Last month, Ramit Sawhney, Tower’s Global Head of Core AI & ML, joined industry peers at the 2026 AI and Future of Finance Conference hosted by Georgia Tech’s Scheller College of Business.
Ramit participated in the “Hedge Fund Perspectives on Implementing AI in Capital Markets” panel alongside leaders from Balyasny, Millennium, and Schonfeld. The discussion was part of a broader program focused on the evolving role of AI across financial services – and, more specifically, what it takes to move AI from experimentation to production-grade implementation.
For many capital markets firms, the conversation around AI has shifted. The question is no longer simply whether AI can create value, but how firms can apply it responsibly, securely, efficiently and at scale. As organizations move beyond proof-of-concept work and begin exploring more advanced AI agents and structured AI systems, Ramit emphasized that success depends as much on the architecture around AI as on the underlying models themselves.
That means building systems with the right foundations: strong governance, standardized development practices, clear cost controls, robust security, and a disciplined approach to identifying where AI can create strategic value.
From Isolated Tools to Structured Systems
One of the key themes Ramit discussed was the move away from isolated AI tools or standalone chatbots and toward more integrated, structured AI environments.
In practice, this means designing AI systems that can work with the full context of an organization’s data, workflows, and operational requirements. Rather than relying solely on static documents or flat information retrieval, firms are increasingly exploring evolving knowledge bases and knowledge graphs that map relationships between entities, data points, systems, and processes.
This kind of structure gives AI agents a richer foundation for reasoning. By helping systems understand not just individual pieces of information, but how those pieces relate to one another, firms can improve the consistency, accuracy, and usefulness of AI-supported outputs, especially in complex environments like capital markets.
Building More Disciplined Agents
The panel also explored the growing role of AI agents, including auto-research-style systems that can browse, synthesize, analyze, and report on large volumes of information.
These capabilities are creating new possibilities for research, engineering, operations, and decision support. But Ramit also noted that autonomy without structure can quickly become a risk. As AI agents become more capable, firms need the right controls around how those agents operate.
That includes what Ramit described as an agent “harness” – the surrounding framework that governs how agents interact with internal tools, external APIs, data sources, and feedback loops. A strong harness can help standardize execution, monitor performance, benchmark outputs against defined KPIs, and provide observability into how agents are reaching conclusions.
In other words, the goal is not simply to build smarter agents, but to build more disciplined ones.
Governance, Security, and Cost Management
As AI moves closer to production, operational rigor becomes even more important. Without clear controls, AI systems can become expensive to run, difficult to monitor, and challenging to govern. Agents that perform recursive tasks, consume large volumes of tokens, or interact with multiple systems can create cost and operational complexity if they are not carefully managed.
Ramit emphasized the importance of cost management, including token budgeting, execution limits, and ROI measurement. He also highlighted the need for security and data governance as AI systems become more deeply embedded in enterprise workflows.
For capital markets firms, this is especially critical. AI systems must operate within secure, permissioned environments that protect sensitive data, support compliance obligations, and prevent unintended information exposure. A zero-trust approach to agent permissions – where systems are granted only the access they need, within clearly defined boundaries – can help firms capture the benefits of AI while managing the risks.
Applying AI with Strategic Discipline
A recurring theme throughout the discussion was the importance of use-case discipline.
As AI capabilities continue to expand, firms may feel pressure to apply the technology everywhere. But Ramit emphasized that successful implementation depends on clearly defining where AI can create meaningful business value – and recognizing where it may not be the right fit.
That requires evaluating AI initiatives through the lens of adoption, engagement, ROI, KPIs, security, and operational impact. It also requires balancing ambition with practicality: moving quickly enough to innovate, but carefully enough to ensure that AI systems are reliable, measurable, and aligned with real business needs.
The panel reflected a broader shift underway across the industry. Capital markets firms are no longer just experimenting with AI. They are working to understand how AI can become part of durable, production-grade infrastructure.
Tower is proud to contribute to these conversations and to help shape thoughtful dialogue around the future of AI in finance.