What challenges did Toyota Motor Europe face before AI implementation?

TME manages more than 70 custom-built applications on legacy mainframe and AS400 platforms. The core warranty-handling application contains more than 1.3 million lines of NCL code. Scarce NCL expertise and limited documentation created risk of losing embedded business logic during modernization.

What was the impact of the AI solution?

All three documentation types were validated positively by TME technical and business experts. The solution established a reusable agentic architecture for other legacy systems across TME’s modernization portfolio.

Source

Proof of conceptEvidence: Medium50/100

Toyota Motor Europe (TME) automates legacy mainframe code documentation with Amazon Bedrock AgentCore

Use case typeCode modernizationUpdated Mar 4, 2026

Toyota Motor Europe (TME) built a proof of concept with Deloitte and the AWS Generative AI Innovation Center to automatically generate documentation from legacy NCL source code. The solution produces technical YAML documentation, business HTML reports, and Mermaid process-flow diagrams from a legacy warranty-handling application, using Amazon Bedrock, Strands Agents SDK, Amazon Bedrock AgentCore, and an Amazon Bedrock Knowledge Base. The workflow uses agentic orchestration, retrieval-augmented generation, and bottom-up diagram composition to overcome context-window limits and preserve embedded business logic for modernization.

Organization: Toyota Motor Europe
Industry: Automotive
Location: Belgium
Published: March 2026

Reported outcomes

Modules documented: 2 modules

Automation & deflection

Planned next steps

The source says the pilot could deliver: Established reusable documentation architecture.
The PoC automatically documented 2 of the 10 modules in the core warranty-handling application.

Why do we believe this?Outcome claims, sources, and evidence checks

Normalized claim

Modules documented: 2 modules

AWS BlogMar 4, 2026Blog postExplicit claimMedium evidence strength

Automated legacy code documentation across 2 of 10 modules in a single core business application using generative AI.

Why do we believe this deployment?Customer identity, provider attribution, maturity, and source checks

Customer: Toyota Motor Europe
Provider: AWS
Maturity: PoC
Linked source: AWS Blog

Toyota Motor Europe (TME) built a proof of concept with Deloitte and the AWS Generative AI Innovation Center to automatically generate documentation from legacy NCL source code

Customer identity supportedSource describes one deploymentMaturity supported

Primary read

Use case focus

Showing 2 of 2

1Code modernization
2Document automation

TME partnered with Deloitte and the AWS Generative AI Innovation Center to build a PoC that turns legacy NCL source code into multiple documentation formats.
The solution uses Amazon Bedrock orchestrated through the Strands Agents SDK and designed for future scalability with Amazon Bedrock AgentCore.
For technical documentation, a Strands Agent uses an Amazon Bedrock Knowledge Base built on Amazon Titan Text Embeddings V2 and a FAISS vector index stored in Amazon S3 to retrieve language-specific context on demand.
The pipeline generates structured YAML technical documentation, converts it into business HTML reports, and creates Mermaid process-flow diagrams, using Amazon Bedrock Converse API and Anthropic Claude Sonnet 4 for iterative reasoning and prompt-driven transformation.

Technologies

Amazon Bedrock Amazon Bedrock AgentCore Strands Agents SDK Anthropic Claude Sonnet 4.0 Amazon Bedrock Knowledge Bases Amazon Titan Text Embeddings V2 Amazon Bedrock Converse API Amazon S3 FAISS Mermaid NetworkX

All three documentation types were validated positively by TME technical and business experts.
The solution established a reusable agentic architecture for other legacy systems across TME’s modernization portfolio.

Architecture

A Strands Agents SDK workflow on Amazon Bedrock orchestrates multiple stages. Technical documentation uses a Strands Agent plus Amazon Bedrock Knowledge Bases and Amazon Titan Text Embeddings V2 over a FAISS index in Amazon S3 to resolve unknown NCL constructs. Business documentation uses the Amazon Bedrock Converse API with structured prompts over YAML outputs. Process-flow diagrams are generated in a bottom-up composition pipeline using Python graph construction and Amazon Bedrock/Anthropic Claude Sonnet 4 prompts. Amazon Bedrock AgentCore is cited as the future scalable agentic platform, and Amazon S3 stores source code, reference material, and generated documentation.

Implementation partners2

Deloitte AWS Generative AI Innovation Center

Sources & evidence1

Evidence: Medium50/100Evidence strength