FCA Mills Review Sees Agentic AI Run Retail Finance by 2030

The FCA’s Mills Review puts a date on a structural shift most financial services firms are still treating as distant: by 2030, agentic AI will move retail finance from assisted decision-making to autonomous delegation. That forecast, from a major regulator rather than a vendor, changes the compliance and governance calculus for firms building or procuring AI systems now.

What Agentic AI Actually Does

Current robo-advisors automate within fixed parameters. Platforms like Schwab Intelligent Portfolios or Wealthfront’s Automated Bond Portfolio rebalance portfolios, harvest tax losses and allocate assets according to predefined rules. They execute what they are told, once the user has set the parameters. Agentic AI operates differently. Rather than following an “if-this-then-that” logic, it pursues a goal, say, maximise retirement savings by 2050, by breaking that goal into sub-tasks, executing them across multiple systems and adapting as conditions change.

An agentic financial system might identify a rebalancing opportunity, execute the trades, adjust future contributions based on real-time market data and flag a tax implication, all without a direct instruction for each step. The Mills Review describes this as a shift from “assistance to delegation,” with agentic systems taking on “longer tasks and more actions within firm and consumer workflows.” That is a materially different liability and oversight profile from anything current regulatory frameworks were built to handle. The Review frames this around four systemic shifts it expects by 2030, the transformation of firms, new consumer journeys, reshaped competition and amplified financial crime and sets out seven priority recommendations for the FCA Board to act on.

Risks for Individual Investors

The risks are less obvious than they appear. When an autonomous agent makes a complex, multi-step financial decision, the reasoning behind it may not be visible to the investor. The Mills Review explicitly flags this opacity problem: misplaced trust in an opaque algorithm can produce unforeseen losses, and consumers will both be shaped by AI outputs and will themselves influence markets through changed behaviour. Transparency mechanisms are not a regulatory nicety here, they are a practical requirement for informed consent.

The Shrinking Role of Human Planners

As agentic systems absorb more execution tasks, the human financial planner’s role narrows toward areas automated systems still cannot handle: managing psychological biases, navigating family financial dynamics, estate planning and contexts where a client’s emotional attachment to an asset matters as much as its tax profile. An AI agent can identify the most efficient way to liquidate a holding. It cannot weigh what that holding means to the person holding it.

The Financial Services Skills Commission estimated in a May 2026 report that between 30% and 50% of tasks in most financial services roles will see significant automation over time, and that figure covers generative AI already underway. Agentic AI sits on top of that. For planners, the strategic question is not whether to compete with automated execution but how to reposition around judgment calls that automation cannot replicate.

Accountability When Things Go Wrong

The governance question is the hardest one the Mills Review raises. When an AI agent makes an investment decision that produces a loss, the liability chain across investor, developer and deploying institution is not settled by existing frameworks. The review states directly that as firms move toward greater delegation, “questions of accountability, governance and oversight become increasingly important,” and that capable models “still require controls for reliability, consistency, explainability and accountability, alongside human oversight.”

The Architecture Behind Delegation

Agentic systems are not single models. They require orchestration frameworks that chain multiple AI components, connect to external data sources and interact with live services. A retirement planning agent might need to coordinate a market forecasting model, a tax optimisation engine and a risk assessment layer, reconciling their outputs in real time. That architecture introduces failure points and dependency risks that a single-model deployment does not. For financial services specifically, this means building the data layer and API standards that allow agents to interact securely with payment systems and brokerage platforms, not just process information, but act on it. The data pipeline is as load-bearing as the model itself.

Today’s Tools vs the 2030 Horizon

The gap between current AI applications in personal finance and the agentic future is wide. Today, tools like Monarch and Cleo classify transactions, flag anomalies and surface insights, tasks that remain human-initiated and human-concluded. Fraud detection models run continuously in the background, monitoring transaction velocity and geolocation in milliseconds, but they alert; they do not act.

An agentic system does not wait to be asked. Origin’s AI Advisor already moves beyond expense categorisation to evaluate how rising costs affect savings rates and delay financial targets, a step toward the proactive, multi-domain decision-making that defines agentic behaviour. The compliance distinction matters: a system that informs sits under one set of obligations; a system that acts autonomously sits under another, and most current frameworks were written for the former.

The Regulatory Path to 2030

Regulators are moving, but the distance between current frameworks and what agentic deployment requires is significant. The FCA’s Mills Review is the clearest signal yet from a major financial regulator that existing rules are likely insufficient. In the US, an Executive Order dated June 2, 2026, directed federal agencies to promote AI innovation while addressing national security risks, calling for expanded AI-enabled cybersecurity tools and a voluntary framework for engaging with frontier AI models. Voluntary frameworks and regulatory reviews are starts; binding requirements for agentic systems in financial services have not yet materialised in either jurisdiction.

The infrastructure is being built faster than the rules governing it.