TLDR
- AI is turning modernization from a periodic project into an ongoing process, as companies race to update legacy systems for new capabilities.
- Technical debt remains a major constraint, consuming budgets and slowing innovation as older systems become harder to maintain.
- Tools like AWS Transform aim to automate code updates at scale, but enterprises still face challenges around complexity, validation and adoption.
As companies look to deploy generative AI, they typically have to confront a longstanding challenge: Updating millions of lines of legacy code to make their systems ready for the AI era.
This technical debt is a persistent burden for companies, especially as older programming languages like COBOL fall out of favor and engineers who know them are retiring from the workforce.
But Sriram Devanathan, director of AWS Transform, sees agentic AI changing the game and ushering in an era of continuous modernization. Companies may no longer have to revisit their modernization projects every decade or so; agentic AI will be modernizing computer code continuously.
“Previously, you modernized, and then you’re done for a while. Ten years later, you’ll come back and look at it,” he said in an interview with The AI Innovator. “But with AI, there’s an opportunity to get into a continuous modernization paradigm where you’re not falling behind anymore.”
Alleviating technical debt can be costly. According to McKinsey, 70% of companies allocate between 6% and 20% of their tech budget to reducing their tech debt. Meanwhile, the U.S. government typically spends 80% of its tech budget on operations and maintenance of existing IT, including legacy systems, according to the 2025 Government Accountability Office report.
The cost is not merely financial. Legacy frameworks, aging runtimes, brittle code patterns and the like have historically slowed innovation and tied up engineering resources. In 2024, AWS previewed a tool to update legacy code more quickly using AI agents. AWS Transform became generally available last May and added a customization capability in December.
Now, the desire to deploy AI means that companies are more pressed to update their legacy systems so they can interact with generative AI tools and agents. The result is a shift in urgency: What was once a gradual modernization effort is now increasingly treated as a business imperative.
“You’re starting to see that AI adoption is itself becoming the reason why companies have to do something about their old systems,” Devanathan said. “They can’t just leave them the way they are, and every company is figuring out the path there.”
Agent-based approach to updating code
AWS Transform introduces an agent-based approach to modernization. The service combines prebuilt transformations — such as upgrading Java, Node.js and Python runtimes — with the ability to define custom transformations tailored to an organization’s codebase and standards.
By learning patterns from documentation, code samples and developer feedback, the system can apply consistent changes across large codebases, including hundreds or thousands of repositories.
This is designed to replace repetitive, manual work with automated workflows, allowing developers to focus on higher-value tasks. Customers using AWS Transform custom have reported 60% to 80% reduction in execution time for certain modernization tasks, according to Devanathan.
In one example, Thomson Reuters used AWS Transform to modernize .NET applications, cutting Windows licensing costs by 30% and increasing migration speed fourfold. The company is now modernizing about 1.5 million lines of code per month, turning modernization into a continuous, assembly-line process rather than a one-off project.
“They’ve essentially turned this into a factory kind of process, continuously modernizing their systems,” he said.
Those gains can compound over time. “You migrate 10 modules, you save 30% on those, but the next month you’re saving 30% on those 10 plus the next batch,” Devanathan said. “It becomes this cascading effect that unlocks innovation.”
Overall, customers have saved more than one million hours of manual work and analyzed 1.8 billion lines of code using Transform, according to the company. In addition, AWS said, porting applications to Linux instead of .NET also cut operating costs by 40%.
How the system works
AWS Transform is designed to fit into existing developer workflows, with both command-line and web interfaces, according to a company blog post.
Developers can define transformations using natural language, documentation or example code, and apply them locally or across repositories. The system can also integrate into continuous integration and continuous delivery pipelines for large-scale automation.
Beyond standardized upgrades, AWS Transform can capture organization-specific coding practices and apply them consistently across codebases, helping preserve institutional knowledge.
Devanathan said the system is designed to be largely self-service, allowing customers to upload their code for AI-driven modernization. If their code is spread across systems, it can be consolidated and uploaded to Amazon’s S3 cloud storage service, where Transform’s agents can process it.
For customers that prefer not to share proprietary code, AWS offers a client-side tool that runs within their own environment. The tool has access to internal code and test suites and communicates with the AI agents to carry out transformations.
Testing is a critical part of the process. The system runs unit and integration tests to validate changes and identify errors. If a test fails, the agents can adjust the code and try again, refining the output over time. Human oversight remains part of the workflow. “A human can review them and decide how to take that forward,” he said.
Transform is built on AWS Bedrock, the company’s platform for working with multiple AI models. The system continuously evaluates new models to balance cost, speed and accuracy.
