The news
Thought Machine, the core banking software company, and Amazon Web Services (AWS), the cloud arm of Amazon (AMZN), announced on September 29, 2026 an AI-powered pipeline meant to speed the migration of banks off legacy mainframe systems that many banks still depend on.
The pipeline links AWS Transform, Amazon's set of AI agents for reverse engineering old applications, with Vault Forge, Thought Machine's AI tooling suite, which runs on Amazon Bedrock, AWS's service for accessing large AI models. According to the companies, banks can take millions of lines of mainframe code, such as COBOL, an older programming language, and turn it into tested financial products in a live sandbox. The release says this compresses multi-year processes into days.
The companies describe four stages. AWS Transform first extracts business logic from legacy applications into structured requirement documents written in a format called Easy Approach to Requirements Syntax. The pipeline then consolidates what the release calls hundreds of redundant legacy product variations into leaner specifications. Vault Forge next uses specialised AI agents to write Python financial products, test suites and configurations for Thought Machine's Vault platform. Finally, bank engineering and risk teams review the output and hold explicit sign-off authority before it is deployed to a Vault sandbox for further evaluation.
The approach rests on a design choice. Thought Machine says older cores hardcode products such as interest calculations, fee schedules and repayment rules into database infrastructure, while Vault defines every product as Python code separate from the underlying system. That separation, the company argues, lets AI agents skip database complexity and translate the rules directly. AWS and Thought Machine also said the pipeline runs entirely inside the bank's own AWS environment.
Thought Machine is also launching a joint Core Modernisation Accelerator programme, which pairs its forward-deployed engineers, architects and product specialists with bank teams on delivery, governance and setup. The companies did not disclose pricing, named customers or measured results from banks using the pipeline.
Thought Machine CEO Paul Taylor said the partnership rewrites "the economics of core modernisation." Charith Mendis, AWS head of worldwide banking, said the companies believe it will change how the industry approaches core transformation. Thought Machine says its Vault platform is used by more than 65 banks and financial institutions and that it has raised more than £500 million.
The numbers
- Banks and institutions using Vault (company figure)
- More than 65
- Thought Machine funding raised
- More than £500 million
- Pipeline stages
- 4
- Claimed timeline change
- Multi-year processes into days (company claim)
Why CEOs should care
For bank CIOs and CFOs, replacing the core system is among the largest technology projects a bank can take on. A tool that shortens the discovery and rule-mapping phase could change the business case, but the claim of days rather than years comes from the vendors, not from a published customer result. Ask Thought Machine and AWS which stage of a migration the days figure covers, whether it includes data migration, testing with regulators and parallel running, and which banks have used the pipeline in production.
Chief risk officers and CISOs should focus on the controls. The companies say processing stays inside the bank's own AWS environment and that humans sign off before anything reaches a sandbox. Useful questions: how are AI-generated products audited line by line against the old system's behaviour, who owns errors in generated code, and how will examiners be shown that interest and fee calculations match what customers were promised. Any rule that the extraction step misses could become a customer-facing pricing error later.
Boards and procurement teams should also weigh lock-in. The pipeline outputs products for Thought Machine's Vault platform and runs on AWS services, so adopting it means committing to both. That may be the right call, but it should be a deliberate choice with exit terms in the contract, not a side effect of a faster migration.
The bigger picture
AWS itself calls extracting business logic from legacy systems one of the hardest banking modernisation challenges. Thought Machine has landed at least one large mandate recently: The Paypers reported on September 1, 2026 that Deutsche Bank signed a global licensing agreement for Vault Core, starting with retail banking in Germany, with product migration planned from 2027. Tools that promise to lower migration risk give core vendors and cloud providers a new way to compete for those mandates.
Using AI to read old code is a sensible target: business rules buried in decades of mainframe code can be hard to trace by hand. The harder question is proof. Banks will want evidence that generated products behave exactly like the systems they replace before trusting customer balances to them.
What’s next
Watch for the first named bank to use the pipeline publicly, any measured timelines or cost figures, and whether Deutsche Bank's planned migration from 2027 draws on the new tooling. Regulators' views on AI-generated product code in core systems will also shape how fast banks move.
What “Fact-checked” means
Fact-checking means testing a story’s facts against the evidence before it is published. This story went through at least two separate checks before this version was published.
- What we checked
- Its names, figures, dates, job titles, quotes and who said what were checked against the story’s sources, including its main source where it could be opened. The headline was checked for accuracy and overstatement.
- How
- A first check reviewed the whole story. If it passed, a second, skeptical check went back to the sources to look for mistakes in the most important facts. If a check flagged the story, it was edited to fix the problems found, and a separate re-check then reviewed the whole story again.
- Who
- The checks are made by our newsroom, as steps kept separate from the writing, under rules set by our editor, Hussein Mukhtar. A story the checks still flag is held for the editor, who decides whether it is fixed, published or dropped.
- If something is wrong
- “Fact-checked” does not mean error-free. If a material error is found after publication, we correct the story and add a note saying what changed. Report an error
