Beyond automation: Agentic AI and scaling fragmented financial markets

Agentic AI applications go beyond passive assistance and allow for dynamic problem-solving, making decisions and executing tasks with minimal human input, unlocking huge potential across industries. This potential should not be underestimated: When used appropriately, agentic AI solutions will further catalyze and accelerate change in how capital markets are funded and scaled. 

Over the past decade, the rise of private credit has enabled the seismic growth of nonstandardized credit instruments with bespoke or nonstandard contracts. Unlike the broadly syndicated market, private credit has no real market-based framework to standardize documentation terms and conditions. This presents a growing challenge for asset managers grappling with a large and diverse pool of instruments, where smaller tranche sizes make scale difficult to achieve. Here, AI agents can bring a step-change in efficiency: These systems can interpret unstructured data, evaluate nonstandard contracts and autonomously generate decisions that assist portfolio managers in identifying patterns, risks or arbitrage opportunities at scale.

Meanwhile, digital financial markets, such as crypto exchanges, insurance trading platforms and algorithmic trading platforms, pose challenges to investors due to the complex and interrelated workings between smart contracts and the illiquidity that stems from market fragmentation across multiple chains and protocols. In both cases, the variety and complexity of assets contribute to the inefficiency (or lack) of a secondary market, which is key to a digital financial market’s maturation. With agentic AI solutions, firms can move beyond automation and toward intelligent orchestration, optimizing cross-protocol decision-making, asset monitoring and liquidity management to navigate complexity and scale their participation in emerging digital markets.

Agency refers to an entity’s capacity to act independently and make decisions. That entity may be a person, corporation, machine or AI software. AI agents for business embody this concept by operating as autonomous intelligent systems that interact with their environment, collect data and perform self-determined tasks to meet predetermined goals with limited human intervention. Levels of agency range from lower to higher autonomy. 

For agentic AI systems to meet the needs of a fragmented, unmapped private credit universe, the market needs sophisticated AI agents, not just AI assistants. A fully autonomous AI agent is the most self-sufficient form of agentic AI because it can resolve issues through holistic sensing, planning, acting and reflecting. It not only learns from feedback and environmental interaction but also adapts based on real-time data and analyzes its performance for self-improvement with a robust long-term memory. These agents can better act autonomously to explore different scenarios or generate predictions at, essentially, human-level reasoning capability.

AI agents serve different purposes depending on their applications (e.g., consumer, industrial or financial). Execution AI agents already exist in industrial applications, largely based on machine and deep learning, such as preventive maintenance of machinery using telemetry, predicting potential failure that could otherwise lead to costly repairs, defective products, wasted materials, and even health and human safety concerns. As with other AI-related technologies, the integration and adoption of generative AI in many organizations has helped galvanize research and development into how intelligent agents can be used in various value streams, categorized here as automation, wide complexity and deep complexity. 

The financial services industry depends on complex administrative processes for decision-making due to its deep history in regulation and level of documentation. Automating these processes would be a natural evolution, owing to technological advancement and incentives to streamline activities that otherwise compress margins for many organizations. Some 60% of financial services companies anticipate that AI agents will bring the most value in 2025 through task automation, according to a survey by S&P Global Market Intelligence 451 Research. 

In credit underwriting, for example, AI agents could automate data collection, digitization of physical documents (e.g., deeds, titles, legal documents) and analysis (e.g., cataloging digital information for processing and calibration). Decision-making could be automated as well, but measures would need to be taken to reduce implicit bias and ensure compliance with applicable accountability standards. For tokenization, agentic AI’s capabilities could help enable on-chain securitizations, where smart contracts are part of the origination process. The AI agent would be like a digital attorney; it could have a custodial relationship that would allow it to make decisions and execute on the client’s behalf. 

Wide complexity

Modern financial decision-making, particularly for institutional financial products, involves an increasingly large set of complex factors. These range from traditional elements, such as financial, legal and firmographic considerations, to geopolitical, market (interest rates, volatility, pricing), economic, environmental and digital factors. AI agents could distill this information (e.g., indexing) so a portfolio manager, risk analyst or orchestration agent can consider various factors in a vacuum and understand their interplay, with the AI agent providing transparent, reasoned insights that a human can use in their decision. Human oversight of AI agents will be key to fostering a critical thinking and safety-first mindset. Similarly, agentic AI workflows can be used to create products such as index funds with a trading strategy that aims to exploit some of these factors or hedge on others. In a private credit market with many diverse borrowers for which limited information is available, this could be a useful application for agentic AI in finance. 

Derivative products, including options or securitized products, use longitudinal data and analytical processes to estimate future curves and forecasts to better understand expected outcomes, risks and adequate pricing schemes. These processes are less about the number of factors than their nuances (“deep complexity”), which allow underwriters, asset managers or analysts to make recommendations with conviction. In this context, AI agents are “digital research assistants” that can help make such recommendations. For example, in private credit, a “deep complex agent” could understand the nuances of unique complex transactions, such as bespoke structuring terms or multilayered capital structures, interacting autonomously with multiple stakeholders. For tokenization, AI agents could help the manager navigate the complexities of smart contracts. Human involvement — scrutinizing the insights deep complex agents provide and connecting them to real-world context and human-aligned values — will be key. 

Financial markets may benefit from AI agentic systems across functions and actors, including banks, insurers, asset managers, private equity and brokers. AI agents tend to specialize in specific tasks and can interact with other AI agents or multi-agent systems to perform complex workflows that comprise multiple tasks, where orchestration to ensure coordination and control is crucial. AI agents tend to rely on stochastic models, which is a limitation for zero-error risk tolerance situations. For that reason, human judgment and oversight will remain critical. Areas where agentic AI may benefit efficiency, enhance risk management and boost new revenue streams include: 

• Trading and investing-related activities, including investment research and advice, sentiment analysis, algorithmic high-frequency trading, and trading assistants

• Document management tasks, including back- and middle-office operations, automated due diligence, client onboarding, and claims streamlining

• Risk management of liquidity needs and market risk, regulatory compliance, fraud and market manipulation monitoring, supply chain analysis, “what if” scenarios, simulations, and stress test modeling

• Capital flows prediction, including refinancing and capital needs, as well as issuance forecasts and predictive cash flows 

• Transaction optimization, including payments and money transfers, purchase of financial and nonfinancial products, order routing, matching, and surveillance

• New products, such as index funds or dynamic exchange-traded funds that offer proactive risk mitigation to underperforming, asset-volatile sectors and can capitalize on short-term market opportunities

Agentic AI capabilities could offer significant opportunities for financial markets. Deep and wide analysis of information takes time, and increasing operational efficiency can improve productivity and save time for more value-added activities.

Qualitative benefits such as error reduction, consistency and timeliness may similarly add value for practitioners and customers in financial services. Furthermore, agentic AI solutions can improve price discovery, such as adjusting prices in real time based on supply and demand changes, and generate timely predictions to optimize pricing strategies, expand revenue opportunities and improve transparency. By leveraging advanced algorithms and data analytics through well-orchestrated agentic AI systems, firms can make more informed decisions, respond swiftly to market changes and ultimately drive profitability in an increasingly competitive landscape.

However, we expect that agentic AI in finance will take time to gain scale, with challenges surrounding: 

• Financial stability risks: AI agents’ capacity to increase the complexity and opacity of workflows if not correctly managed, as well as their ability to execute transactions at high speed, means they can amplify systemic risks. Agentic AI systems interacting at scale may multiply the speed of execution and spread of contagion in situations of high volatility, disinformation, cyberattacks or market turmoil. 

• Regulatory concerns: For use cases involving trading and investment advice, new products and systems can potentially affect financial stability (e.g., certain risk management and transaction optimization systems). These would most likely fall under the high-risk category of the EU AI Act and may be highly scrutinized by other regulators to ensure investor protection, transparency in decision-making and market integrity. Reporting requirements for counterparty credit risk exposures may need to become real-time instead of daily or weekly. 

• AI governance challenges: The autonomous nature of agentic AI systems in handling confidential information and their potential to make mistakes, make unethical decisions and cause harm (e.g., through hacker AI agents) pose considerable risks in accountability and liability for market participants. AI governance is crucial to mitigate these risks through conscious design and oversight. We expect that financial players will be cautious when scaling agentic AI, as they are ultimately liable for their agents’ misbehavior.

Furthermore, an adverse consequence of AI agents' operational speed and persistence may be reducing the natural latencies of traditional market mechanisms, potentially eliminating some investment opportunities and arbitrage and making some asset classes unprofitable. Guarding against this would require very thorough analysis and process design to promote true agency; this is largely uncommon in AI agents today because processes feeding the agents and managing their implications downstream lack causal foundations. 

Crypto technology such as blockchain, thanks to its immutable and decentralized nature, helps provide solutions to agentic AI in finance and enhances transparency. AI-powered smart contracts, carefully engineered to avoid introducing vulnerabilities (e.g., prompt injection, model inversion and credential leakage), can automate complex processes and enhance efficiency. Furthermore, instant settlements for agent queries in smart contracts can facilitate multi-agent interaction and collaboration.

Regulatory concerns will inevitably grow alongside development of AI agents and crypto solutions, and the blockchains that will likely provide the information infrastructure for them. AI agents could use wallet verification tools or soulbound non-fungible tokens (NFTs) to identify parties involved in a financial transaction, whether directly or via a custodial relationship. These identifiers could function in authentication and security operations. But they have vulnerabilities: While designed to mitigate security risk, they could instead amplify it and contribute to contagion if automated systems are unchecked, creating a vector for decentralized-finance-related cyberattacks. Examples include sybil attacks, routing attacks, logic errors (which could be amplified with agents), key theft/mismanagement and decryption of keys (e.g., quantum safe). Unsurprisingly, regulations and legal oversight are rapidly evolving in crypto technology as AI becomes more intertwined with decentralized systems.

Several key developments are necessary to scale agentic AI applications across financial markets: increased transparency and explainability of agentic processes to build trust among market participants and regulators, adaptive real-time regulation both for AI and crypto technology to ensure accountability and liability, and the interoperability of data and infrastructure to power financial markets. 

The future has never looked this precarious and bright at the same time, and agentic AI may play a key role in shaping this transformation. The disruptive and transformative effects of AI on the financial markets may be a double-edged sword, and as with any innovation of broad scale and utilization, it requires prudent management. The benefits it brings extend beyond efficiency and productivity, but its potential for inspiring further innovation and creativity — enabling and accelerating progress at a rate not seen before — is counterbalanced by its ability to amplify systemic risks of similar magnitude. This irreversible transition to AI in financial markets must be harnessed thoughtfully and responsibly by regulators and market participants alike.