Driving Efficiency in Software Delivery
Mashreq Bank in Dubai has run a pilot using agentic AI to support software engineering tasks behind its digital banking, payments and API services. The bank reports 60 to 70 percent reductions in time for integration design and troubleshooting during the pilot, a level of acceleration that directly impacts time to market for new features and fixes.
The Technology Behind the Transformation
The pilot layered IBM Bob, an AI-powered software engineering platform, with IBM Context Forge to feed the agents with technical context such as API specifications and architecture standards. Agents interacted through a secure gateway handling roughly 4,000 requests per month during the trial. The stack sits on cloud-native infrastructure components including Red Hat OpenShift and Confluent for event streaming, enabling agents to operate against live engineering artifacts and CI/CD pipelines.
Balancing Speed with Quality Assurance
Mashreq’s results demonstrate clear productivity gains but also highlight governance and resilience questions. Faster delivery creates pressure on testing regimes, incident response and operational monitoring. Banks must treat agentic AI as an active collaborator rather than a passive tool, and introduce controls around output validation, traceability and failure-mode analysis. For mission-critical components, the bank is evaluating on-premises deployments to keep sensitive processes within firm-managed boundaries.
The Future of AI in Banking Engineering
The pilot suggests agentic AI can shift routine engineering work—integration design, code review and initial troubleshooting—from humans to agents, freeing senior engineers for higher-value design and oversight. The strategic task for CIOs and CTOs is to scale those gains across broader workloads while maintaining demonstrable control over quality and resilience.
- Operational takeaway: pilot measurable, repeatable metrics for productivity and defect rates.
- Control takeaway: build validation layers and audit trails for agent outputs.
- Technology takeaway: combine agent platforms with context services and cloud-native infrastructure, with on-prem options for critical flows.
Mashreq’s experience offers a practical blueprint: agentic AI can accelerate software delivery, but sustaining those gains requires new QA patterns, governance and operational resilience planning.
