Agentic finance

Autonomous AI agents that can plan, decide, transact, and spawn sub-agents are entering the economy. These are not algorithmic trading systems, robotic process automation, or smart contracts operating within parameters fixed at the point of design. Autonomous agents pursue goals across contexts, form economic relationships that emerge without human direction or anticipation, and, as early experiments in products and financial markets show, develop their own strategies that challenge existing financial systems. Agents differ from conventional borrowers in ways that break existing frameworks: they currently lack legal personhood and attachable assets; their behaviour is simultaneously data-rich and deeply opaque; they operate at machine speed; they exhibit correlated failures through shared foundation models; and they create recursive delegation chains, each link further from human oversight, compounding information asymmetry. The regulated financial system, as currently constituted, is too slow and too constrained to serve agent-driven credit demand at the scale and speed at which it is likely to emerge. Without purpose-built infrastructure, shadow channels will emerge by default: agent-to-agent resource flows that perform functions equivalent to financial intermediation but remain invisible to regulators, generating systemic risk, enabling financial crime at machine speed, and creating predatory lending at scale. Previous expansion of economic participation, from consumer credit to shadow banking, outran existing financial infrastructure until purpose-built architecture caught up. The pace and opacity of agentic finance compress the window for building that architecture before costly failures occur.

The sustainability governance architecture built over the past decade, comprising disclosure frameworks, taxonomies, stewardship codes, and fiduciary duties, assumes human decision-makers at the point of capital allocation. Agentic finance challenges that architecture through three mechanisms. First, principal-agent monitoring breaks down: agent reasoning is opaque at scale, and the standard tools for aligning agent behaviour with principal objectives, monitoring, bonding, and reputation, were designed for humans making observable decisions at human speed, and fail when applied to autonomous software making millions of opaque decisions faster than any audit cycle can follow. Second, specifying sustainability objectives for agents is harder than it appears: principles-based regulation depends on interpretative oversight that, however imperfect for humans, ceases to function at the scale and speed of autonomous agents. Third, unregulated agent-to-agent financial channels emerge outside supervised infrastructure, with no sustainability constraints. If governed well, agents operating at machine speed could screen investments against sustainability criteria more consistently than human decision-makers, apply rules uniformly across global portfolios, lower the transaction costs of sustainable investment, and create opportunities for greater impact. If governed poorly or not at all, agentic finance could accelerate the allocation of capital to unsustainable activities at a speed and scale that human-designed enforcement cannot match. The question is not whether agentic finance will emerge but whether the research community, regulators, and policymakers will shape it for societal benefit.

The Initiative is pursuing several research streams:

• Agentic finance and sustainability. Defining the concept, establishing the mechanisms through which agentic finance challenges sustainability governance, and developing responses.
• Monitoring and tracking. Developing frameworks and methodologies for measuring agentic finance flows, producing regular state-of-the-field assessments, and making the case to regulators that monitoring these flows is an urgent priority.
• Regulatory engagement and convening. Direct engagement with financial regulators and policymakers to advance proactive architectural design.
• Experiments and simulation. Empirical and simulation work on whether and how agents trade with each other, correlated agent losses, and emergent credit networks.
• Public agentic finance. The role of public financial institutions, central banks, development banks, and sovereign wealth funds in developing their own agentic finance capabilities, setting specification norms, and providing observable, regulated infrastructure that pre-empts shadow channels.
• Underlying data, protocol, and infrastructure layer. The role of distributed ledgers, smart contracts, alternative data, and improvements in data quality, consistency, and assurance in supporting agentic finance.

Work on the financial dimensions of the agentic economy is at an early stage. Current approaches are starting to recognise the challenge of agents as autonomous economic actors, but the work remains scattered across disciplines and has yet to be consolidated into a coherent analytical framework for financial intermediation. The Agentic Finance Initiative is working to fill that gap: to establish the intellectual foundations, build the evidence base, inform the policy response, and train the next generation of practitioners. The Group brings interdisciplinary research strengths, established policy networks, and a track record of work on the financial system to bear on this question, with the aim of contributing to the design of the financial architecture for the agentic economy and the evidence base on which sound regulation and risk management depend.

• Sustainable Finance Theme at The Alan Turing Institute
• State of Agentic Finance reports