Agentic AI is moving fast in financial services. Governance has to move faster

The financial services sector is rapidly adopting agentic AI, not as a future ambition, but as a live capability. One of the central questions firms face today is whether engineering and governance models are mature enough to support it. Systems that can reason, act and orchestrate multi‑step workflows are already reshaping financial crime investigations, underwriting, claims management, lending and due diligence. But as the latest PwC and Microsoft Agentic AI in Financial Services report shows, the industry’s challenge has shifted. The question is no longer whether agentic AI works. It is whether organisations are engineering it in a way that regulators, customers and boards can trust.

We are entering a new dawn where AI agents are coordinating entire end‑to‑end journeys rather than just supporting isolated tasks. Agents can now gather evidence, validate documents, apply policy rules and trigger downstream actions across multiple systems. In a sector governed by strict expectations around auditability, explainability and accountability, that level of autonomy changes everything.

This is why the report points to the growing importance of what I would describe as an AI “control engineering tower”. Financial services firms need a central control plane that acts as both a system of record and a system of control for agentic workflows. This layer continuously monitors agent behaviour, enforces risk thresholds, maintains end‑to‑end audit trails and ensures that high‑impact decisions are escalated to humans. Without this capability, scaling autonomy risks breaching regulatory obligations under frameworks such as the EU AI Act, DORA and existing conduct and consumer‑protection regimes. At the same time, an overly cautious response risks thwarting adoption altogether, inhibiting firms from realising the full business and customer impact agentic AI can already deliver at scale.

Crucially, this is not just a governance problem. It is a software engineering one. Agentic AI does not fit neatly into delivery pipelines designed for deterministic code and static automation. Organisations are having to redesign how they design, verify, implement and track AI‑enabled workflows that can adapt to new information while remaining transparent and compliant. This is often where early initiatives stall. A proof of concept can demonstrate technical promise, but production systems must withstand regulatory scrutiny and operational stress.

The report is equally clear on what enterprise AI looks like in practice today. The value is not coming from chat interfaces or standalone use cases. It is emerging where repeatable, multi‑agent blueprints are embedded directly into core value chains. In banking, these blueprint patterns are compressing KYC and financial crime processes from days into real time activities. In insurance, they are enabling near‑autonomous claims handling and continuously refreshed underwriting decisions. In capital markets, they are reducing weeks‑long research and due diligence cycles to minutes.

That value only compounds when capabilities are reusable. Frontier organisations are investing in shared data foundations, governed platforms and repeatable blueprints that allow successful agentic patterns to be redeployed across business lines. This is how AI moves from experimentation to regulated, revenue‑driving infrastructure.

Agentic AI is no longer a distant promise for financial services. Success will be defined less by the sophistication of the models, and more by how deliberately we engineer control, trust and scale into the systems that deploy them. Ultimately, this means we need to fundamentally consider the implications of this transformation to the operating model.

Financial Services Programme activities

The techUK Financial Services programme connects tech firms, the FS industry, and regulators to ensure innovation and technology can be fully embraced. Through market engagement activities and events, we help to empower decision makers and aid collaboration.