Agentic AI in Finance
Year: 2026
Type: article
Abstract: This working paper examines how agentic AI (systems that plan, call tools, retain state and act within defined constraints) is being applied across financial services, and what it takes to deploy it responsibly. Ten chapters cover the shift from generative to agentic AI; the building blocks of agentic systems (agent architectures, large language models and reasoning, tools, memory, orchestration and knowledge graphs); customer-facing agents in retail banking; fraud, compliance and risk management; investment management and capital markets; operations and back-office transformation; building, deploying and scaling agents; the regulatory landscape and ethical guardrails; case studies and lessons learned; and the outlook for the coming decade. The case studies include the authors' own prototypes: an agentic Monte Carlo Value-at-Risk simulator, a cross-jurisdiction regulatory comparison tool and a continuous policy gap-analysis loop. First drafted in 2025, the paper was reviewed and updated in November 2026. Statistics, case studies and statements about law and regulation were checked against current sources (position as of 13 November 2026), and external claims that could be traced are tied to a numbered list of 62 references. A Review and Update Note sets out what was corrected and added. Nothing in this paper is legal, regulatory or investment advice.
Authors: Sharma, Suyash, Shukla, Aakanksha
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7233575486
FWCI:
Cites:
Cited by:
Related to:
