Artificial intelligence is quickly becoming a topic of conversation in every IT organization. In the IBM i ecosystem, the discussion has largely focused on coding assistants powered by generative AI models, tools designed to help developers write more consistent RPG applications and mitigate the impact of the declining pool of programming talent.

This is an important development. But it's only part of the story.

By Marc Dallas, R&D Director, ARCAD Software

At ARCAD, we believe AI can play a much broader and more strategic role in modernization initiatives, not only at the code level, but also at the application level. And today, with the launch of the ARCAD MCP (Model Context Protocol) Server, we are introducing the architectural foundation that makes AI truly usable, reliable, and controlled in IBM i environments.

Because in legacy systems, especially business-critical IBM i systems, AI cannot be "improvised".

Several large language models (LLMs) are already capable of understanding and generating reasonably accurate RPG and COBOL code. This opens up exciting possibilities:

• Helping junior developers learn RPG or COBOL
• Explaining legacy programs to speed up onboarding
• Suggesting improvements in syntax or structure
• Converting logic from one language to another

AI can help transfer knowledge, especially in organizations where experienced IBM i developers are approaching retirement and institutional memory is at risk of being lost.

But to leverage AI for modernization, you need to think on two distinct levels:

1. Code-level modernization
2. Application-level modernization

At the code level, AI can:

• Generate new code
• Explain existing programs
• Provide contextual advice on syntax
• Help maintain internal programming standards

With retrieval-augmented generation (RAG), it is possible to inject company-specific source code into a model so that it can analyze and reflect the organization's programming style and conventions. In theory, this allows AI to produce code that complies with internal standards.

This type of approach can work in a virtuous circle: AI helps explain legacy programs, developers gain clarity, and new code becomes more consistent and better documented.

But there is one significant limitation. Code is only part of an application.

A program may be clear at the line level, but remain opaque when viewed as part of a larger system. Developers need to understand the architecture, dependencies, services, and data flows, especially in IBM i environments where applications have evolved over decades.

Thanks to the stability of the IBM i architecture and its technology-independent machine interface, programs written many years ago still work reliably today. This durability is an asset. But it also means that modernization projects must manage layers of accumulated logic, integrations, and business rules. And this is where AI must go beyond code.

In addition, writing relevant source code sometimes requires the use of external elements that interact with that code (such as the structure of the files used or dependent elements). It is therefore necessary to understand the overall context of the application for greater efficiency.

An application is not just its source code. It includes:

• The architectural structure
• Dependencies between programs
• External services
• Data definitions
• Business rules
• Metadata

Modernization requires a precise understanding of how all these elements interact.

Within the ARCAD ecosystem, the metadata repository and DISCOVER already provide what we call application intelligence, which is a structured representation of the system's architecture and dependencies.

DISCOVER, enhanced with AI capabilities, allows users to retrieve architectural specifications and dependency information using natural language queries. This is particularly important for business analysts who need architectural visibility but may not have the deep technical expertise required to manually analyze complex IBM i environments.

Instead of relying on spreadsheets, fragmented documents, or institutional memory, organizations benefit from a consolidated, structured view of their application landscape.

However, introducing AI into this equation raises a crucial question:

How can we ensure that AI operates in a specific, controlled, and secure context in IBM i environments?

Without a structured and reliable context, LLMs can:

• Misinterpret complex RPG dependencies
• Ignore DDS structures
• Miss business rules built up over decades of development
• Produce approximate or inconsistent responses

In business-critical systems, approximation is not acceptable. Without an architectural layer to structure and govern context, AI risks generating responses that are incomplete, misleading, or inconsistent with internal processes.

This is precisely why we are launching the ARCAD MCP (Model Context Protocol) Server as a strategic component of the infrastructure.

It is the nerve center for control, information, and validation between:

• IBM i applications
• The ARCAD suite
• AI agents (including Bob)
• LLMs

Its objective is simple but fundamental: to structure and ensure the reliability of the context provided to AI models.

1. Providing a stable reference

The main role of the MCP server is to provide a deterministic, comprehensive, and reliable reference for interactions between system elements.

It:

• Exposes application metadata
• Structures dependencies
• Injects accurate contextual information into AI queries
• Performs mature and proven sub-processes

Without this layer, AI models attempt to deduce meaning from incomplete inputs. With MCP, AI operates on structured and validated context.

This significantly reduces the risk of hallucination or approximation. The model no longer simply "guesses" based on partial clues, but leverages organized and controlled application intelligence.

In modernization scenarios, this means:

• Faster architectural analysis
• More consistent recommendations
• Better alignment with actual system dependencies

AI becomes a controlled accelerator rather than an unpredictable assistant.

2. Governance and security: integrating AI into DevSecOps

Introducing AI into critical IBM i environments raises legitimate concerns:

• Is source code exposed to external services?
• Are access rights being respected?
• Are DevOps processes being followed?
• Is context shared securely?

The MCP server addresses these concerns by providing:

• An access control layer
• Secure management of shared context
• Integration with existing DevSecOps workflows

AI does not bypass processes. It integrates with them.

This distinction is critical for IT leaders responsible for governance and risk management. Adopting AI should not compromise compliance, traceability, or security standards.

With MCP, AI is an integral part of the controlled application lifecycle, not an external shortcut.

3. Enabling AI across the ARCAD suite

The MCP server is not an isolated component. It strengthens the entire ARCAD ecosystem.

It enhances:

• DISCOVER — with AI-augmented application analysis
• ARCAD CodeChecker — with contextualized recommendations
• ARCAD Transformer RPG — with intelligent assistance during modernization

By centralizing context orchestration, the MCP server becomes the hub of industrialized AI in the modernization lifecycle.

This architectural approach ensures consistency between tools and prevents fragmentation of AI initiatives.

This approach aligns with the reality of IBM i environments: sustainable, business-critical systems that require rigorous modernization strategies.

AI can accelerate modernization projects.
It can help developers.
It can help fill knowledge gaps.

But only if it is based on structured application intelligence and a governed architecture.

The ARCAD MCP server provides that structure. By consolidating context, applying governance, and integrating AI into the DevSecOps lifecycle, it transforms AI from an experimental tool into a strategic catalyst for modernization.

For IT managers and developers navigating the complexity of legacy system transformation, this marks an important milestone:

AI is no longer just generating code.
It now operates within a governed architectural framework.

And that's what makes it usable in real-world modernization projects.

__________________________________________________________________________________

About the Author: Marc Dallas holds a Software Engineering degree from the Integral International Institute, Marc started his career in 1994 as Analyst Programmer at Nestle Cereal Partners, and was appointed Product Manager at ADSM Software, prior to joining ARCAD Software in 1997.