AI in finance is advancing fast, but markets still decide what works

Artificial intelligence is moving rapidly across financial markets, but its real test is unfolding more slowly than many expected.

While much of the broader AI industry is defined by speed, faster models, quicker deployment, instant feedback loops, finance continues to operate on a very different timeline. In markets, performance is not validated in days. It is revealed over cycles.

A model can appear robust in stable conditions, only to fail when volatility rises, liquidity tightens, or correlations break down. These are not edge cases. They are the defining characteristics of real markets.

This growing tension between technological progress and market reality was a key theme at the recent Agentic AI and Automation in Finance Summit in Atlanta. Discussions increasingly centred not on what AI systems can do, but how they behave when conditions deteriorate.

As Kaushal Sheth, chief technology officer at GFT Technologies, noted during a panel alongside Juan Mendez of BlackRock, the real challenge is not building sophisticated models,  it is understanding their behaviour when markets stop behaving normally.

That distinction is becoming increasingly important as firms push toward more autonomous systems.

Agentic AI, systems capable of making or executing decisions across workflows, introduces a different level of risk. In these environments, an incorrect output is no longer just a flawed signal. It can become a flawed action, with direct financial consequences.

This shifts the conversation from capability to reliability.

Unlike other sectors where errors can be quickly identified and corrected, financial systems often reveal weaknesses only over time. A strategy that performs well over weeks may fail across a full market cycle. Stability in low-volatility environments does not guarantee resilience under stress.

As a result, experience is emerging as one of the hardest advantages to replicate.

The rapid expansion of AI tools has lowered the barrier to entry for building models. However, it has not shortened the time required to observe how those systems behave across different regimes. Real-world validation still depends on exposure to market cycles that cannot be compressed or simulated fully.

This is where infrastructure-led approaches are gaining attention. Through both his role at GFT Technologies and his work with Otonomii, Sheth has focused on integrating AI systems that are designed not just for performance, but for durability, systems that are observed, refined and tested over extended periods rather than optimised for short-term outputs.

That approach reflects a broader shift across institutional finance.

For asset managers and capital allocators, the question is no longer simply whether a system works. It is whether it continues to work when conditions change — and how it behaves when it does not.

In a market environment defined by uncertainty, that distinction carries weight.

As artificial intelligence becomes more embedded in trading, portfolio construction and risk management, the competitive edge may not lie with those who build the fastest systems, but with those who have seen them operate through stress.

In finance, time remains the ultimate benchmark. And for AI, that may be the one variable that cannot be accelerated.