AWS deploys AI agent to demolish tech debt and automate code modernization at scale

Amazon wants to demolish your tech debt like it did this AWS server - here's the plan

Transform modernizes Windows stacks and slashes costly licensing. Well... that happened. PR folks the world over can take a lesson from how Amazon announced the enhancement of its AWS Transform service. They hauled an old AWS server 150 feet up on a crane in the middle of Las Vegas, and then dropped it on a pile of explosives. This, ladies and gentlemen, is how you get the attention of tech journalists. I mean, dropping a server 150 feet and blowing it up is a happy place I didn't even know I had. Also: Amazon says new DevOps agents need no babysitting - you can try them here The AWS team pulled off this outrageous PR stunt to demonstrate that its new AWS Transform service has integrated agentic AI into its legacy modernization system. Let's take a step back for a moment and discuss the pain point that AWS Transform is designed to solve. It all revolves around tech debt. We're all familiar with financial debt. That's an amount of cash you borrowed that you're required to pay back, usually over time. Although the analogy is a bit rough, tech debt is the payback (usually in terms of fixing and re-engineering) that an organization has to do because of the result of an earlier decision or implementation action. Also: Not a developer? AI could still take your job, MIT study finds In terms of coding, tech debt is basically all that old code (which uses out-of-date foundations, languages, APIs, and more) that has to be rewritten to be modern, maintainable, and less costly. In 1992, Ward Cunningham, creator of the first wiki, coined the term tech debt, saying: Shipping first-time code is like going into debt. A little debt speeds development so long as it is paid back promptly with refactoring. The danger occurs when the debt is not repaid. Throughout my many years of writing and marketing software products, I've encountered two instances of tech debt so substantial that I decided to sell the products rather than spend additional months or years rewriting them to bring them back up to their original level of functionality. One of those products was one of the very first embedded database engines for what was then called multimedia. In order to work in the platform it was built for, it needed to be structured around that platform's API framework. This wasn't done capriciously. When the company pitched me on using its framework, it promised me it would remain compatible for a minimum of five years. That was a lie. Also: How AI can magnify your tech debt - and 4 ways to avoid that trap So I spent about a year coding to that framework. Eighteen months after their promise, the company introduced a whole new, better-than-ever framework that was, incidentally, completely incompatible with all the previous work. If I wanted my database engine to keep working on its platform, I would have had to do all that work all over again. One of the attributes of tech debt is that it's work you do just to keep up, not work you do to add new features, services, or innovations. Fortunately, I found a buyer for that database engine (and yes, I fully disclosed the tech debt), and I moved on to new things. More recently, for about a decade, I produced an open-source donation management system that, during the course of its life, channeled about $40 million in donations to non-profits. It was open source, so while it was good work, my personal take was about enough for a car payment each month. The gotcha was that I had to support payment gateways (like Stripe and PayPal). These services were constantly updating their interfaces, both to add more capabilities and to combat fraud. But it meant that for every year of programming I put in, about 9 months were spent just updating the payment gateway interfaces. Also: You should still learn to code, says top Google AI exec - here's why These payment gateway update emergencies always came with the threat that my non-profit users would be cut off if I didn't do the work in time. This constant cycle of urgency left me with precious little time for actual innovation and product improvement. I sold off that product last year, primarily because I just couldn't face rewriting it yet again for yet another change in PayPal's interface code. Those are two examples of tech debt for a lone programmer. Now, scale that up to the enterprise level. Add in commitments to increasingly expensive licenses, reliance on hardware or applications that are out of date, and you have a massive problem. Many of our challenges stem from aging or unsupported technologies, systems that don't talk to each other, and unreliable data. In several areas, we're missing essential tools, still relying on manual steps that should be automated, or operating without adequate training and documentation. Software also often lacks the seamless, multi-channel capabilities customers now expect. According to Amazon, a typical organization "spends 30% of its teams' time on manual modernization work, otherwise known as tech debt." According to Accenture's 2024 Digital Core report, "tech debt costs $2.41 trillion a year and would require $1.52 trillion to fix." Also: AI agents see explosive growth on AWS Marketplace - over 40x the team's initial targets Speaking from experience, this sucks. Beyond all the expense involved, it's soul-destroying to spend a huge percentage of your time simply trying to claw your way back to where you were, just to keep your systems maintained. It results in lost revenue, diminishment of brand value, huge opportunity cost, possible security issues, and life-sucking rework that leads to burnout. For me, it was the incentive to divest myself of two products I had spent years refining and crafting. It was heartbreaking, but freeing. When it was introduced earlier in the year, AWS Transform was billed as a service for modernizing applications and legacy systems. At the time, while AI-based, it was more procedural in nature. The announcement this week ups that game considerably by adding agentic capabilities, which gives businesses the ability to do large-scale modernization and transformation projects at a much faster pace than had been previously possible. The big headline capability is one Microsoft is sure to be uncomfortable with. Transform provides agentic AI tools for customers to "modernize their complete Windows environments to reduce expensive licensing costs and improve security and performance." By "modernization," Amazon means "moving off of Windows" to open source solutions. Once the AI agent has completely analyzed the Windows server stack, it proposes a plan to update .NET applications and UI frameworks. It creates a plan to move from Microsoft's proprietary (and costly) SQL Server to PostgreSQL and other environments that don't involve sending license fees to Microsoft. Also: Microsoft is packing more AI into Windows, ready or not - here's what's new Amazon makes a fairly bold claim here, contending that it can reduce operating costs by up to 70%. To backup that claim, the company points to two users of Transform: Teamfront and Thomson Reuters. Teamfront is a vertical SaaS consolidator that centralizes operations and growth support for its partner companies. According to Bobby Land, Teamfront's chief product and technology officer, the company modernized 800,000 lines of code in two weeks. That's pretty astounding. Although mostly unproven, a metric in the industry (pre-generative AI) suggests that the typical programmer writes 10-25 lines of code per day. Therefore, 800,000 lines of code would be 32,000 person-days or roughly 87 person-years of code updating. Land says, "This breakthrough showed us a clear path to retiring technical debt and gave us the confidence to expand our modernization efforts. We're now moving from SQL Server to PostgreSQL while simultaneously transforming our applications, accelerating our modernization journey, and enabling us to better serve our portfolio of field service software companies." Also: I found a powerful Microsoft Office alternative that doesn't push AI - and it's free Then there's Thomson Reuters, a global information-services provider that delivers trusted data and workflow software. They used AWS Transform to "move from Windows to open source alternatives to achieve better performance and lower costs." According to Amazon, "Using agentic AI-powered automation, they now boost velocity by migrating 1.5 million lines of code per month, achieving 30% lower costs, and reducing technical debt by 50%." 1.5 million lines of code per month. It's a number that's almost impossible to wrap your head around, but apparently, the AI agent can do it. With this announcement, Amazon is showcasing three additional enhancements, designed to cater to various usage models. Mainframe modernization: New AWS Transform agents build on existing capabilities to produce activity analysis, blueprints for "reimagining legacy code into clear business functions," and task agents to speed up automated test planning and validation. VMware migration: According to Amazon, "New capabilities in AWS Transform for VMware simplify and accelerate large-scale discovery, planning, and network migration." Agents orchestrate the entire process, utilizing an on-premises discovery tool and enhanced network migration agents. Also: Mistral's latest open-source release bets on smaller models over large ones - here's why AWS Transform composability initiative: This "empowers AWS Partners, such as Accenture, Capgemini, and Pegasystems," to integrate their proprietary tools, agents, and knowledge bases directly into the AWS Transform product experience to build customized modernization workflows for customers, particularly in specialized industries like financial services and healthcare. In other words, for truly challenging verticals, partners who know their way around those solutions can build out their own custom capabilities. Amazon claims that AWS Transform can "Achieve transformation up to 5x faster than when done manually." I'm always wary of such claims, but my own experience using AI agents to assist in coding has shown that such mind-blowing productivity benefits are possible. Amazon offers two more customers as proof points: Also: I got 4 years of product development done in 4 days for $200, and I'm still stunned Air Canada used AWS Transform to "coordinate and execute the modernization across thousands of Lambda functions (i.e., small tasks in response to events or triggers)." The airline achieved an 80% reduction in time and cost compared to doing the migration by hand. QAD is a software company that provides cloud-based supply chain solutions for manufacturers. According to Sanjay Brahmawar, chief executive officer, "Modernizations that used to take two weeks now take just three days, driving 60%-70% productivity gains and saving more than 7,500 developer hours a year. We've already processed more than 180,000 lines of legacy code with exceptional accuracy -- and the agent improves with each project." Migrations and modernizations are mission-critical necessities for all technology companies, as well as for most companies that rely on technology for any part of their business. But the problem is that this type of work is essentially drudge work. It doesn't move the needle forward, enabling businesses to do more or offer more. All it does is attempt to prevent entropy. A typical employee works about 2,000 hours per year. In the highly unlikely scenario where all those hours are devoted to coding, that 7,500 hours saved would translate to four coders' jobs. I am concerned that all of these productivity savings will simply mean fewer jobs for my coding colleagues. Also: I tested Opus 4.5 to see if it's really 'the best in the world' at coding - and things got weird fast But grunt work, like updates and migrations, does not grow businesses. If a tool like AWS Transform can reduce the cost of the mechanical maintenance work and free up programming talent to innovate, that has to be a boon for programmers and IT personnel. Of course, it always comes down to management decisions. If companies use AWS Transform simply to cut jobs without an eye to new opportunities and growth, it's not only bad for coders, it's also bad for the companies themselves. However, if companies utilize AWS Transform to reduce costs, thereby channeling their talent into focusing on greater productivity and innovation, that's a very good thing. Most likely, across the entire spectrum of users, we'll find companies that choose one and others that take the second path. Regardless, reducing technical debt and saving us all from having to spend months on even one more treading-water upgrade or migration is a relief. Also: Google's Antigravity puts coding productivity before AI hype - and the result is astonishing What about you? Have you encountered tech debt that has drained time, money, or morale? Have you tried using AI tools to modernize old systems, or are you considering something like AWS Transform? Do Amazon's claims about code migration speed and cost reduction seem realistic to you, or do they raise concerns about accuracy or job impact? If you've tackled a major migration, how did it go? What would you do differently next time? Share your thoughts in the comments below.

AWS wants to take the strain out of modernizing all your old code - and ending tech debt quicker than ever before

AWS says upgrades can lead to significant cost and time savings Tackling your organization's tech debt could soon be a lot less painful thanks to a new upgrade from Amazon Web Services (AWS). At its AWS re:Invent 2025 event, the company revealed new agentic AI capabilities for AWS Transform which it says will allow firms to modernize any code or application. This includes custom programming languages and internal or specific apps, with AWS Transform able to tackle a wide range of challenges to potentially help businesses of all sizes. The company claims many organizations are seeing around 30% of their team's time being spent on manual modernization work, which takes valuable resources away from innovation. It claims customers have already used AWS Transform to analyze an estimated 1.1 billion lines of code and save more than 810,000 hours of manual effort - but have demanded more, particularly to harness the benefits of agentic AI tools. AWS says Transform can eliminate up to 70% of maintenance and licensing costs, with AWS Transform able to handle full-stack Windows modernization, accelerated up to 5x, operating across .NET apps, SQL Server, and user interface frameworks, and deployment layers. Its agents are able to analyze a company's complete Windows stack before proposing coordinated modernization plans across all layers. Once these are approved, the agent transforms the application, UI framework, database, and operating system, while providing updates and comprehensive transformation summaries. As it goes along, AWS Transform also automatically captures feedback and continues to improve over time, so each subsequent transformation becomes more reliable and efficient.

AWS Launches AI Agent to Automate Large-Scale Code Modernisation & Reduce Technical Debt | AIM

Early customers have reported up to 80% reduction in execution time, allowing teams to redeploy developer hours toward product work. AWS has introduced Transform custom, a new agent to automate large-scale code modernisation and reduce technical debt across enterprises. The company said the service "changes how organisations approach modernisation at scale" by combining pre-built transformations with custom, organisation-specific rules. According to AWS, early customers have reported up to 80% reduction in execution time, allowing teams to redeploy developer hours toward product work. The service is available through the AWS Transform CLI and web console. Transform custom applies its learned transformation patterns across large codebases, including hundreds or even thousands of repositories. It learns from documentation, natural-language instructions, and code samples, and then improves over time by analysing developer feedback and the manual fixes teams make during the modernisation process. The service includes a CLI and a web interface. The CLI supports conversational inputs for defining and executing transformations locally or in CI/CD workflows, while the web interface offers campaign-level tracking for modernisation projects across teams. AWS said the agent supports runtime upgrades for Java, Python and Node.js, and can execute complex transformations such as migrating Spring Boot applications or shifting workloads to AWS Graviton. It can also learn enterprise-specific coding patterns and apply them consistently. "The service understands not only the mechanical aspects of API changes, but also recognises best practices and optimisation opportunities available in newer SDK versions," the company said. Transform custom extends to Infrastructure as Code, with support for CDK-to-Terraform conversions and CloudFormation updates. A demonstration shared by AWS showed a Python 3.8 Lambda function being upgraded to Python 3.13 using the AWS/python-version-upgrade transformation. The agent analysed the codebase, changed deprecated syntax, updated dependencies, and produced evidence logs. The migrated version was stored in a new branch for developer review. AWS said that users can iterate as much as needed to refine transformation definitions before publishing them to an internal registry. Once published, custom transformations can be reused and applied repeatedly across different repositories. The agent generates a tailored JSON plan for each codebase and executes it step by step, supplying detailed evidence for each stage. AWS said the system centralises modernisation efforts that were previously fragmented across teams. "It keeps institutional knowledge available as scalable assets," the company said, positioning Transform custom as a way to standardise modernisation while limiting manual rework.