AI Revolutionizes Legacy Code Migration: 70K-Line COBOL-to-Rust Port Achieved in Just 3 Days

Breaking: Developers Clone GNU COBOL Compiler in Rust Using AI, Finish in 72 Hours

In a stunning demonstration of AI-driven software engineering, a development team has successfully produced a behavioral clone of the GNU COBOL compiler—rewritten entirely in Rust—in just three days. The resulting codebase is 70,000 lines long.

AI Revolutionizes Legacy Code Migration: 70K-Line COBOL-to-Rust Port Achieved in Just 3 Days — Source: martinfowler.com

This breakthrough, shared during a confidential industry retreat held under the Chatham House Rule, underscores the growing power of large language models (LLMs) to port legacy code to modern platforms with unprecedented speed.

"Good regression tests were likely invaluable here," said one attendee, speaking on condition of anonymity. "If you have access to an existing implementation, you can even build a test suite from scratch."

Interrogatory LLM: AI Interviewing Humans to Validate Specs

Another attendee described a novel technique called the "Interrogatory LLM." Instead of having humans struggle through massive specification documents, the LLM interviews an expert, asking targeted questions to verify correctness.

"Large spec documents are complex for humans to review," the developer explained. "This method flips the dynamic—the AI probes the human for gaps."

Change-Control Boards as 'Scar Tissue' of Organizational History

A separate insight focused on organizational learning. One consultant revealed that their first step when working with a client is to read the change-control board guidelines.

"That document is the scar tissue of everything that has gone wrong," the attendee said. "To understand why a system is the way it is, you must understand the history that shaped it."

Revisiting 'Lift and Shift': AI Makes It the First Step, Not a Dead End

For years, many modernization experts dismissed "lift and shift"—porting a legacy system to a new platform while keeping feature parity—as a missed opportunity. They argued that legacy systems are bloated, with up to 50% of features unused (according to a 2014 Standish Group report).

But the advent of LLM-powered code porting is changing minds. "One attendee, who specializes in legacy migration, said lift and shift should now always be the first step," the group noted. "The cost is no longer prohibitive, and a modern platform makes further evolution far cheaper. Just don't stop there."

Background: The Retreat and the Rise of Agentic Programming

The retreat, which took place in mid-May, brought together software professionals to discuss the future of programming in an era of agentic AI systems. Attendees included engineers from major financial institutions, cloud providers, and open-source communities.

Chatham House Rule was enforced, meaning all quotes are unattributed to individuals. The discussions reflected a growing consensus that LLMs are not just coding assistants but transformative tools for software archaeology and migration.

What This Means: Faster Legacy Migration, Lower Risk, But New Challenges

The ability to port 70,000 lines of COBOL to Rust in three days dramatically lowers the barrier for modernizing critical legacy systems. Financial institutions, which operate under strict regulatory controls and face immense risk from software errors, stand to benefit most.

"Several attendees from the finance sector emphasized the complexity of legacy environments combined with regulatory controls," the report states. "This AI-driven capability could cut migration timelines from years to weeks."

However, experts caution that lift-and-shift is only a starting point. Organizations must still analyze which features users actually need—and resist the temptation to blindly replicate past bloat. The technique also requires robust test suites and human oversight to avoid introducing subtle bugs.

As one attendee summarized: "The cost of porting has fallen dramatically. But if you treat it as a complete solution, you'll just digitize your old problems faster."