Modernising legacy software with AI support

In older systems, documentation, tests and current system knowledge are often incomplete. AI can support analysis, the creation of a test baseline and clearly bounded migration work.

Three practical uses of AI

Understanding existing code

Language models can explain unfamiliar code, summarise modules and indicate possible data flows. Their output is a working hypothesis and must be checked against the actual code and system behaviour.

Building a test baseline

Before making changes, tests should capture the current behaviour of the affected areas. AI tools can draft initial characterisation tests, suggest edge cases and prepare suitable test data.

The team must verify that the generated tests represent the observed and functionally relevant behaviour correctly. The actual change should begin only once this foundation is reliable.

Repetitive migration work

Framework upgrades, API changes and syntax migrations often contain many similar, clearly describable steps. These tasks can be divided into small, reviewable units.

Coding agents can accelerate this work when the task is clearly bounded and automated tests verify the result. Architecture changes or functionally ambiguous modifications still require a separate technical decision.

Where technical decisions remain with the team

Architecture decisions require full context. Business objectives, existing systems, team capabilities and operational requirements all shape the decision. AI can prepare options; responsibility remains with the team.

A complete rewrite carries its own risks. Older systems often contain business rules that are neither documented nor covered by tests. A full replacement can lose these rules without anyone noticing.

The new code must remain understandable to the team. Modernisation is sustainable only when the responsible people can review, explain and maintain the changes over time.

Three prerequisites for getting started

A target state with clear stages. Define which areas will change first, which will remain for now and which criteria determine when a stage is complete. A reviewed test baseline. Characterisation tests capture the current behaviour of the areas being changed. AI can assist with drafting; the team verifies their meaning and coverage. Small, reversible changes. Each change should be independently reviewed, integrated and reverted if necessary. AI support can accelerate the work, but does not justify larger change sets.

A controlled sequence

Start by mapping modules, dependencies and risk areas. Then build the test baseline for the first stage and use coding agents to support clearly bounded, repetitive changes.

Structural changes follow separately and require a deliberate architecture decision and technical review.

Assess and document the result after each stage, then define the next unit of work. This keeps the initiative interruptible and limits the scope of each subsequent decision.

Conclusion

AI can accelerate analysis, test development and repetitive migration work. Sustainable modernisation depends on reviewed results, deliberate architecture decisions and small, reversible changes.