NAB Picks Cursor AI Over Rivals

National Australia Bank (NAB) has standardized 6,000 developers on Cursor AI, a move that follows an evaluation against Amazon Q and GitHub Copilot. The decision has reportedly led to legacy modernizations and refactoring efforts running three times faster and at higher quality. This pivot marks a significant adoption of AI coding assistants within one of Asia-Pacific's largest financial institutions, underscoring the growing role of AI in banking.

NAB's engineering teams are now tackling projects previously deemed out of reach, including complex monolith refactors to microservices and mainframe migrations away from Assembly. One merchant services team even developed a hardware-agnostic payment app in just three weeks, a task originally scoped for four months.

Enterprise AI Adoption

NAB's initial exploration involved Amazon Q and GitHub Copilot, but a broader evaluation led to the selection of Cursor for an initial cohort of 6,000 developers. Key factors influencing the decision included model flexibility, allowing engineers to select optimal AI models based on cost, latency, and task complexity.

The platform's deep codebase understanding across thousands of repositories and diverse tech stacks, from Java and React to COBOL and Assembly, proved critical. This capability is essential for effective legacy system modernization, a challenge many established financial firms face.

Furthermore, NAB developed an internal context engineering library, NAB CEL, leveraging Cursor's primitives for rules, skills, and hooks. This approach centralizes shared knowledge, enforces development standards, and provides guardrails for AI agent behavior, offering greater extensibility and control compared to plugin-based solutions.

"With Cursor, the agent understands our codebase and works the way NAB works," stated Chris De Lorenzo, Principal Engineer at NAB.

NAB is now scaling Cursor's use to over 10,000 employees, encompassing engineers, product managers, designers, and leadership, with tailored training paths for each function. This strategic investment in enablement, including dedicated sprint days using Cursor on production projects, is yielding tangible productivity gains.

Accelerating Legacy Modernization

The bank's business lending unit, for instance, is refactoring its BizCalc fee calculator. Originally a Silverlight/.NET monolith, the application must transition to Java microservices and React before Silverlight's extended support ends in 2026.

The pre-development phase alone, initially scoped for two months of documentation and requirement gathering, was completed by Principal Engineer Coby Paterson in just one week using Cursor's Ask and Plan modes with custom NAB-CEL skills. This rapid documentation and API specification generation is a stark contrast to manual reverse-engineering efforts.

Paterson anticipates the full migration will conclude in two months, a threefold improvement over the original six-month estimate. This efficiency gain is crucial for banks navigating the complexities of AI in banking and modernizing critical infrastructure.

Mainframe Migration Gains Traction

NAB's core banking systems, running on Assembly-based mainframes, manage vital functions like customer balances and interest accrual. Migrating these key programs off the legacy infrastructure has been a long-standing goal, hampered by resource constraints and the specialized Assembly expertise required.

Previously, the manual process of separating low-level machine instructions from business logic was a significant bottleneck. Harjot Singh, Engineering Manager at NAB, noted, "Before Cursor, we couldn't even think about moving away from Assembly."

Using Cursor, Singh's team can now generate flowcharts and business summaries directly from Assembly code, significantly accelerating the migration. The project is progressing three times faster than anticipated, making the substantial undertaking financially viable.

Greenfield Development Boost

The merchant services team aimed to build a hardware-agnostic payment app, avoiding vendor lock-in. This greenfield project, initially projected to take four months due to the team's unfamiliarity with Android frameworks like Kotlin, was completed in under three weeks using Cursor.

Principal Engineer Chris De Lorenzo utilized Cursor for detailed product requirements and a phased implementation plan, leveraging advanced coding models. De Lorenzo credits Cursor with a 5-8x improvement in development velocity for this project, emphasizing that the initiative might not have been attempted without the AI assistant.

He also highlighted Cursor's ability to integrate engineers, architects, and product teams into a unified workflow, fundamentally changing the organization's approach to software development.

NAB plans to embed Cursor further into the software lifecycle, including code review, QA testing, and deployment, signaling a broad strategic re-evaluation of engineering processes around AI agents.