The landscape of investment management is evolving rapidly, particularly with the integration of artificial intelligence (AI). As firms increasingly incorporate AI into their operations, understanding its applications and implications becomes vital. While many organizations utilize AI tools in their daily routines, they often struggle to articulate the specific types of intelligence they employ. This article examines the classification of AI agents and their roles within the investment sector.
As the industry embraces AI technology, recognizing the difference between traditional AI and advanced systems—referred to as agentic AI—is essential. Unlike basic models that provide straightforward answers, agentic AI can observe, analyze, make decisions, and even take actions on behalf of human operators within set limits. Investment firms must classify their AI systems: are they decision-support tools, autonomous researchers, or delegated traders?
Classifying AI agents for effective governance
The adoption of AI necessitates a well-defined framework to set boundaries and ensure proper governance. Without a classification system, organizations may find it challenging to manage and scale their AI capabilities effectively. To address this issue, a collaborative research initiative between DePaul University and Panthera Solutions has developed a multi-dimensional classification system for AI agents within the investment management industry.
This classification framework aims to provide a common language for practitioners, board members, and regulators. By evaluating AI systems based on their autonomy, functionality, learning capabilities, and governance, investment leaders can better understand the processes required to develop an AI taxonomy tailored to their firms.
The risks of lacking a shared taxonomy
A collective understanding of AI classifications is essential. Without it, organizations risk either over-relying on or underutilizing a technology that is fundamentally altering capital allocation. Such misalignment can lead to unforeseen complications. A well-conceived AI taxonomy should not stifle innovation; instead, it should enable firms to define the problems their AI agents address, establish accountability, and mitigate model risk. In the absence of this clarity, AI adoption tends to remain tactical rather than strategic.
Approaches to AI implementation in investment management
Investment managers currently approach AI in two primary ways: treating it as a set of functional tools or integrating it as a systemic part of the investment decision-making process. The functional perspective focuses on utilizing AI for tasks such as risk scoring, employing natural language processing for sentiment analysis, and integrating co-pilots for summarizing portfolio exposures. While this enhances efficiency and consistency, it keeps the decision-making process human-centric, with AI functioning as a supportive element.
Conversely, an increasing number of firms are embracing a systemic approach, recognizing AI agents as adaptive participants in the investment design process. In this model, the roles of autonomy, learning capacity, and governance are explicitly defined, allowing human judgment and machine reasoning to coexist and evolve together. This distinction is crucial, as functional implementations may lead to faster tools, while systemic applications cultivate smarter, more agile organizations.
Achieving balance in decision-making
Neuroscientist Antonio Damasio has noted that all forms of intelligence aim for homeostasis—a state of balance with their surroundings. The financial markets function as complex adaptive systems that require equilibrium between data and judgment, automation and accountability, as well as profitability and sustainability. A robust AI framework should reflect this ecological balance by mapping AI agents across three critical dimensions.
The first dimension is the investment process. Organizations should evaluate where in the value chain each agent operates. The investment process typically involves five stages: idea generation, assessment, decision-making, execution, and monitoring, all of which are intertwined with compliance and stakeholder reporting. AI agents can enhance each of these stages, but decision-making authority must remain proportional to the interpretability of the AI’s actions.
Comparative advantage and the complexity range
The second dimension focuses on the comparative advantage that an AI agent enhances, whether it be informational, analytical, or behavioral. While AI does not inherently create alpha, it can amplify existing advantages. A useful method for taxonomy mapping is to differentiate among three archetypes based on strategic alignment with specific investor skill sets.
As the industry embraces AI technology, recognizing the difference between traditional AI and advanced systems—referred to as agentic AI—is essential. Unlike basic models that provide straightforward answers, agentic AI can observe, analyze, make decisions, and even take actions on behalf of human operators within set limits. Investment firms must classify their AI systems: are they decision-support tools, autonomous researchers, or delegated traders?0
As the industry embraces AI technology, recognizing the difference between traditional AI and advanced systems—referred to as agentic AI—is essential. Unlike basic models that provide straightforward answers, agentic AI can observe, analyze, make decisions, and even take actions on behalf of human operators within set limits. Investment firms must classify their AI systems: are they decision-support tools, autonomous researchers, or delegated traders?1
As the industry embraces AI technology, recognizing the difference between traditional AI and advanced systems—referred to as agentic AI—is essential. Unlike basic models that provide straightforward answers, agentic AI can observe, analyze, make decisions, and even take actions on behalf of human operators within set limits. Investment firms must classify their AI systems: are they decision-support tools, autonomous researchers, or delegated traders?2