NatWest automates 33% of software code with agentic AI, achieving 10x productivity in financial crime units

NatWest's engineering organization of 12,000 engineers faced mounting pressure to accelerate software delivery across complex financial systems without proportional headcount growth. In retail banking — where customer-facing digital products, fraud detection, and compliance tooling require continuous iteration — the pace of manual code authoring, review, and testing had become a structural bottleneck. Financial crime units in particular operated under strict regulatory timelines, yet their development workflows were no faster than any other team. The status quo meant slower product cycles, higher engineering costs, and limited capacity to address compliance-driven development demands at scale.

NatWest deployed AI coding tools across its full engineering organization to assist with drafting, reviewing, and testing software. For the majority of engineers, this meant AI-augmented workflows integrated directly into existing development environments. In financial crime units, NatWest went further — piloting agentic engineering, where autonomous AI agents handle multi-step development tasks end-to-end without requiring per-step human prompting. This agentic approach, distinct from standard code completion, allowed agents to plan, write, test, and iterate on code independently. The deployment moved from targeted pilots in structured, high-compliance workflows — where task boundaries are well-defined — before broader organizational rollout across the 12,000-engineer base.

AI now generates over one-third (33%) of all software code written at NatWest, representing a fundamental shift in how the engineering organization operates at scale. In financial crime unit trials, agentic engineering produced a 10x increase in developer productivity — the most significant gain reported in any unit. Separately, the retail division reclaimed 70,000 hours of staff time through AI-powered automation of call summaries and complaint drafting, directly reducing administrative overhead in customer-facing operations. Key outcomes include:

• 33% of total software output now AI-generated
• 10x productivity gain in financial crime engineering pilots
• 70,000 hours saved in retail operations through documentation automation
• Relationship managers freed up 30% more time for client-facing work via AI meeting summaries

• Agentic AI differs from copilots in kind, not degree — autonomous agents completing full development tasks, not just suggesting lines, are what drives order-of-magnitude productivity gains.
• Structured, compliance-heavy workflows make ideal pilot targets — financial crime and regulatory units have well-defined task boundaries that constrain agent scope and reduce risk during initial rollout.
• Scale requires systemic adoption — deploying to 12,000 engineers means embedding AI into standard tooling and workflows, not running isolated experiments.
• Pair engineering AI with operational AI — NatWest's 70,000-hour savings came from automating documentation in retail operations, not just code; both layers compound the ROI.