AWS for Industries

Agentic AI for Capital Markets: Building a Foundation for Success

This article was originally published in the Autumn 2025 edition of FinTech BoostUP Magazine.

Agentic AI is an inflection point for Capital Markets firms

AI-powered innovation and automation within capital markets has evolved rapidly. What began with assistants has progressed to agents, and now to fully autonomous, multi-agent systems. These systems mimic human logic and reasoning to execute complex, multi-step workflows. Or has it? Gartner predicts that over 40 percent of agentic AI projects will be canceled by the end of 2027, due to escalating costs, unclear business value, or inadequate risk controls.

From my vantage point as leader of Agentic & Generative AI for Capital Markets at AWS, capital markets firms are starting to appreciate the power and promise of agentic AI but many perceive agentic AI as difficult to implement and deploy due to all the required risk mitigation and quality measures.

This caution can be justified. On the one hand agentic AI can be seen as a natural continuation of a company’s generative AI journey, as previously implemented point solutions (each representing a single agent) can now be deployed in a multi-agent, collaborative fashion. On the other hand, agentic systems are more complex than gen AI assistants or single gen AI agents. They comprise domain-adapted foundational models along with orchestration agents, customized execution plans, and tool use. Moreover, agentic AI is likely to have a more direct impact on operational risk, as agentic AI inherits the risks of the processes that it automates.

In short, agentic AI is not a simple next step for any company, even one that is already successfully employing gen AI. It’s an inflection point that requires management of new components and careful consideration of operational risks.

But, as I’ll explain here, now is a great time for Capital Markets firms to be building a foundation for success in agentic AI. While some of the initial excitement across the industry has been replaced by a more measured focus on practicalities, tools and standards are maturing rapidly and are now capable of delivering agentic AI with the flexibility and security that Capital Markets firms require.

Solving complex, multi-step problems in Capital Markets

Due to the modular, specialized nature of their components, multi-agent systems provide significant benefits over single AI agents, including distributed problem-solving and improved accuracy, scalability, and transparency.

Within Capital Markets, where AI has long been used for analyzing both structured and unstructured data, we’re seeing significant untapped potential for agentic AI to help investment professionals not only gather, analyze, and synthesize information faster, but also to take fresh approaches to long established processes.

For example, an asset manager could use gen AI to process large amounts of information on an investment opportunity, summarize this information in a research report, and provide the basis for an investment recommendation in hours rather than days or weeks. But consider how much further this asset manager could go. Using agentic AI, they could produce similar research reports for all companies in the investable universe. This could extend to any type of investment opportunity. Other agents could identify clusters of similar investment themes. These agents would then create descriptions for each cluster or theme. Finally, they would link each theme to a historical performance proxy. This proxy would be calculated using the performance data from companies within that theme. Then, another category of agents could seek to assess how each investment would influence the portfolio’s positioning, characteristics, risk, and potential performance. The portfolio manager would be able to adjust individual agents and the overall execution plan as their investment objectives, processes, and theses change.

Agentic AI can similarly help other Capital Markets professionals where data-heavy workflows are complex and multi-step, such as product development, trading, financial crime investigations, back-office workflows, investor relations, and sales and marketing.

Agentic AI tools and standards are maturing rapidly

Many compelling agentic AI use cases are still in early phases of development. The ecosystem is maturing rapidly, however, and it’s getting easier to build effective AI agents and deploy them at scale.

First, frameworks such as Amazon Bedrock AgentCore are making it easier to build and deploy agentic systems that respect existing paradigms for access control and identity management while providing transparency and auditability for agent interactions, tool use, and ultimate output and/or external actions.

Second, the foundational models on which these agents are built are becoming increasingly capable (in many cases using internal agentic flows) and smaller, distilled versions of large models lend themselves well to being customized for more specialization in a set of tasks performed by an individual agent. Capital markets firms can now customize their agentic systems to reflect their unique intellectual property, workflows, brand requirements, and specialized data.

Third, AI agents are getting easier to build as the ecosystem is becoming more robust. Agentic AI can require a lot of distracting, undifferentiated heavy lifting to build the technology infrastructure required to get agents into production. It’s no wonder that a recent Forrester study found that 57% of financial services organizations are still developing the internal capabilities needed to fully leverage agentic AI’s potential.

This is likely to change quickly, as companies now have many new choices when deciding whether to build in-house or to buy a wide range of pre-built AI agents, ready-to-integrate agent tools, agent infrastructure, agent development solutions, and professional services offered by third parties. Capital Markets firms can also leverage fully managed services purpose-built for scaling AI agents, eliminating capacity planning and infrastructure maintenance. These developments allow capital markets firms to focus on orchestrating the agents for maximum effectiveness.

How to build and deploy agentic AI easily with peace of mind

Despite the challenges, there are firms succeeding today with agentic AI. In a recent Forrester study, one financial services Vice President revealed their organization already had 60 agents in production, with plans to deploy an additional 200 agents by 2026.

Although agentic AI is becoming easier to manage technically, it still requires significant organizational changes. Agentic AI is reshaping processes and the nature of employees’ work. This transformation requires both team education and training. Additionally, operational risk policies and controls will need to be updated. To ensure a positive return on investment, it is crucial for capital markets firms to be selective about the use cases to which they apply agentic AI. They should look for significant:

  • Complexity: Is this a complex, multi-step process that requires a fully autonomous system that mimics human logic and reasoning? Or could it be solved by a simpler, task-oriented single agent, or by traditional rules-based systems?
  • Business value: Does the use case solve a real problem that will save time, save money, or delight customers?

Capital Markets firms should also keep their long-term agentic AI vision in mind from day one by focusing on the data pipelines, security, identity management, and observability they’ll need as they launch more agentic use cases over time. Another best practice is for Capital Markets firms to make sure they fully trust their individual agents before orchestrating them into complex multi-agent workflows.

Looking ahead

As we look ahead, agentic AI represents an exciting opportunity for Capital Markets firms to rethink and optimize their workflows. With the rapid maturation of tools, frameworks, and standards currently underway, now is a great time for Capital Markets firms to be building a solid foundation for success and safety in agentic AI.

Learn more:

Ruben Falk

Ruben Falk

Ruben is a Capital Markets Specialist with focus on Data Architecture, Analytics, Machine Learning & AI. Ruben joined AWS from S&P Global Market Intelligence where he was Global Head of Investment Management Solutions and ran product strategy and market development for S&P’s fundamental and quantitative investment management products including desktop, data feeds, NLP, and the ClariFI quant platform. Previously Ruben was a Director with UBS Investment Bank and also spent time as a management consultant. Ruben has a Computer Science degree from Brandeis University and an MBA from UC Berkeley.