Modernize Software touts deterministic AI for COBOL modernization

Modernize Software said it is pairing generative AI with a deterministic translation engine to modernize mission-critical COBOL and other legacy systems without losing business rules. The Austin-based division of Semantic Designs is pitching the approach as a way to make code migration faster, repeatable and auditable for regulated industries and government users.

Why it matters: - Modernization buyers in aerospace, finance, government and other regulated sectors need more than faster code conversion. They need proof that migrated systems preserve every business rule and behave the same way as the originals. - Modernize Software is positioning deterministic translation as the safeguard that generative AI alone cannot provide. - The company is also targeting environments where source code cannot leave the client network.

What happened: - Modernize Software, the legacy modernization division of Austin-based Semantic Designs, outlined its approach to mission-critical software modernization on October 16, 2026. - The company said it uses generative AI to speed up code analysis, documentation and testing. - Modernize Software said its deterministic DMS engine then verifies that translated code preserves the original system’s business rules. - DMS measures transformation completeness by checking that all nodes in the original program’s abstract syntax tree are translated. - Semantic Designs said testing helps confirm the resulting application preserves the original system’s business rules and behavior.

The details: - DMS, short for Design Maintenance System, combines symbolic AI with compiler technology. - The platform parses source code into abstract syntax trees and analyzes control and data flow through the program. - DMS applies mathematically defined rewrite rules to generate maintainable code in the target language. - The same input produces the same output on every run. - Each change can be traced to the rule that produced it. - Code that cannot be converted deterministically is flagged for engineer review. - Generative AI runs alongside DMS to extract business rules, document architecture and design tests, while the deterministic engine controls final translation. - Semantic Designs said it has built code transformation tools since 1995. - The company said its published work includes a conversion of the Northrop Grumman B-2 Spirit flight program from JOVIAL to C with 100% accurate translation, validated jointly with Northrop Grumman engineers. - The company said it also converted a 7-million-line COBOL application to C# with 100% of the conversion fully automated. - Semantic Designs said it built a custom DMS-based tool for Dow to migrate chemical plant process control software, under contract in 2012 and still in active use. - The company said it supports more than 40 languages and dialects, including 13 COBOL dialects and mainframe technologies such as CICS, IMS, VSAM and JCL. - Modernize Software said organizations can submit code under mutual NDA for an assessment that includes complexity analysis, an automation estimate and a rough timeline. - The company said DMS can be deployed on-premise for classified and regulated environments so no source code leaves the client’s network. - Modernize Software said more information is available at the company’s announcement. - The company also listed Semantic Designs on LinkedIn.

Between the lines: - The announcement lands as interest in AI-assisted COBOL modernization is rising. - The company is drawing a sharp line between probabilistic AI and deterministic transformation, arguing that legacy systems with high stakes require repeatable results and an audit trail. - Dr. Ira D. Baxter, founder and CEO of Semantic Designs, said understanding code and proving a translation is correct are different problems, and systems that move money, pay benefits or fly aircraft need deterministic verification. - Baxter said organizations should ask whether running the translation twice produces the same code, whether every change can be traced to a defined rule and how errors will be found before production. - The company said large legacy systems can contain millions of lines of code, undocumented rules and retired engineers who no longer know the original design intent. - Modernize Software said language-model translation is probabilistic, meaning the same input can produce different output on separate runs. - The company said errors may compile cleanly and appear correct, making manual review at multi-million-line scale impractical.

What’s next: - Modernize Software is offering assessments for organizations considering COBOL or other legacy migration projects. - The company is also pitching on-premise deployment for agencies and regulated enterprises that need tighter control over source code and data. - Buyers evaluating modernization providers are being pushed to focus on repeatability, traceability and pre-production error detection rather than automation claims alone.

The bottom line: - Modernize Software is betting that the winning AI modernization stack will pair generative speed with deterministic proof.