Modernizing these systems to provide customers with digital, self-service experiences was necessary to meet the expectations of Californians. However, the DMV couldn’t meet these expectations before understanding decades of embedded business logic, which could carry unacceptable levels of operational risk. The DMV needed to regain an understanding of its COBOL and Assembly codebases quickly, efficiently and accurately.
The DMV relied on aging technology with a core architecture built in the late 1970s that depended on mainframe applications written in COBOL and Assembler. Those systems continued to run critical services, but decades of modifications to the embedded logic by personnel who are no longer available made it risky to reliably update the code.
This issue is a familiar modernization challenge for many large organizations: the legacy system works, but teams are unable to provide the embedded business logic from legacy code reliably to inform the modern design before replacing or redesigning core systems.
To complete this task, DMV brought in IBM Consulting®, IBM Analysis and Renovation Catalyst (ARC) and IBM watsonx.ai to help design a path forward. This partnership resulted in a solution that uses a flexible platform that can effectively deliver cutting-edge DMV services to millions of Californians.
The modernization problem hiding inside working code
The DMV’s Digital eXperience Platform program is designed to replace aging systems with modern technology and improve customer experience through digital, self-service channels. But a modern front end can only go so far if the business logic behind it remains locked inside decades-old code.
The DMV needed to extract and validate the business rules embedded across roughly 6 million lines of code spanning 2500 COBOL and Assembler programs. Those rules govern vehicle registration, driver licensing and other high-volume services. Manual extraction was an impractical approach. Mainframe skills were scarce due to retirement, attrition and hiring challenges.
Limited access to expert programmers proved to be just one challenge. Reading code line by line would have taken years. A change to one rule might also trigger regression risk across related modules, batch jobs, partners and downstream systems.
The value for users from this modernization is significant: fewer service disruptions, shorter lines at the DMV and better access to online services.
From COBOL and Assembler to business-readable rules
ARC is an IBM accelerator that analyzes mainframe and legacy application code to uncover dependencies, complexity, data relationships and embedded business logic. It turns legacy code into understandable modernization assets that teams can use for documentation, impact analysis and microservices planning.
The technical architecture paired the ARC accelerator with IBM watsonx.ai.
The flow started with legacy source code. ARC performed static analysis on COBOL and Assembler programs to identify structure, dependencies, variables, conditional logic and business-rule patterns. From there, ARC generated a set of work products that gave teams both technical and business views of the system.
Those outputs included:
- Size and complexity metrics, including lines of code, instructions, assignments, conditional constructs and complexity ratings
- Program call chains, showing dependencies across COBOL programs
- Pseudocode, which converted program logic into a more readable form
- Business rule summaries, tied back to program names and source line numbers for traceability
- Program-to-data dependency reports, including create, read, update and delete activity
- Structure diagrams, showing program hierarchy
- Logical flow diagrams, showing execution flow within a program
- Pattern-based rules, expressed as conditions and outcomes
The pipeline produced spreadsheet-based outputs that combined original code snippets, pseudocode and watsonx.ai-generated summaries. That format gave DMV experts a practical review surface. They were able to see the original technical source, the intermediate interpretation and the business-readable explanation in one place.
Human validation remained a nonnegotiable part of the process.
AI generated the draft interpretation, while IBM mainframe experts validated the outputs to ensure the accurate recognition of source language (COBOL and Assembler) syntax. The DMV’s SMEs reviewed the output for policy and regulatory accuracy. Feedback from those reviews helped inform improvements in the process across iterations.
The process also reduced risk by limiting exposure to operational systems. The AI workflow didn’t need to connect directly to production applications or process live citizen data. It worked from source code, generated pseudocode and created useful artifacts. That gave the team a safer way to analyze core logic without interfering with daily operations.
What IBM changed: Speed, traceability and safer delivery
The IBM approach dramatically accelerated modernization discovery for the DMV. Experts estimated that it would take up to five years to conduct a manual effort to extract, convert and validate business rules across the DMV’s legacy systems. Instead, using ARC and watsonx.ai reduced the timeline to just 15 months.
Within the first year, the team analyzed nearly 5 million lines of COBOL and Assembler code. As the solution evolved, extraction accuracy improved significantly, with COBOL outputs increasing from roughly 30% to 80% accuracy and Assembler outputs improving from 0% to 75%.
Just as important, IBM helped make decades of hidden mainframe logic understandable and traceable. Every generated rule was linked back to the original program and source line, enabling DMV experts to validate outputs with confidence. Visualizations like call graphs and flow diagrams also helped teams understand dependencies and reduce modernization risk before making changes.
The result was more than faster code analysis. IBM created a reusable modernization foundation that supported work across vehicle registration, driver licensing and other critical DMV systems—while also enabling future capabilities such as automated test generation, impact analysis and user story creation.
What the DMV learned about modernizing systems that still run
The DMV’s modernization journey highlights several lessons for organizations managing complex legacy environments:
- Understanding before replacement: Modernizing legacy systems requires more than code conversion. Teams first need visibility into the business rules, policies and dependencies embedded across decades of applications.
- Use AI to accelerate expert productivity: AI can rapidly extract, organize and convert legacy logic, enabling subject matter experts to focus on validation, decision-making and exceptions instead of manual code analysis.
- Keep humans in the loop to build trust and consistency: Enterprise modernization requires repeatable, explainable outputs. Controlled prompts, standardized workflows and expert validation helped improve consistency, reduce variation and ensure regulatory and operational accuracy.
- Make modernization assets fully traceable: Linking extracted rules back to original programs and source lines improved auditability, impact analysis and confidence in modernization decisions.
- Create reusable knowledge for future transformation: Extracted rules, dependency maps and visualizations became long-term assets that can support future testing, onboarding, planning, audits and broader modernization initiatives.
From extracted rules to the next phase of DMV modernization
The first phase of the DMV’s modernization journey answered a challenging question: what does its legacy system actually do? The next phase builds on that answer. With business rules extracted, validated and connected back to source programs, the DMV can use that knowledge to design modern applications faster and with less risk.
The work’s outputs are more than documentation. Business-readable rule summaries, dependency maps, logical flow diagrams and program-to-data reports give teams a structured foundation for forward engineering. They can help generate user stories, test cases, microservices and APIs based on real production logic rather than assumptions about how the old system behaves.
That foundation also changes the role of governance. As AI-assisted development expands, the DMV needs clear controls that preserve traceability from generated outputs back to the original business rule and source program. Validation workflows, version control, data access policies and privacy controls help make AI-generated artifacts useful for modernization while keeping them ready for audit, compliance review and change management.
The skills model also evolves. AI-augmented engineers, business analysts, quality engineers, enterprise architects and AI governance leads all need to work from the same managed knowledge base. Treated as a living asset, the extracted ruleset can feed design, development, testing and ongoing policy updates across modernization workstreams.
That is where the first phase becomes more than analysis. It gives California DMV a reusable modernization layer. Platforms including IBM watsonx.ai, IBM Bob™ and Claude can employ this structured knowledge to accelerate application development, reduce regression risk and compress delivery timelines for new digital services—all without losing the business logic that keeps public services working.
Read more about IBM watsonx.ai
Ajay Gupta, Chief Digital Transformation Officer, California DMV
Syed Ali, Distinguished Engineer, CTO, IBM
