Banks Move AI Agents From Experiments Toward Daily Work

Banks are moving AI agents into more operational workflows while keeping tighter controls around customer-facing and high-consequence decisions. Reuters reports activity across wealth management, client vetting, trading, treasury, and internal operations. The reporting includes plans for digital assistants that support advisers and interact with clients, while retaining human oversight for critical functions.

KPMG's banking survey shows 51% of respondents piloting AI agents, while smaller groups are exploring, orchestrating, or scaling them. The underlying KPMG survey captured responses from 204 U.S.-based C-suite and business leaders at organizations with annual revenue of at least $1 billion. The survey also identifies data readiness, agentic-system complexity, human oversight skills, workforce resistance, and cost literacy as deployment challenges.

Reuters reports banks moving agents into wealth, client vetting, trading, and treasury workflows, with customer-facing tests retaining human oversight. Morgan Stanley described planned client-facing assistant tests connected to adviser workflows. BNY described digital employees with system identities, assigned tasks, and human managers. UBS described agents that assemble internal information and flag adviser actions. These examples differ in autonomy and risk, so they should not be treated as one deployment category.

A pilot count measures experimentation, not durable productivity or safe autonomy. It does not show how many workflows reached production, how often employees override outputs, whether client outcomes improved, or whether the agent's operating cost is lower than the manual process. The survey's deployment challenges also indicate that data access and governance remain part of the bottleneck.

LDS recommends separating pilot activity from production value by tracking task completion, human overrides, escalation quality, latency, incidents, and audited business outcomes before an autonomous agent receives broader system access. Banks should also give each agent a unique service identity, narrowly scoped permissions, immutable action logs, spending or transaction limits, and a named human owner.

The practical signal is not that banks have solved autonomous work. It is that agent governance is becoming an operating-model question. The strongest deployments will be the ones that make autonomy measurable and reversible rather than merely adding a conversational interface to existing systems.