Why a clear code architecture is fundamental in the age of AI: our expert’s insight on legacy systems

By Robert van Uden, Practice Lead & Software Architect at ALTEN  

At ALTEN, we know that generative AI has changed the way legacy system modernisation projects are approached. Robert van Uden, Practice Lead and Software Architect at ALTEN Nederland, has developed the tooling and methodologies that help teams successfully adopt AI. Here is his take on the future of code architecture at the age of AI.

The problem: legacy systems were facing issues before AI

When the code needs to be changed, here is where the vicious cycle begins.

Refactoring the existing code or adding more code to fix the problem? In most cases, the preferred solution is to add more code and postpone the refactoring, making the system larger and increasingly more difficult to understand. And as it grows more and more complex, even an experienced software engineer struggles to understand it. This leads to a vicious cycle.

A strong architecture breaks this cycle. By decomposing the system into clear, well-defined components, engineers can focus on local changes without needing to grasp the entire codebase. This not only contains complexity but also preserves design intent, ensuring knowledge persists beyond individual team members.

Architecture as a solution to the AI trap

When we let AI solve a code issue without supervision, AI will figure out a solution by creating extra code, creating a bigger problem. However, a clear architecture decomposed into components will make sure that AI will not increase the complexity.

Here is where AI brings unique value in the legacy modernisation context: it can actively assist in the process of decomposing itself by:

• Analysing existing codebases
• Identifying clusters of related functionalities
• Visualising the architecture

How to break the vicious cycle of legacy systems?

AI and humans now play distinct but complementary roles.

AI thrives in navigating complexity by analysing unfamiliar code, identifying patterns, generating solutions, and spotting potential issues across vast codebases. 

Humans own the strategic decisions by shaping the system’s architecture, defining component roles, and ensuring AI-generated solutions align with long-term design principles.

Specification skills, architectural thinking, and solid verification and validation practices are still “as essential as ever”.

However, new competencies need to be developed, such as:

• Formulating precise prompts
• Validating critically AI-generated code 
• Defining architectural guardrails that keep automated generation aligned with design intent

If you want to know more about legacy system modernisation, read Robert’s full article on the Bits & Chips website.