“I wrote "A Blueprint for the AI-Native Lender" as a report earlier this year. But reports sit in inboxes. This series is my attempt to make the same argument in a format that travels further - broken into three parts, written directly to the people running mortgage operations across Europe. The ideas are the same. The urgency hasn't changed.” - Geert van Kerckhoven

Every few years, someone declares that AI will transform banking. Most of the time, it doesn't - not because the vision was wrong, but because the technology wasn't ready.

This time is different. And if you're running a mortgage operation in 2026, you already feel it.

The models have crossed a threshold. Not in one dimension - in all of them simultaneously. Reasoning, vision, multimodality, tool use, code generation, legal text interpretation. Each capability was interesting in isolation. Together, they're transformative. Because for the first time, we can build AI systems that don't just answer questions or classify documents - they deliver outcomes. They reason through edge cases. They know your processes. They adapt to circumstances rather than following predefined workflows.

That's what we mean by agentic. Not a chatbot. Not an automation script. Not a thousand if-then rules stitched together. An agent is more like a thousand workflows distilled into a single intelligence - one that handles the variability, the exceptions, the judgment calls that previously required a human sitting at a desk.

The implications for mortgage origination, credit decisioning, and loan lifecycle management are profound. Over this three-part series, I'll lay out a front-to-back blueprint: where the opportunities are, what patterns are already emerging, and how a truly agentic lender should think about redesigning its value chain. We'll start where disruption will be most visible - the front office.

The AI-Assisted Advisor

The most immediate application is augmenting human loan advisors. Picture an advisor sitting with a borrower, backed by an agent that has already ingested the lender's full risk policy framework, current rate structures, and regulatory requirements. The advisor asks questions; the agent surfaces tailored, policy-compliant recommendations in real time.

This is already happening. In the US, platforms like Addy AI train on lender-specific guidelines - Fannie Mae, Freddie Mac, non-QM - and provide real-time analysis during borrower conversations. Frost Bank, which re-entered mortgage lending with technology at its core, ended 2025 with $595 million in unpaid mortgage balance, exceeding its goals by 19%. One credit union reported its AI assistant resolved over 90% of customer inquiries instantly, cutting support calls by 20%.

But here's the thing: if the agent can deliver that advice in real time, calibrated to policy, across every product - do you need the advisor in the same form at all?

The Borrower-Facing Agent

That question leads directly to putting the agent in front of the borrower. An AI model - available 24/7, in any language, infinitely patient - that guides a borrower from initial inquiry through affordability assessment, document collection, and all the way to pre-approval. Not a form with a chatbot bolted on. A genuine conversational experience that understands context, answers specific questions, and moves the process forward.

Infosys has described this as a "Conversation Agent" - an always-on assistant that serves as the borrower's single point of contact from rate shopping through post-close. ServisBOT already deploys AI borrower assistants that manage servicing interactions via voice, chat, or email, assisting through pre-qualification and application. In March 2026, Mortgage Automation Technologies launched BIG AI - a ChatGPT-like interface purpose-built for mortgage, supporting natural language queries, amortization schedules, and client proposals, powered by the lender's own origination data.

Only 9% of American consumers now prefer branch visits. 55% use mobile banking apps as their primary channel. We don't think Europeans are far off. The direction is clear: borrowers want digital-first, and agents can deliver an experience that's not just digital, but genuinely intelligent.

When Borrowers Bring Their Own Agents

Now here's where it gets really interesting - and where most banks stop thinking too early.

People are already using ChatGPT and Claude to advise them on mortgages. This isn't speculation. A LendingTree survey found 51% of respondents turn to AI for financial advice, with 30% saying AI influenced their decision to open or close a financial account or loan. A Veterans United poll showed 32% of homebuyers use AI tools, and 22% specifically use them to compare mortgage lenders. Among Gen Z, 74% have used an AI chatbot at least once.

And it's not just consumers. 82% of mortgage brokers have used ChatGPT or other AI tools within the past three months. Nearly half use AI daily.

We're not a year away from consumers using agents to shop for mortgages. We're already there.

Now extrapolate. Today, a consumer asks ChatGPT for general mortgage advice. Tomorrow - and "tomorrow" may be months away, not years - the consumer's agent contacts your bank directly. It queries your rates. It submits a pre-qualification request with the borrower's financial profile. It compares your offer against three competitors. It negotiates. All machine-to-machine.

As a bank, how do you respond to that?

The MCP Moment: Building Machine-Readable Interfaces

We believe banks need to start thinking about programmatic interfaces for external agents - the equivalent of what APIs did for open banking, but for AI-to-AI interaction.

The infrastructure is emerging. Anthropic's Model Context Protocol (MCP), introduced in late 2024 and donated to the Linux Foundation's Agentic AI Foundation in December 2025, provides an open standard for connecting AI agents to external systems. Google's Agent-to-Agent (A2A) protocol offers another pathway. Both are seeing rapid adoption - hundreds of tool providers and major platforms have integrated MCP within its first year.

In banking specifically, Block (Square) has integrated MCP into internal payment workflows. Research shows banks implementing MCP-like protocols can give AI agents real-time access to transaction data, enable dynamic risk assessment, and integrate with existing systems - all while maintaining OAuth 2.1 security compatible with PSD2, BSA/AML, GLBA, and EU DORA requirements.

And the regulatory enablement is coming. PSD3, which reached provisional political agreement in November 2025, standardizes Open Banking APIs as the main access point for data exchange, with stronger governance and customer dashboards for permission management. Targeted applicability: Q2/Q3 2028. The regulatory infrastructure for agent-to-bank communication is being built in parallel with the technology.

What we expect to see - and what forward-thinking banks should prepare for:

Agentic mortgage brokers. Not human brokers using AI tools, but AI-native brokerage platforms that contact banks for quotes programmatically, compare offers across dozens of lenders in seconds, and present borrowers with optimised recommendations. The pattern is clear: 82% of brokers already use AI, adoption is accelerating, and the economic logic of replacing human intermediation with agent-mediated comparison is overwhelming. Whether this emerges as standalone "agentic broker" platforms, as features within consumer-facing AI assistants, or as specialist marketplaces remains to be seen - but the direction is unmistakable.

Consumer agents negotiating directly with bank agents. This creates a new interaction paradigm: what information do you want to disclose to an external agent? What don't you? How do you price competitively when every offer is compared algorithmically in real time? Banks that build these interfaces early - with appropriate compliance safeguards - will have a structural advantage in a world where the borrower's first touchpoint isn't a website or a branch, but an agent acting on their behalf.

The banks that dismiss this as futuristic will find themselves scrambling when a significant share of mortgage inquiries arrives not from browsers, but from agents.

‍

‍

Next in this series: I'll look at where AI delivers the most immediate ROI - the mid-office and back-office - and why the most transformative implications in lending may have nothing to do with origination at all.