How Do I Modernize My Old System with AI? Learn from Amazon

You can now hand a large part of modernizing your old system to AI: it reads the code, extracts the business rules buried in it, proposes a plan and rewrites it, then your team reviews and approves the result. The clearest live example is Amazon's AWS Transform. AWS says it modernizes Windows, mainframe and VMware applications up to 5 times faster, and it charges $0.035 for every minute an agent actually works. But it does not replace your team, and it does not yet run from a cloud region in the Middle East.

This opens a new series for Saudi companies whose old system still works but has become costly or risky. For modernizing without downtime in general, see our earlier legacy modernization guide; here we focus on what AI adds.

What a legacy system is, and why it is hard to modernize

A legacy system is software your business cannot run without, but that is built on technology that is no longer supported or has become expensive to change. Common examples: an accounting system built on .NET Framework, an older version of Microsoft's platform for building software. A banking system written in COBOL, a programming language from the 1960s that still runs many financial systems, on a mainframe, a single large machine that runs every operation. Or virtual servers on VMware, software that runs several virtual servers on one physical machine.

The hard part is not writing new software. The real business rules, such as how a discount, a commission or interest is calculated, live inside the code rather than in any document, and the people who wrote them have often left. This is what technical debt means: every old shortcut or patch raises the cost of each later change. These systems are not a failure; they kept the business running for years. But modernizing them has become a risk-management decision, not a fashion.

How AI does the work: the AWS Transform example

According to the official service page, AWS Transform is a workbench where your team works alongside specialized AI agents, and it usually goes through four stages:

• Discovery and analysis. The agent reads the code and current systems and maps the applications and what each one depends on.
• Business-rule extraction and planning. It extracts and documents the business logic in the old code and proposes a prioritized modernization plan.
• Transformation. It rewrites the code in a modern technology or moves servers to the cloud.
• Human review. Your team reviews, tests and approves the results before go-live. AI speeds the work up, but the decision stays with you.

These four stages are not unique to Amazon. They are the same method you will see in any tool or team that modernizes a system with AI, which is why the rest of the series builds on them.

What you can modernize with it

The figures in the last column are AWS's own claims, so treat them as an upper bound the company advertises, not a guaranteed result for your project.

What happened for those who tried it

Thomson Reuters: according to the case study on the AWS website, the company now modernizes 1.5 million lines of code a month, 4 times faster, saved 30% in costs by moving from Windows to Linux, cut application conversion from several months to a single two-week cycle, and halved its technical debt.

Banks: on 29 September 2026, Thought Machine and AWS announced a solution that extracts banking product logic from COBOL code into clear specifications, then builds tested modern products from them. The idea is not to translate code line by line but to extract the «business intent», with human review gates that give the bank's engineering and risk teams final sign-off.

What it costs, and whether it fits your business in Saudi Arabia

According to the official pricing page, AWS charges for actual agent minutes at $0.035 per minute, about SAR 0.13 at the official rate of 3.75, and does not charge for your idle time or builds on your machine. Its example: transforming a 50,000-line .NET solution takes about 2,800 agent minutes, and AWS-managed .NET transformations include up to 50,000 agent minutes a month within the quota at no charge. You still pay for the cloud you move to, and for your team's review time.

• It fits if you have large systems on .NET Framework, a mainframe or VMware, you plan to move to AWS, and you have people who can review and test what the AI produces.
• It may not fit if your system is small and built on technologies outside its scope, or you do not plan to use the cloud. The same method still works for you with other tools.
• Watch where your data goes. According to AWS documentation, workspaces are available in 9 regions, none in the Middle East, and your data may be processed in another region. AWS has said its first cloud region in Saudi Arabia is on track to launch in December 2026, according to the Saudi Press Agency. If the code or databases contain personal data, read our PDPL guide before you start, and if you plan to move to the cloud anyway, read our cloud migration guide for Saudi businesses.

The Origami view

Modernizing legacy systems is a core part of our work building systems and integrations, and we see AI as a powerful way to speed up the hardest stage: understanding what the old system actually does. But we do not start with the tool; we start with the list of systems and the business rules that must not get lost on the way.

So we treat what the AI writes as a draft that an engineer reviews and tests on real data before it touches operations, and we choose where the data is processed before we choose the tool. If you are thinking about modernizing your system, take a look at our services.

This week's exercise: the systems that cannot stop

• Write down every system your company could not run a single day without: accounting, sales, inventory, payroll and any custom software.
• Next to each one, note the technology it was built with if you know it, and who fixes it when it breaks.
• Mark every system that only one person or one outside supplier understands. That is your first risk, and you will need this list in the next parts.

Coming up in the series

The next part answers a question that comes before everything else: does your legacy system need modernizing at all? We cover the signs not to ignore, and when it is better not to modernize yet. Then we inventory the systems, extract business rules with AI, compare rewriting, migrating and replacing, cover testing and cutover without stopping work, and close with where your data goes during modernization and a 90-day plan.

Sources

• AWS Transform: official service page — stages, system types, and AWS's speed and savings claims.
• AWS Transform: pricing page — $0.035 per agent minute, the 50,000-line example and the monthly .NET quota.
• AWS documentation: supported Regions — 9 workspace regions, and processing in another region.
• Thomson Reuters case study on the AWS website.
• Thought Machine and AWS announcement (29 September 2026).
• Saudi Press Agency: AWS region in Saudi Arabia in December 2026.
• SDAIA: Personal Data Protection Law.