Agentic Capital Markets: When AI Becomes the Counterparty

A woman analyzes a real-time risk dashboard on multiple monitors.

KEY TAKEAWAYS (DISPONIBLE EN ANGLAIS SEULEMENT)

AI has the potential to automate tasks and workflows across capital markets
Technology is less of a barrier today but building trust among stakeholders is a constraint
Every AI decision must be transparent, auditable and reversible to meet financial institutions’ risk-management standards
Firms will reach full autonomy only after governance, controls and trust evolve alongside the technology

Financial solutions providers are anticipating the next frontier where autonomous systems, powered by artificial intelligence, can execute trades, manage liquidity and oversee critical workflows. For the industry to fully deploy these agentic systems, it will take more than technological advances – it will also require clear lines of accountability and governance frameworks.

How the industry thinks about opportunities and necessary safeguards was the focus of a panel discussion I joined at Sibos 2026 in Miami, entitled: “Agentic Capital Markets: When AI Becomes the Counterparty.” On stage with me were:

Agnel Kagoo, Principal, KPMG International (Moderator)
Manoj Bohra, Chief Data and AI Officer, State Street
Akber Jaffer, CEO, Smartstream
Gloria Lio, Managing Director, Head of Enterprise Services, DTCC

This is a Markets Plus podcast based on insights from the panel discussion:

Below is a summary of the conversation’s insights:

Organizational constraints to automation

While AI can perform tasks at remarkable speed, integrating AI solutions into workflows requires trust in the output. That is one of the biggest current constraints to fully automating workflows. Autonomy and trust go hand in hand, and trust must be earned.

For banks, it is paramount to maintain oversight of agents. The technology must be fully transparent and reversible – and that’s increasingly more important as many tools are being developed outside of traditional engineering, governance and risk controls.

One of my fellow panelists used this analogy to explain the challenge of what’s called “agent sprawl”. In the past, a complex company might have had thousands of spreadsheets feeding into its financial reports, making it difficult to understand the sources and contexts of input data. That complexity has the potential to allow issues to go undetected and lead organizations astray.

The same is true for AI agents. If you have AI making 10,000 decisions, even small errors or deviations can compound at that scale.

Before financial firms can develop fully autonomous systems, they need guardrails in place that limit or shape what an AI agent is allowed to do, while ensuring human oversight to catch and correct problems.

The need for accountability

For financial institutions, accountability isn’t the technology’s responsibility. If an AI agent carries out a task, the institution is on the hook for the decisions and their consequences.

That’s why firms must fully understand and govern the entire AI process, including what an agent is authorized to do, how much risk the task carries and whether its actions can be reviewed or reversed.

An effective approach is to begin with measurable, lower-risk tasks and increase autonomy gradually as systems prove reliable. As the panel noted, you need to monitor both the actions and outcomes, keeping detailed records of agent decisions and actions.

As agents take on more work, controls will need to operate at the speed and scale of those systems. That means more than periodic reviews or controls that simply check a box: firms need ways to monitor agent actions and outcomes, identify when behavior drifts and intervene in time.

As financial industry participants move from using AI in support and recommendation roles toward fully autonomous ones, greater autonomy will need to be earned through testing, clear guardrails and controls tailored to each workflow’s risk. That will come with time.