Today, AI is remarkably adept at migrating individual code components. However, a single prompt is insufficient for large-scale systems. AI-powered code migration only becomes scalable when analysis, prioritization, code generation, testing, quality assurance, and governance work together within a controlled process.
For companies with complex, organically grown IT landscapes, this is the decisive factor. Modernization efforts often fail due to unclear dependencies, a lack of transparency, undefined priorities, inadequate testing, and the challenge of maintaining a traceable record of changes.
AI-powered code migration can significantly accelerate modernization initiatives—provided it is viewed as more than just simple prompting and code translation.
Individual LLMs are well-suited for clearly defined tasks. However, in large-scale system landscapes, the absence of an overarching control mechanism leads to a loss of visibility regarding dependencies, priorities, testing, and approvals.
Agent systems help bridge this gap. They handle specialized tasks such as analysis, classification, code generation, test validation, error handling, and documentation.
The primary benefit lies in creating a migration process that is more predictable, verifiable, and controllable.
Regulated sectors—such as banking, insurance, pharmaceuticals, and public sector organizations—have a particular need for traceability and auditable systems. Every technical change must be explainable, verifiable, and documented.
Simple prompts work well when the task is clearly defined: understanding a module, translating a class, creating a test file, or applying a known pattern.
In a large-scale migration, however, tasks are rarely isolated. Programs access shared data structures and a wide variety of heterogeneous data sources. Business logic is distributed across scripts, workflows, data flows, and manual processes. Added to this is the fact that documentation is typically incomplete or outdated.
Consequently, code cannot simply be translated from A to B. First, one must understand:
A single LLM can assist with these tasks. However, it cannot reliably manage them, as outputs can vary from user to user given the same input.
In large-scale modernization initiatives, code translation itself is only part of the work. The greater effort lies in analysis, classification, prioritization, quality assurance, business alignment, and risk management.
This is particularly true in regulated industries. There, it is not enough for the result to be correct; the process used to achieve it must also be transparent and traceable. Companies must be able to explain what was changed, why it was changed, which risks were assessed, and who authorized the decision.
A "prompt-only" approach reaches its limits here.
Typical weaknesses include:
| Weakness | Impact on migration programs |
| Lack of orchestration | Individual migration results do not fit together seamlessly from a technical or business perspective. |
| Lack of traceability | hanges, decisions, and risks are difficult to substantiate for governance and audit purposes. |
| Lack of testing strategy | It remains unclear whether the new solution is functionally equivalent. |
| Lack of prioritization | Teams invest effort in less critical areas while key dependencies remain unaddressed. |
| Lack of scalability | Manual management becomes a bottleneck itself when dealing with thousands of objects. |
| Lack of target architecture | Without a modern, agent-compatible target architecture, the result is merely "old wine in new wineskins." |
Agent systems do not view AI merely as a tool for isolated tasks. Instead, they operate within a managed workflow where various agents handle specialized tasks throughout the migration process.
One agent might analyze source code, while another classifies risks and dependencies. A third generates target code. Additional agents create tests, verify results, document decisions, or flag cases for human approval.
Naturally, agents do not fully automate the migration process. Their advantage lies in structuring repetitive tasks, making results verifiable, and enabling more targeted involvement from human experts.
This leads to a different approach to modernization for companies.
Recent real-world examples demonstrate that LLM-assisted migrations within large software portfolios can succeed when supported by automation, testing, and human oversight.
Airbnb migrated approximately 3,500 React test files from Enzyme to React Testing Library. The manual effort involved was initially estimated at around 1.5 years; however, with the help of LLM-assisted automation, the migration was completed in six weeks.
In a research report, Google details 39 internal code migrations carried out over a twelve-month period, involving the submission of 595 code changes comprising 93,574 edits. A significant portion of these changes was generated by LLMs and subsequently verified within an automated workflow.
Slack describes its transition of more than 15,000 Enzyme test cases to React Testing Library. Here, too, success hinged not on a single prompt, but on a combination of traditional code analysis, automation, LLM support, and human oversight.
While these examples are significant, they must be viewed in the proper context. They demonstrate that LLMs can be effective for large-scale migrations when integrated into a controlled process.
The added value of AI-supported code migration lies in the ability to better manage the entire modernization initiative.
For companies, this means:
This can be the deciding factor in the success of a modernization program, especially given large and complex legacy landscapes. Without structure, AI can quickly become a tool that offers no added value within an already complex project.
The most important question is whether the prerequisites for a scalable migration are in place.
Five points should be clarified before getting started:
A sensible starting point is to conduct an assessment before beginning the actual migration. An assessment analyzes the existing application landscape, source code complexity, dependencies, and business-critical processes to establish a solid foundation for modernization and migration.
HMS supports companies in the structured planning, technical implementation, and functional validation of complex migration projects. The focus is on the entire journey—from initial analysis to the validated target environment.
Particularly in regulated industries, an approach that combines technical efficiency with traceability is essential. This entails a clear assessment of the initial situation, a robust target state, appropriate automation, quality assurance, and documented decision-making.
AI-supported code migration works particularly well when it is not viewed merely as an isolated task. While a language model can boost developer productivity, a managed process utilizing specialized agent systems can make a migration program more predictable, verifiable, and scalable.
For companies with extensive legacy landscapes, this makes all the difference.
The next logical step, therefore, is not a broad AI rollout, but rather a structured assessment aimed at establishing a clear baseline—including representative artifacts, a defined target architecture, and measurable quality criteria—upon which a structured migration can be built, true to the motto:
Better to do it right once than quickly three times.
Are you facing a migration and want to get off to a solid start?
You can find more information about our approach in our Code2X white paper.
