Agentic AI in Treasury: From Copilot to Autonomous Execution

What happened

Agentic artificial intelligence is moving from a supporting role in treasury operations to one of autonomous execution — and the financial services industry is beginning to reckon with what that shift actually means in practice. The conversation, surfaced through Finextra's event coverage, centres on how bank and corporate treasurers can harness AI systems that do not merely recommend actions but carry them out independently, end to end.

The framing of "copilot to autopilot" captures the essential transition: earlier AI deployments assisted human decision-makers with analysis, forecasting and alerts. Agentic AI, by contrast, is designed to perceive its environment, set sub-goals and act — managing liquidity positions, executing payments or rebalancing portfolios — with minimal human intervention at each step. The question for treasury professionals is no longer whether this capability exists, but how to govern, trust and integrate it responsibly.

Why it matters

Treasury may seem distant from the customer experience conversation, but it sits at the operational core of every financial promise a bank or corporate makes to its clients. Payment timing, credit availability, foreign-exchange execution and cash-flow reliability are all treasury outputs — and customers feel every failure in those outputs acutely, even if they never see the back-office machinery behind them. When AI agents begin making autonomous decisions in that machinery, the tolerance for error narrows sharply, because the consequences land directly on customer outcomes.

From a behavioural economics perspective, the shift to agentic AI also restructures accountability in ways that matter for trust. When a human treasurer makes a decision, there is a legible chain of reasoning and a person to whom consequences attach. Autonomous agents dissolve that chain. For service designers and CX leaders, this is a signal: the institutions that will win customer confidence in an agentic era are those that make AI decision-making interpretable and contestable — not just fast and efficient.

The Renascence take

Most of the industry debate around agentic AI in treasury focuses on capability and risk controls — the engineering problem. Far fewer practitioners are asking the more important question: what happens to the customer relationship when the entity serving them is no longer meaningfully human?

The real design challenge of agentic AI is not automation — it is the preservation of perceived agency on both sides of the transaction. Customers and counterparties need to feel that someone is accountable, even when no single human made the call. Treasury teams deploying autonomous agents should invest as heavily in explainability interfaces and escalation rituals as they do in the agents themselves. The institutions that treat interpretability as a customer-experience feature — not a compliance checkbox — will be the ones that retain trust when, inevitably, an autonomous system does something unexpected. Autonomy without legibility is not efficiency; it is a latent churn driver.