Toyota Motor Europe runs more than 70 custom applications on mainframe and AS400 platforms. Several run on NCL, a proprietary language, with documentation that is either thin or long out of date, and the warranty handling application alone is over 1.3 million lines of it. The problem isn't really the code. It's that the people who understand it are retiring, and the business logic lives in their heads rather than anywhere written down, so migrating without recovering it first means rebuilding something subtly different from what the business has relied on for decades.

With Deloitte and the AWS Generative AI Innovation Center, we built a pipeline on Amazon Bedrock, orchestrated with the Strands Agents SDK, which reads NCL source and writes documentation from it. The proof of concept covered two of the ten modules in the warranty application. The requirement shaping the design is that one document won't do: migration engineers need technical detail, business stakeholders need to confirm the system does what they think it does, and coordination between them needs something in between. Each audience needs the same logic at a different altitude, which is expensive by hand and cheap once the extraction is done.

Toyota's technical senior manager described the result as a comprehensive view of the technical and business flows of one of their oldest applications, and a basis for deciding how to modernise it. The proof of concept has since moved to production rollout, and the approach generalises to the rest of the NCL estate.