development cycles, reduce the learning curve for new team members, and improve the overall maintainability of IRIS applications. It transforms the AI from a simple autocompletion tool into a more integral part of the development workflow, capable of acting as a knowledgeable pair programmer with full context.
The series promises to delve deeper into the tool catalog, ObjectScript skills, and benchmarking. This initial part, however, lays the crucial groundwork by defining the problem and introducing the solution that promises to unlock the full potential of AI for IRIS developers working on complex, server-side applications.