Corporate treasury sits at the intersection of every financial decision and every risk a company carries. The tools have improved over the years. The decisions have stayed human. Agentic artificial intelligence is beginning to change that.
AI agents are now performing core treasury functions autonomously, from intraday liquidity decisions in wholesale payment systems to FX exposure forecasting and cash flow optimization. The shift is happening inside central bank research, inside corporate treasury teams across EMEA, and inside the largest banks in the world.
What central bank research found
A BIS Working Paper by Iñaki Aldasoro and Ajit Desai of the Bank for International Settlements and the Bank of Canada tested whether a generative AI agent could perform intraday liquidity management inside a wholesale payment system without domain-specific training. The researchers used ChatGPT's o3 reasoning model across a series of simulated cash manager scenarios.
The paper said the agent closely replicated key prudential cash management practices. When facing two small pending payments and the possibility of a large urgent payment shortly after, the agent chose to delay the smaller payments to preserve liquidity, a strategy consistent with how experienced human cash managers operate. When complexity increased, with probabilistic inflows and competing payment priorities, the agent adapted its reasoning, though consistency dropped slightly as trade-offs became more layered.
The paper tested the agent in an operator mode, running it through a structured set of stylized cash management scenarios. The agent completed the exercise autonomously, responding correctly across routine liquidity prioritization scenarios while deferring to human oversight when it encountered potentially anomalous payment patterns. The paper concludes that routine cash management tasks could be automated using general-purpose large language models, potentially reducing operational costs and improving intraday liquidity efficiency.
What EMEA treasurers are actually focused on
A Treasurer Magazine article covering the JP Morgan EMEA Treasurers Forum, drawing on polling of finance leaders from 26 countries and more than 60 industries, found that AI is shifting from experimentation to targeted implementation inside corporate treasury teams across the region. Sixty-three percent of respondents expect employee productivity to be the area most affected by technology in the next 12 months, with automated reporting at 45% and financial forecasting and planning at 35% close behind. Only 6% said they did not expect AI to have a material impact.
Sara Castelhano, head of U.K., Europe and Canada Treasury Services at J.P. Morgan Payments, said treasury teams will stay focused on resilience, liquidity, and improving cash flow, running stress-tested assumptions to maintain clear visibility into what breaks, when, and what actions to take. She added that AI and tokenization are moving from pilots to targeted rollouts in forecasting, payments operations, and controls, but only scale with reliable data, clear ownership, and solid governance.
For finance leaders looking to build those skills, the AI for Finance training resources cover the practical applications emerging in treasury operations. A more structured option is the AI Learning Path for CFOs, which addresses the governance and implementation questions banks and corporate teams are now confronting.
What the largest banks are already running
A World Economic Forum article covering emerging trends for 2026 found that the banking industry is moving from AI assistance to transactional authority, with autonomous agents integrated as semi-autonomous digital co-workers designed to settle routine trades and manage compliance under human oversight.
Goldman Sachs is developing autonomous agents powered by Anthropic's Claude to handle core trade accounting and client onboarding, with agents acting as digital co-workers that reduce the time process-intensive functions take. Lloyds Banking Group has committed to enterprisewide agentic AI deployment in 2026, expecting the systems to add £100 million in value by automating fraud investigations and complex complaints, diverting routine cases to AI while reserving human staff for the most nuanced escalations.
The shift is structural, not incremental. Banks are no longer asking whether to integrate agentic AI into core treasury operations. They are asking how fast they can move from pilots to production, and what governance needs to be in place before they do.
Why this matters for management
The evidence from central bank research and bank deployments points to a specific conclusion: routine treasury work is becoming automatable, and the teams that adapt will reallocate their people to judgment-heavy work. The BIS paper shows general-purpose models can handle routine liquidity prioritization without domain-specific training. The bank rollouts show production systems are already running.
For treasury managers, the practical question is no longer whether agentic AI works. It is which processes to automate first, what data quality standards to enforce, and how to structure human oversight for the cases agents flag as anomalous. The teams that answer those questions in the next 12 to 24 months will determine whether AI displaces their routine work or frees their people to focus on the decisions that still require human judgment.
Your membership also unlocks:
