Over the past year I’ve spent a lot of time reflecting on my journey in digital banking — from one of the first internet banks, to building world-class mobile experiences, to scaling platforms globally. And more recently, of course, on the impact AI is already having — and will have — on financial services. I’ve always been fascinated by the intersection of technology and financial systems — because every technological wave rewrites the economics of the industry. But what we’re seeing now isn’t another incremental rewrite. It’s the first time I’ve seen execution itself become programmable.

Since leaving Galileo last fall, I’ve had hundreds of conversations — with CEOs, boards, regulators, CIOs, data scientists and engineers. The same question comes up again and again: is this just another tech cycle? Or is this different? For me, the answer is clear. It’s not generative AI that will fundamentally reshape the enterprise. It’s Agentic AI. Because generative AI still waits for instructions. Agentic AI doesn’t. It plans. It coordinates. It executes. It verifies. Within guardrails. Under supervision. At scale. And when systems can act — not just advise — you’re no longer improving work. You’re redesigning how work gets done.

We’ve spent the last two years being dazzled by large language models. They write, summarize, code, generate. They are extraordinary tools for knowledge work. But they are still tools. What’s emerging now is something structurally different: directed intelligence that can assess, test, approve and execute tasks within defined guardrails. Not copilots whispering suggestions. Not chat interfaces producing drafts. Agents that act. This is the shift from intelligence as advice to intelligence as execution capacity. And when execution becomes programmable, measurable and governed, the operating model of the enterprise changes.

For 30 years, digital transformation has largely been about distribution. ATMs. Internet banking. Mobile apps. Cloud. Each wave improved access and lowered marginal transaction costs. The story of banking technology adoption shows that every cycle compressed cost and expanded reach. But inside the enterprise, a stubborn truth remained: growth was still tightly coupled to people. More customers? Hire more service reps. More loans? Hire more underwriters. More compliance complexity? Hire more risk analysts. Digital improved the front door. Humans still operated the engine room.

Agentic AI changes that equation. If you can 3–5x execution capacity per team, if you can free up 20–40% of human time from low-value orchestration, if unit cost per workflow drops 60–80% — then revenue and operating expense begin to decouple. Efficiency ratios fall. Cycle times compress. Risk consistency improves. New product experimentation becomes viable again. This is what we mean by elastic execution capacity — the ability to scale output without linear cost expansion. That is not incremental automation. That is structural leverage.

Let’s be clear about something important: this is not about replacing people, more about amplifying them. In regulated industries especially, humans remain accountable. Agents operate within policy, approvals and embedded controls. The model is not autonomy without oversight. It is governed execution capacity. Think about commercial lending. Today, analysts gather data, associates extract financials, credit teams draft memos, risk reviews documentation, compliance verifies adherence, and operations books the deal. Multiple handoffs. Re-keying. Manual interpretation. Latency everywhere. Now imagine the reality of multi-agent orchestration: an extraction agent ingests structured and unstructured documents, a financial logic agent analyses covenant ratios, a compliance agent validates policy thresholds, a drafting agent produces the first credit memo, and a control layer traces every step, every decision, every input. Humans review, challenge, approve and build relationships. But the orchestration, coordination and documentation burden collapses. Processing time drops 60–80%. Consistency improves. Throughput multiplies. That is how Agentic lands: not as a chatbot — but as an execution fabric across the enterprise.

Here’s what worries me more than model hallucinations. Uncontrolled proliferation. Shadow AI. Fragmented pilots. Unmanaged token spend. No shared institutional memory. No traceability. In regulated environments, unmanaged agents introduce systemic risk. Without a control layer, you don’t get leverage — you get entropy. The future enterprise cannot run thousands of autonomous agents without policy enforcement, audit trails, supervisory transparency, and clear accountability structures. Agentic AI must be institutionalized — not improvised.

If you study the history of banking technology, one pattern is undeniable: adoption cycles are accelerating. ATMs took years to scale. Internet banking took less. Mobile went mainstream in under a decade. Generative AI reached 100 million users in months. The friction of distribution is collapsing. Which means competitive gaps will open faster than ever before. Organizations that move from experimentation to production deployment — safely — will compound advantage. Those that hesitate will not just lag in productivity. They will structurally disadvantage themselves. Because once your competitor can launch products faster, resolve customer issues in half the time, reduce cost per transaction materially, and reallocate talent toward innovation instead of administration — you are no longer competing on the same economic curve.

Several CEOs I’ve spoken with recently have said a version of the same thing: “We need to pause and rethink from first principles.” They feel the weight of legacy structures. Matrixed reporting. Siloed systems. Process built on process. Agentic AI forces that first-principles reset. If execution capacity is no longer constrained by human throughput, then: how should we design teams? How do we measure productivity? What is a manager’s span of control when they are leading humans and agents? How do we reward value creation when capacity expands non-linearly? This isn’t about inserting AI into yesterday’s org chart. It’s about redesigning the operating model around hybrid human–agent teams. That requires new KPIs (cycle time, autonomy %, agent reuse rates), new governance layers, new cultural norms, and yes, new leadership muscle.

Done right, Agentic AI is profoundly positive. It can remove drudgery, increase consistency, reduce operational risk, lower cost-to-serve, free capital for innovation, expand access to financial services, and increase shareholder value. But it will not happen by accident. It requires architectural discipline. It requires governance. It requires courage to redesign. And it requires speed — because the adoption curve is not waiting. Over the coming weeks, I’ll share more about how I believe this next phase of digital banking — and enterprise operating models more broadly — will land. And yes, I’ll share more about what we’re building. Because this isn’t just another AI cycle. This is the moment execution itself becomes programmable. And when execution becomes programmable, everything changes. Thank you.