Summary

• Infosys will provide ABN AMRO with application development, testing, and support while using its Topaz generative and agentic AI portfolio across the engagement.
• ABN AMRO already had almost 50 AI use cases in production by its second-quarter results and says AI is being embedded alongside legacy-system simplification and wider API use.
• Moving AI into mainstream technology delivery links model adoption to the slower work of reducing application complexity, controlling operational risk, and modernising long-lived banking systems.

ABN AMRO is moving artificial intelligence further into the machinery of its technology estate, extending its relationship with Infosys across application development, testing, and support as the Dutch bank tries to combine AI adoption with the slower work of simplifying systems and reducing operating costs.

ABN AMRO has renewed its strategic collaboration with Infosys, which will use its Topaz portfolio of generative and agentic AI technologies as part of work to modernise the bank’s IT landscape. Financial terms and the duration of the renewed agreement have not been disclosed.

The contract covers application development, testing, and support rather than a single AI application, making the scope more significant than another chatbot or isolated proof of concept. Infosys says the objective is to move beyond individual AI initiatives towards wider adoption across the bank’s technology environment while simplifying platforms and supporting digital services.

That description carries more weight because ABN AMRO is already making a broader structural change. The bank’s strategy calls for legacy systems to be phased out, API use expanded, processes simplified, and AI embedded more deeply into operations as part of a programme to reduce its cost base and modernise service delivery.

AI adoption is already widespread inside the bank

ABN AMRO reported in August that almost 50 AI use cases were in production across the organisation, while earlier in the year it said 85% of employees were using AI. Examples include an AI lending assistant that drafts credit proposals from documents, automated support for advisers, fraud and risk applications, and tools intended to reduce administrative work after customer conversations.

The bank has also established an internal AI register, standards, and compliance processes intended to govern deployments under requirements including the EU AI Act. Its public AI policy emphasises human oversight where necessary rather than treating adoption simply as a question of making models available to staff.

Extending AI into mainstream application work creates a different challenge from encouraging employees to use a productivity assistant. Development and testing systems sit closer to the operational technology on which banking services depend, meaning errors can propagate into applications, customer processes, regulatory controls, and infrastructure if automation is badly governed.

That tension is becoming familiar across financial services, where institutions are experimenting widely with generative and agentic systems while production deployment remains constrained by security, auditability, data controls, model risk, and legacy architecture. Recent banking research has shown a similar gap between AI experimentation and operational use in regulatory reporting.

Legacy simplification sits underneath the AI programme

ABN AMRO’s technology plans also form part of a broader cost programme. The bank intends to reduce its workforce by a net 5,200 full-time-equivalent positions by 2028 compared with 2024, with roughly half of that reduction expected to occur through attrition. Its strategy explicitly connects organisational simplification with digital processes, API expansion, legacy-system retirement, and AI.

By the end of the second quarter, the bank said it had completed around 45% of that planned workforce reduction. The figures should not be read as a direct measure of AI replacing employees, because ABN AMRO’s programme covers organisational restructuring, attrition, process changes, acquisitions, and technology modernisation rather than a single automation initiative.

Instead, they show the commercial environment surrounding the Infosys agreement. AI is being introduced while management is under pressure to make the bank less complex and cheaper to operate, which means technology projects will ultimately be judged against service quality, risk, productivity, and cost rather than simply the number of models deployed.

Legacy systems can be particularly resistant to that sort of change. Core banking estates often contain applications developed or acquired over decades, with dependencies that are difficult to document completely and migration programmes that can become expensive precisely because the old systems are still performing critical work.

Generative AI can assist with software development, documentation, testing, migration, and analysis, but it does not remove those dependencies. An automatically generated change still has to be understood, reviewed, tested, secured, and introduced without disrupting accounts, payments, lending, compliance, or customer channels.

Supplier dependence becomes part of AI governance

The renewed Infosys relationship also makes third-party technology governance more important. European banks operate under the Digital Operational Resilience Act, which places greater emphasis on understanding and managing ICT supplier risk, while AI introduces additional dependencies involving models, software libraries, data, cloud services, and specialised engineering expertise.

ABN AMRO is not relying on a single AI partnership. The bank has also entered a strategic relationship with French model developer Mistral AI and has joined Amsterdam AI initiatives, while its internal technology teams continue to develop their own applications. That mix suggests the bank is building an ecosystem rather than outsourcing its entire AI strategy to one services provider.

Even so, a long-running application-development partner occupies a different position from an experimental model supplier. Infosys engineers can influence how applications are designed, tested, maintained, and modernised across the estate, making the governance of AI-assisted delivery consequential even where the resulting software is not itself an AI product.

The renewed collaboration therefore places generative and agentic AI inside a familiar enterprise problem rather than beside it. ABN AMRO still has to retire old systems, control costs, maintain regulatory resilience, and keep banking services running while technology changes underneath them.

If AI produces durable gains, they are likely to emerge through that routine engineering work — fewer manual steps, faster testing, better documentation, simpler applications, and more consistent operations — rather than through a single dramatic deployment. Moving from dozens of AI use cases to a bank-wide capability means proving that the technology can survive precisely the systems and controls that made enterprise modernisation difficult in the first place.