✦ 8-9 October 2026 · Hilton Munich City
AI in banking is moving into a more consequential phase.
The first generation answered questions. The next can retrieve information, initiate workflows, prepare decisions and increasingly act across enterprise systems.
That changes the question. It is no longer only Can the AI do it?
Banks now have to decide what the AI should be allowed to do, under which controls and how every important action can be reconstructed afterwards.
At Digital Banking Summit 2026, MNB Solutions will bring that question into both the panel discussion and the keynote that follows it.
Four banking perspectives on where autonomy should stop
Martin Bačík will moderate a discussion bringing together four different perspectives on AI-powered customer interaction:
Thorsten Seeger
Member of the Management Board
SME Bank, Lithuania
Felim O'Donnell
Director of Financial Crime Operations
Starling Bank, UK
Egill Ingolfsson
Head of Product & Pre-Sales
Meniga, UK
Costis Paikos
Group Chief Digital Officer
Eurobank, Greece
The interesting part is the combination.
Board responsibility, financial crime operations, banking product and group digital strategy all meet once AI starts interacting with customers and acting across processes.
The panel will explore where automation creates value, where human judgement remains essential and how much autonomy banks should actually give AI systems.
When AI Agents Start Acting
Immediately after the panel, Martin will take the discussion one level deeper: from customer experience into architecture, economics and control.
✦ Keynote · When AI Agents Start Acting
Does the agent create value?
Measure work removed, time saved, exceptions and operating cost.
How much autonomy is enough?
Define what it may do alone and where humans approve.
Can you explain every important action?
Reconstruct who acted, on what evidence and through which systems.
An agent that can act needs more than intelligence. It needs boundaries.
An AI agent is not just a chatbot. It is an integration problem.
A chatbot can sit at the edge of the organization. An agent that acts cannot.
The moment an AI system starts retrieving customer information, initiating processes or executing actions, it crosses multiple layers of the bank.
Customer interaction
Chat, email, service requests and digital channels.
Identity & authority
Who is the customer? Who is the agent acting for? What is it permitted to do?
Enterprise data
Customer information, documents, knowledge, transactions and context.
Core systems & APIs
CRM, workflow platforms, document systems, core banking and legacy applications.
Controls & evidence
Validation, confidence gates, approvals, monitoring, traceability and audit trails.
Human operations
Exceptions, escalations, review and responsibility.
A failure in any one of these layers becomes part of the customer experience.
That is why agentic AI in banking is as much an architecture and operating-model problem as it is an AI problem.
What production control actually means
Before an action-taking AI system gets meaningful authority, we want six things to be explicit.
✦ Identity and authority
Who the agent represents and what it may do.
✦ Permission boundaries
Which data, tools and actions it can access.
✦ Confidence and approval gates
Where automation stops and human approval begins.
✦ Traceability
Reconstruct each important action end to end.
✦ Exception handling
Defined fallbacks for ambiguity, failure and outages.
✦ Change control
Test and record every material change.
The goal is not maximum autonomy. It is useful autonomy under control.
Why this is familiar territory for MNB
MNB Solutions has spent years building and operating business-critical systems in regulated environments.
Agentic AI introduces new technology, but many of the hard production questions are familiar: architecture, permissions, integration, monitoring, operational responsibility, security, change control and auditability.
40+
enterprise applications in active production
7+ years
without a major outage across managed solutions
ISO 42001
governed AI management system
100% EU
delivery and GDPR-compliant operations
ISO 27001 · ISO 9001 · ISO 42001
If you are working on customer-service automation, agentic workflows or AI that will eventually be allowed to act inside a bank, we should compare architecture and control models in Munich.