How can AI impact the finance industry?

Key Concepts

• Agentic AI: Generative AI combined with the ability to take action. It's a "do it for me" technology that sets goals, plans tasks, acts, and adapts to feedback.
• Generative AI: A "create for me" technology that builds upon traditional AI and machine learning.
• Traditional AI/Machine Learning: AI technologies that have existed since the 1950s, primarily effective with structured data.
• Unstructured Data: Data that does not have a predefined format, which is prevalent in financial services.
• Agentic Commerce: The concept of AI agents transacting on behalf of individuals, comparing financial products, and managing funds.
• AI Wallets: Pre-funded, capped wallets for AI agents to transact within a defined budget.
• Automated Authentication Layer: Mechanisms to authenticate AI agents, such as contextual authentication based on chat history.
• Human-in-the-Loop Controls: Processes that ensure human oversight and intervention in AI operations.

Citi Report: Agentic AI's Impact on Finance

Citi has released a report highlighting the significant potential impact of Agentic AI on the finance industry, suggesting it could be even more transformative than the internet era. The report, authored by Sophia, a Future of Finance Analyst at Citi, details the evolution of AI and its implications for financial services.

Defining Agentic AI

Agentic AI is defined as Generative AI plus Action. It represents a "do it for me" technology capable of:

• Setting a goal.
• Planning a task.
• Taking action.
• Adapting to feedback.

An example provided in the financial services context is an agent resolving a payment error. This agent can autonomously navigate unstructured data, cross-check information, identify the mistake, amend the payment, submit it for approval, and log the transaction in the audit trail. This capability is enabled by Generative AI, which is described as a "create for me" technology.

Traditional AI and machine learning, while valuable, have historically been limited by their effectiveness with structured data. The prevalence of unstructured data in financial services makes Generative AI and Agentic AI particularly relevant and impactful.

2025: The Year of Agentic AI

Sophia highlights the sentiment, echoed by Ren Ito (Co-founder and CEO of AI), that 2025 is a "wow year" for AI in finance, comparable to 2024 for Large Language Models (LLMs). This year is significant because several key components have converged:

• Advanced Models: The development and accessibility of LLMs.
• Cloud Infrastructure: Major cloud providers (hyperscalers) have heavily invested in and released tools for agents.
• Enterprise Controls: The emergence of enterprise-level controls and guardrails, moving beyond demos and prototypes into actual use cases.

The conversation at major financial and fintech conferences, such as Money 2020, has shifted to Agentic AI and its practical applications. Significant capital has also been invested in this space, with a noticeable shift from pilot programs to production deployments.

Future Implementation and Impact (2026 and Beyond)

Looking ahead to 2026, Sophia anticipates:

• Wider Deployment: Increased adoption and implementation of Agentic AI.
• Measurable Commercial Impact: The emergence of the first Return on Investment (ROI) stories as the focus shifts from the novelty of the technology to its financial benefits.
• Agentic Commerce: A significant development where AI agents will transact on behalf of users. This includes comparing financial products, optimizing savings by moving funds, or managing mortgage renewals. This will be a game-changer when bank clients begin using AI agents with their financial institutions.

Impact on Careers and the Workforce

The discussion around job displacement due to AI is acknowledged as polarizing. Sophia's perspective, supported by the Citi report, is that:

• Tasks Will Change: The nature of tasks performed in jobs will evolve.
• Increased Productivity: Generative AI and Agentic AI will lead to higher productivity, allowing individuals to focus on higher-value activities. These include tasks requiring oversight, judgment, client engagement, exception handling, and strategic work.
• Shrinking Roles: Some repetitive operational roles, particularly those traditionally outsourced, may decrease over time.
• Emergence of New Jobs: New categories of work and job roles will be created, such as:
  • Agentic AI Ethics Officers
  • Agentic AI Guardrails Officers
  • Workflow Designers
  • Data Quality Managers
  • Governance Enforcers

In essence, Agentic AI is seen as reshaping the workforce rather than simply replacing it. The emphasis is on the need for curiosity and adoption of the technology to adapt to these changes.

Building Trust and Ensuring Accuracy

Addressing concerns about the accuracy and trustworthiness of AI, similar to issues with Generative AI "hallucinations," Sophia emphasizes the need for:

• Robust Guardrails and Governance Frameworks: Implementing similar controls as those being developed for Generative AI.
• Regulatory Scrutiny: Being prepared for regulatory and policymaker scrutiny, referencing frameworks like Europe's AI regulations and guidance from standard-setting bodies.

Venture Capital Funding in Agentic AI

While specific future funding figures were not provided, the report notes that a substantial amount of money has already been invested. The momentum suggests that this investment is unlikely to decrease. The number of deals has also grown significantly, with a 150% increase from 2023 to 2024 in deals involving autonomous agents, digital co-workers, and generative AI for chats.

Agentic AI and the Rise of Malicious Bots

The statistic that bots account for 50% of global internet traffic, with a significant portion being malicious, presents a dual-edged sword for Agentic AI.

• Enabling Fraud at Scale: Agentic AI can be a delivery mechanism for fraud dissemination at scale, as illicit actors can adopt technology rapidly without regulatory constraints and are adept at data sharing.
• Mitigation Strategies: To enable agents to act with sufficient authority without compromising security, the report suggests:
  • AI Wallets: Pre-funded, capped wallets for agents to transact within a budget.
  • Automated Authentication Layer: Implementing contextual authentication, such as tracing an agent's chat history back to the original human prompt.
  • Overarching Governance: Robust governance frameworks, human-in-the-loop controls, and secure enterprise infrastructure are crucial.

The conversation concludes with an acknowledgment of the vastness of the topic and the likelihood of future discussions.