Mambu is pushing agentic AI deeper into banking infrastructure with the launch of Intelligent Core, an architecture that connects AI agents directly with core banking and payments rather than treating artificial intelligence as a separate layer sitting above the technology stack. The approach combines Mambu Core, Mambu Payments and the newly introduced Mambu Agentic platform, giving financial institutions a framework through which AI can access banking context, identify opportunities or anomalies and take authorised actions within defined controls. The launch reflects a broader shift in banking AI, where the focus is moving beyond employee assistants and standalone applications towards systems capable of participating directly in operational workflows.
Mambu brings AI closer to the banking ledger
The Amsterdam-based banking technology provider said Intelligent Core is designed as an open, composable architecture connecting core banking, payments and agentic AI.
Mambu has spent the past 15 years building its position around composable banking, providing technology for deposits, lending and Islamic banking. Intelligent Core extends that model by introducing an intelligence layer capable of interacting more directly with the systems responsible for accounts, transactions and payments.
At the centre of the architecture is Mambu Agentic, which provides AI connectivity, specialised agents and autonomous workflows. Rather than requiring banks to build separate integrations between AI applications and core systems, Mambu wants the intelligence layer to operate as part of the underlying banking architecture.
The company argues that this could allow financial institutions to automate more operational activity while making decisions using real-time information from core and payment systems.
Three layers support Mambu Agentic
Mambu is introducing three capability layers within its agentic offering: a Model Context Protocol solution, AI Insights and configurable Agents.
Its Model Context Protocol, or MCP, solution is designed to connect enterprise AI agents with core banking and payment systems. This provides agents with access to relevant data and operational context while maintaining governance over what information and functions they can use.
The approach is intended to reduce dependence on middleware and custom integrations. For banks experimenting with multiple AI models and agents, standardised connectivity could become increasingly important as institutions seek to prevent another layer of fragmented technology from developing around AI.
AI Insights provides a curated data layer intended to improve the reliability of information used by agents for reasoning and decision-making. Mambu said the capability can also identify growth opportunities, risk indicators and anomalies across portfolios and operations.
The Agents layer moves from analysis towards action. Pre-built agents can gather information, reason within configurable guardrails and perform authorised activities. Financial institutions can define policies, approval thresholds and business logic that determine what an agent is permitted to do.
Mambu said actions and decisions can also be explained and audited, an important consideration as banks explore greater levels of AI autonomy.
Agentic AI moves from assistance towards execution
The distinction between generative AI and agentic systems is becoming increasingly important for banks. Much of the first phase of banking AI has centred on summarising information, assisting employees, generating content and improving customer interactions. Agentic AI introduces the possibility of systems that can complete multi-step tasks and initiate actions rather than simply provide recommendations.
Connecting those capabilities directly with core banking infrastructure raises the potential impact considerably. An agent with governed access to account, transaction or payment information could potentially support activities ranging from servicing and operational processing to portfolio monitoring and personalised customer actions.
It also raises the importance of controls. Banks need to determine not only what information an AI system can access, but which actions it can perform, when human approval is required and how every decision can be reconstructed for audit and regulatory purposes.
Mambu is positioning configurable guardrails and explainability as central elements of its architecture rather than controls added after deployment.
Legacy infrastructure remains part of the AI challenge
Banks have rapidly increased investment in AI, but deploying it across production banking environments remains more complicated than introducing individual AI applications. Core systems were generally designed around deterministic processes, permissions and transactions rather than autonomous software agents capable of interpreting context and deciding what to do next.
This creates an infrastructure question alongside the AI question. Institutions may have increasingly capable models, but their ability to use them operationally depends on how easily those models can access trusted data and interact safely with banking systems.
Mambu CEO Fernando Zandona said financial institutions are increasingly focused on whether AI can operate safely and reliably at scale rather than whether they should adopt the technology.
“Intelligent Core is our vision for the future of financial services: core, payments, and agentic AI working as one,” Zandona said.
According to Mambu, its agents can connect directly into the ledger, operate according to defined guardrails and explain their decisions, with institutions retaining control over configuration and approval requirements.
Core banking could become an AI battleground
The launch also points towards a potentially significant change in competition among banking technology providers. Core banking platforms have traditionally competed around scalability, product configuration, cloud deployment, APIs and the speed with which institutions can launch new products.
Agentic capabilities could add another dimension. Banks may increasingly evaluate core platforms according to how effectively they allow AI models and agents to interact with financial data and execute controlled actions across accounts, lending and payments.
That does not mean banks will hand autonomous control of critical processes to AI systems immediately. Adoption is likely to progress through tightly defined use cases with clear approval boundaries, particularly in regulated or financially sensitive activities.
However, as agentic AI matures, the ability to connect intelligence securely to systems of record could become an increasingly important part of core banking architecture.
What it means for the industry
- AI is moving deeper into banking infrastructure: Agentic systems are beginning to connect with core platforms and payments rather than remaining standalone productivity tools.
- Core banking competition may increasingly include AI architecture: Connectivity, governance and agent capabilities could become differentiators alongside APIs, cloud deployment and product flexibility.
- MCP could simplify AI integration: Standardised ways of providing models with banking context may reduce the need for banks to create separate integrations for every agent or AI application.
- Autonomy will depend on governance: The ability of agents to take action makes permissions, approval thresholds, explainability and audit trails increasingly important.
- Modernisation and AI strategies are converging: Banks may find it difficult to separate decisions about core infrastructure from decisions about how they intend to deploy AI at scale.
Article Source: Mambu
