Proximal for Critical Legacy Systems

This points to a much larger and stickier market than selling tools to AI labs alone. Legacy modernization is the same core job, reading messy code, tracing dependencies, rewriting safely, and proving nothing mission critical broke, but the buyer shifts to institutions running hospitals, grids, and government workflows where old systems cannot simply be replaced. That makes trust, validation, and deployment controls as important as raw coding ability.1 2 3 4 5

• FrontierSWE is built around ultra long horizon engineering tasks, with real problems drawn from performance engineering, computational science, and ML research. That matters because legacy modernization is rarely a single file edit. It is weeks of understanding how old modules interact before touching production code.1 2
• Comparable companies are already aiming at the same budget line. Anthropic and PwC package Claude for COBOL conversion, monolith refactoring, test automation, and migration work in regulated industries. Blitzy and Reflection AI are also framed around enterprise modernization, showing this is becoming a distinct category inside AI coding.3 4 5
• The customer motion changes with this expansion. AI labs buy for model research speed, while enterprises and governments buy for safer upgrades to systems that already run payroll, patient workflows, or public infrastructure. That usually means longer sales cycles, services partners, and heavier review before any generated code reaches production.3 6 7

The likely end state is that coding agents split into two tiers. One tier helps developers ship new software faster. The more valuable tier becomes a modernization engine for old, business critical systems. If Proximal can turn frontier benchmark performance into auditable rewrites for regulated environments, it can move from a lab tooling vendor into core infrastructure spending.1 2 3 4 5