From pilots to agents: How the second wave of AI is transforming asset management

An inflection point

Asset management is in the midst of an inflection point with generative artificial intelligence (GenAI). Traditionally, the industry has favored caution, a stance that can often clash with the uneven results of initial GenAI pilots. However, macroeconomic conditions that once buoyed asset management have shifted, impacting the bottom line. There is now an imperative for the industry to lay the groundwork for more durable, forward-looking, integrated strategies. Although technology spend in the industry has grown, 60 to 80% of that is dedicated to maintaining legacy systems rather than revenue-generating transformation. Efforts around AI will need to reflect not only technical feasibility but also alignment with the investment goals, operational models, and governance standards expected in institutional finance.governance standards expected in institutional finance.

We are now seeing the data confirm what early adopters intuited. KPMG estimates global market spend on agentic AI reached $50 billion in 2025. Wolters Kluwer reports that 44% of finance teams will use agentic AI in 2026, representing an increase of more than 600% year on year. And BCG’s 2026 Global Asset Management Report finds that agentic workflows can increase capacity by 55% to 65% and reduce operational costs by around 40%. The playbook for moving from experimentation to production has matured, the partnership ecosystem has expanded, and the firms that moved early are showing measurable return on investment (ROI).

History shows that transformative technologies go through a period of refinement before they soar. The automobile was invented in 1885 but viewed as an unreliable novelty until the introduction of the Ford Model T around 20 years later. Horses — firmly embedded as core to humanity’s transportation, trade, and culture for 3,000 years — were supplanted in just 20 years, relegated to novelty. At the turn of the 21st century, we saw the internet boom that promised to reinvent finance, enterprises, and our daily lives seemingly end in a bust of failed experiments. Left in its wake was the general idea that having a website was equal to having a business model, which left some businesses scrambling for what to do next. Yet just a few years later, the second wave of internet advancements (such as, broadband rollout, data center virtualization, and cloud computing) geared toward enterprises finally did help deliver the transformative change that has led to technology enterprises comprising approximately 40% of the S&P 500.

We are seeing a similar pattern with GenAI in asset management. The first wave of GenAI pilots, typically isolated proofs of concept using generic large language models (LLMs), created excitement but with broadly variable ROI. PwC noted in its 2026 AI predictions that many agentic deployments in 2025 did not deliver much value and that under the hood, many were not using agents in ways that truly mattered. With this second wave, we are now in the “soaring” inflection point of generative AI, which has been refined with enterprise value in mind. AI agents embedded in real workflows can autonomously execute tasks, coordinate with other agents, integrate into existing systems, and replicate and enrich complex work products. Crucially, they are designed to still ultimately operate under human oversight and within guardrails.
 

From isolated pilots to integrated AI agents

GenAI’s introduction to asset management came in two forms:

1. Broad, off-the-shelf copilots and chatbots implemented across the enterprise, which enabled some productivity gains but did not necessarily transform the day-to-day[M(1]

2. Use cases specific to business functions but often conducted in fragmented pilots with teams operating in isolation and within the confines of legacy processes and tools
 

Agentic AI offers the opportunity to reinvent and enhance the investment process by embedding intelligence directly into the fabric of investment operations. AI agents are not based on static algorithms; they are autonomous digital co-workers that can perceive, plan, and act within defined bounds. These agents carry out multi-step tasks such as monitoring data feeds, analyzing patterns, and drafting reports without needing a human prompting them at each step. They can also call APIs, query databases, or trigger alerts. In practice, this means an agent can continually scan news and market data and proactively alert the team when a relevant risk signal emerges as opposed to a passive dashboard that is only useful when someone actually remembers to view and derive insights from it. Unlike black-box LLMs, AI agents offer traceable, auditable, sourced outputs and can be regulated with access controls and human-in-the-loop (HITL) oversight, which is particularly important for industries like asset management.

Goldman Sachs Asset Management highlighted an important distinction in May 2026: Agentic AI should not be viewed merely as the final step in a linear automation journey. Instead, it represents a discrete capability leap. Organizations facing high-variance challenges (tasks that are unpredictable, inconsistent, or subject to frequent change) can bypass earlier phases of rigid automation to directly deploy agentic systems capable of handling exceptions and rationalizing numerous decision points.
 

Use cases with direct investment relevance: Where is this technology proving its worth?

Early warning and monitoring: Agentic AI is elevating risk management from periodic checkups to continuous oversight. Consider regulatory compliance and operational risk: Instead of solely relying on manual reviews or waiting for quarterly reports, AI agents can constantly audit transactions and communications in the background. They flag anomalies or potential noncompliance issues as they unfold, providing risk teams with the tools to intervene early. For instance, an agent monitoring portfolio trading patterns might catch an unusual spike in exposure to an emerging market currency and instantly alert risk officers if it breaches set limits. BCG’s 2026 report provides concrete evidence: In private markets, AI agents can now process capital calls, interpret unstructured documents, and calculate and validate waterfall distributions autonomously. In public markets, they can reconcile net asset values across custodians, process corporate actions, and differentiate true exceptions for escalation.

Research and analysis acceleration: Perhaps the most immediate impact of GenAI agents is boosting research team productivity. Across equity, fixed income, and multi-asset teams, analysts spend an inordinate amount of time gathering information — reading financial statements, transcribing earnings calls, scouring economic data — before they can even begin to generate insight. Generative AI agents can extract financial metrics; summarize earnings calls; and collect macroeconomic indicators, sourced research reports, and credit assessments, allowing users to devote more attention to insights and decision-making. We are seeing rapid acceleration in adoption across the industry: JPMorganChase has democratized self-service access for 200,000 employees to its LLM Suite in less than a year, with half using it three or more times a day.

Stress-testing and quality assurance: As agentic AI moves into production, the industry is developing dedicated infrastructure to verify and improve agent performance systematically. Franklin Templeton, which oversees more than $1.5 trillion, has partnered with Sentient to develop Arena, a production-grade stress-testing environment that feeds agents incomplete information, ambiguous instructions, and conflicting data, replicating the reality of corporate workflows. Rather than scoring whether a tool generated a correct output, the platform records the full reasoning trace to help engineering teams debug failures over time. This marks an important shift: The conversation has moved from whether agents can perform to how we verify and continuously improve that performance.

The use cases and applications are growing, but the pattern is clear: Agentic AI assists users with finding alpha signals and risks earlier, potentially creating more bandwidth for humans to focus elsewhere.
 

From prototype to production: Integration, oversight, and architecture

If agentic AI is the engine, the operating model is the chassis required to realize value. To truly go from pilot to production, asset managers must focus on the critical enablers: workflow integration, human oversight, data architecture, and regulatory readiness. These are the fundamentals that largely determine whether AI remains in pilot purgatory or becomes transformative infrastructure.

Workflow integration: The highest ROI from AI will come when it is embedded into existing employee workflows. For example, if a portfolio manager works in Excel, an AI agent should be able to align with that environment. This requires close cross-functional collaboration between teams to redesign processes. Model Context Protocol (MCP), an open standard that allows agents to securely connect data across disparate systems, has emerged as a critical piece of this interoperability infrastructure. Forrester predicts 30% of enterprise app vendors will launch MCP servers in 2026, signaling a more interoperable future where agents from different vendors can work together seamlessly. This is already happening in practice: At Moody’s made  MCP servers available across partner platforms, providing customers with access to decision-grade intelligence directly inside the environments where they already work, whether that is Microsoft 365 Copilot, Claude, AWS, Databricks, or Salesforce. to decision-grade intelligence directly inside the environments where they already work, whether that is Microsoft 365 Copilot, Claude, AWS, Databricks, or Salesforce.

Human oversight and accountability: A governance framework is essential. This means defining which actions require human approval or human quality assurance. Every AI output should be auditable, traceable, and sourced. AI agents should have appropriate access controls to data and internal systems, similar to how access controls and privileges are set for human users today, with data minimization and privacy standards enforced and high-impact actions requiring additional signoff. Deloitte’s 2026 guidance recommends safeguards including agent control rooms, real-time auditing, action logging, kill switches, and human override capabilities. PwC adds that firms should follow the 80/20 rule: Technology delivers only about 20% of an initiative’s value while the other 80% comes from redesigning work so agents can handle routine tasks and people can focus on what truly drives impact. This can help mitigate risks and provide transparency for regulators and teams alike that AI is augmenting the decision-making process, not making judgments itself.

Data architecture and infrastructure: Agentic AI’s effectiveness is only as good as the data it can access. Siloed data systems severely limit AI’s efficacy because agents cannot form a holistic picture to make informed decisions. The new wave of AI adoption is forcing a long-needed investment in unified data architecture, including common data lakes or warehouses, standardized data taxonomies, and real-time data pipelines.

Additionally, connectivity infrastructure is key: Agents need secure access to internal systems and external APIs. For example, a portfolio-monitoring agent might need to pull data from an internal risk system, query an external news API, and write an alert in an email. Ensuring those connections are reliable and governed (with proper permissions) directly affects whether AI delivers value. AI agents can analyze swaths of structured and unstructured data simultaneously to find patterns that humans would miss, but it can only do so if the data is there and accessible. Firms that treat data as a strategic asset by cleaning and linking it are more likely to see success in their AI initiatives.

Regulatory readiness: The regulatory landscape has crystallized significantly. The EU AI Act becomes substantially operational on August 2, 2026, with rules for high-risk AI systems taking effect, transparency obligations becoming mandatory, and each member state required to establish AI regulatory sandboxes. For asset managers deploying agents in areas that touch employment, credit, or critical infrastructure, demonstrating compliance with specific transparency, oversight, and risk management requirements is no longer optional. In the UK, a lighter-touch approach through existing regulatory bodies (such as the Information Commissioner’s Office, the Office of Communications, and the Financial Conduct Authority) continues, although a comprehensive AI bill has been delayed. The firms that have built governance frameworks proactively will likely have a significant advantage over those now working to retrofit compliance.
 

What’s next: From edge case to industry standard

After years of AI promises, why should asset management leaders believe that this wave will deliver? In short, they should believe it because we are seeing the pieces come together with agentic AI. The ecosystem around the technology is more mature, users better understand the implementation playbook, and common patterns have emerged from positive ROI case studies. Generative AI itself has advanced dramatically in just the past two years. Today’s models, architectures, and tools are far more capable of understanding finance-specific content and performing complex reasoning than the LLMs of yesterday. The urgency is also greater now. Asset managers face margin pressures due to changing macroeconomic headwinds.

The results speak for themselves. BCG’s data shows agentic workflows increasing capacity by 55% to 65% and reducing costs by 40%. JPMorganChase has embedded AI access across 200,000 employees. Accenture’s 2026 Banking Top Trends report envisions the rise of the “10x bank” in which a single individual leads a team of AI co-workers to deliver exponentially greater output. AI agents are helping redefine what high-performance teams can accomplish. The firms likely to succeed are those that view AI not as a bolt-on experiment but as a strategic capability — something to integrate across desks, asset classes, and functions.

Moody’s Agentic Solutions exemplify this new era of AI-powered workflows. Built on Moody’s comprehensive data estate covering more than 600 million global entities, with more than 50 domain-specific agents coordinating across credit assessment, portfolio monitoring, sales intelligence, and Know Your Customer screening, it’s where our decades of data and risk experience meets cutting-edge AI orchestration. Through strategic partnerships with Microsoft, Anthropic, AWS, Databricks, Salesforce, and OpenAI, Moody’s decision-grade intelligence is now accessible directly inside the environments where financial professionals already work, supporting teams to achieve deeper insights in a fraction of the time.
 

Interested in learning more? Find out how we can help with Moody’s Agentic Solutions.
 

About the author:

Omar Khan is a Senior Director of GenAI at Moody’s, where he shapes generative and agentic AI solutions and partners with executives in financial services and insurance to translate innovations in AI into measurable business impact. Leveraging Moody’s proprietary data, advanced technology, and award-winning insights, Mr. Khan works with financial services institutions to deploy AI-powered workflow automations that support decision-making, reduce complexity, and perform with precision in regulated, high-stakes environments.