AWS has identified a recurring problem among banking, insurance and airline customers: mainframe batch code that has run for decades is often understood by only one maintainer who is approaching retirement. AWS aims to rewrite Assembler into Java Spring Boot using Amazon Bedrock, Kiro IDE and AWS Transform, but the pace of migration still depends on whether business experts can verify the original rules.

What this is

This is a modernization methodology for legacy mainframe systems: AI first extracts business rules from Assembler, experts then review them through COBOL, a more readable intermediate representation, before the code is converted into Java services running on AWS. COBOL, a mature enterprise programming language, is not the end goal; it is a validation layer that reduces the risks of rewriting the code directly.

Amazon Bedrock provides callable large language models, Kiro IDE assists with code analysis and modification, and AWS Transform handles the COBOL-to-Java conversion. AWS also requires access permissions and test data to be configured, while business and technical experts participate in functional-equivalence testing.

Industry view

The method's greatest value lies not in “AI writing code automatically” by itself, but in decomposing code migration into inspectable steps. Converting to COBOL before moving directly from Assembler to Java adds a stage, but it makes it easier for experts familiar with the legacy system to identify whether business logic has been lost.

The risks are equally clear: code can be rewritten without its rules being correctly understood. Mainframe systems often contain special handling, regulatory requirements and implicit rules accumulated over years. If documentation is inadequate and test samples are incomplete, AI-generated Java services may appear to work while producing errors in settlement, risk control or claims processing. AWS itself identifies domain knowledge and expert validation as necessary conditions, indicating that enterprise migration is currently better suited to assisted engineering than automated delivery without human review.

Impact on regular people

For enterprise IT: AI is entering costly work such as code migration, but budgets cannot be based on generation speed alone. Organizations must also pay for architecture mapping, permission governance, regression testing and expert review.

For individual careers: People who understand industry rules and can review AI output will become more valuable within enterprises. Roles measured solely by lines of code will come under pressure as responsibilities shift toward defining problems and validating results.

For consumers: Banking, insurance and airline systems may be updated faster, but what consumers actually notice will remain service stability and processing accuracy, rather than the underlying switch from Assembler to Java.