Cursor Replaces Amazon Q at NAB for 6,000 Developers

National Australia Bank (NAB) has standardized its engineering organization on Cursor, rolling out the AI-native editor to 6,000 developers. The bank plans to scale the deployment to over 10,000 employees, including product managers and designers. This rollout replaces a previous pilot of Amazon Q Developer that involved 350 engineers. The transition marks a strategic shift toward agentic engineering, utilizing multi-model switching to handle both high-reasoning architecture tasks and routine interface work.

Legacy Codebase Modernization

NAB’s migration of legacy systems is moving 3x faster than original estimates. A major focus is reverse-engineering core banking systems, which manage balances and interest, currently written in Assembly. Engineers are bypassing manual 3270 terminal analysis by using AI to generate flowcharts and business logic summaries directly from machine instructions.

In the commercial lending division, the bank is refactoring BizCalc, a pricing application, from a monolithic Silverlight and .NET architecture into Java microservices with a React frontend. Using Cursor’s Ask Mode and Plan Mode, engineers completed two months of API specifications and user stories in one week.

Greenfield Development and Metrics

For new builds, the productivity uplift is similarly aggressive. Internal metrics from NAB show a 40x leap in the speed of generating software requirements and a 5-6x improvement in active development phases. A merchant services team recently delivered a hardware-agnostic payment application in three weeks, a project initially scoped for four months.

Across the organization, AI suggestions currently account for 40% of production code generated at the bank.

The Context Engineering Library

To maintain security and architectural standards across thousands of GitHub repositories, NAB built a proprietary Context Engineering Library (NAB CEL). If you build multi-agent systems in an enterprise setting, you need a mechanism to enforce guardrails. NAB achieved this by mapping Cursor primitives—rules, skills, and hooks—into their internal library, ensuring the models have specific context about the bank’s codebase requirements.

This library-driven approach to context engineering allows developers to switch between heavy-reasoning models for structural tasks and faster, cheaper models for boilerplate code without losing organizational compliance. The bank is embedding the tool into the entire software lifecycle, expanding beyond coding to cover QA testing and deployment.

Evaluating coding assistants solely on line-completion metrics misses the larger opportunity in reverse-engineering and automated refactoring. For engineering teams managing mainframe infrastructure, AI-native editors provide a direct path to untangle legacy logic without relying purely on specialized manual analysis.