AI can price the trade. Who says yes?

Every generation of market participants believes it is witnessing a transformative technology. I speak as someone who joined ICAP in 1999, during the peak of the dot com mania, when some believed the boom in day trading would put every intermediary, market maker and broker out of business.

Over the past three decades, financial markets have absorbed electronic trading, algorithmic execution, high-frequency trading, cloud computing and ever vaster quantities of data. Each development promised to make markets faster, more efficient and better informed. 

Artificial intelligence is the latest chapter in that story.

Much of the discussion focuses on capability. How effectively can AI price risk, identify liquidity, structure transactions or optimise execution? As models improve and computing power increases, the answer appears to be that machines will perform an expanding share of these activities.

The excitement is understandable. Yet recent events have also demonstrated how quickly enthusiasm can collide with the realities of markets. The sharp reversal at the AI-focused hedge fund Situational Awareness, which saw billions of dollars erased as concentrated, highly leveraged positions unravelled, served as a reminder that technological conviction does not suspend the fundamental disciplines of risk management and good governance. 

That distinction matters because much of the current discussion around AI focuses, almost exclusively, on capability. How effectively can AI price risk, identify liquidity, structure transactions or optimise execution? 

But financial markets have never been defined solely by the quality of their technology. They are also systems of governance, responsibility and trust. Every transaction ultimately carries consequences for clients, counterparties, shareholders and regulators. When outcomes are favourable, the allocation of responsibility rarely attracts much attention. When outcomes are unfavourable, it becomes impossible to ignore.

This is where the debate around AI becomes particularly interesting.

The most important question may not be how much of the market process can be delegated to a machine. Markets have been delegating elements of decision-making to technology for decades. The more relevant question is how accountability evolves as those systems become increasingly capable, adaptive and autonomous.

History suggests that technological progress rarely removes the need for human responsibility. More often, it changes where responsibility sits and how it is exercised.

Aviation provides a useful comparison.

Modern commercial aircraft rely extensively on automation. Many aspects of a long-haul flight are managed by systems capable of operating with a level of precision that would be difficult for any human to replicate consistently. The result has been dramatically safer and more efficient air travel.

The role of the pilot did not disappear as that technology advanced - it evolved.

Pilots spend less time manually controlling the aircraft than previous generations, yet they remain ultimately responsible for the outcome. Their value increasingly lies in understanding context, monitoring complex systems, interpreting unusual situations and intervening when circumstances move beyond established parameters.

In other words, automation reduced the need for manual involvement while increasing the importance of oversight, judgement and accountability.

Markets are approaching a similar moment.

As AI becomes capable of pricing, structuring, analysing and recommending transactions, it is tempting to view this as a sharp break from the past. The language certainly encourages it. Artificial Intelligence sounds fundamentally different from the algorithmic, ‘if-this-then-that’, trading systems that have operated across financial markets for decades.

Yet I think the distinction may be less profound than it first appears.

Markets have long relied on machine decision-making. High-frequency trading firms deploy systems capable of analysing market conditions, identifying opportunities and executing transactions in micro-, even nano-, seconds. For years, competitive advantage was measured by ever-faster execution, smarter routing and increasingly sophisticated algorithms. 

Today, that source of advantage is becoming harder to sustain.

As electronic trading has matured, firms increasingly compete not on access alone, but on the insight that surrounds execution. Sophisticated algorithms remain essential, but they are rarely sufficient in themselves. Success is increasingly determined by how effectively firms combine technology with data, liquidity management, client understanding and market expertise.

That shift is revealing.

If machine intelligence alone were the decisive factor, trading would already be a largely commoditised activity. Instead, many of the attributes clients value most remain stubbornly human. Understanding client intent. Advising on execution strategy. Interpreting fragmented liquidity. Navigating periods of market stress when normal patterns break down.

Even in highly electronic markets, participants increasingly talk about "liquidity consulting" rather than simple execution. The objective is no longer merely connecting buyers and sellers but in helping clients determine when, where and how to access liquidity in pursuit of the outcome they actually want.

This brings us back to AI.

Much of the current debate wrongly assumes a clear distinction between traditional algorithms and artificial intelligence. In practice, markets may not experience such a clean dividing line.

The history of electronic trading offers plenty of examples of systems operating at levels of complexity that are already beyond the direct comprehension of any individual trader. The Knight Capital incident in 2012 remains one of the most notable examples. A software deployment error triggered a cascade of unintended trades, generating approximately $440 million of losses in less than an hour. The trades were executed by machines, but responsibility rested entirely with  the firm that designed, deployed and governed the system.

The lesson was not that machines should be removed from markets but that complex systems require robust controls, supervision and accountability. That principle remains unchanged whether the system is described as an algorithm, a machine learning model or an AI agent.

The technological architecture may differ. AI systems may be more adaptive, less transparent and capable of identifying patterns that traditional rule-based models would miss. But the governance challenge remains remarkably familiar.

In every case, firms are delegating elements of decision-making to technology operating at a speed and scale beyond direct human observation. The critical question is not how the recommendation was generated, but whether the surrounding framework is capable of understanding, challenging and controlling it.

Indeed, the next competitive frontier may have less to do with artificial intelligence itself than with how effectively firms integrate it into broader decision-making processes.

The firms that consistently outperform are rarely those with the most impressive technology in isolation. They are often those that combine technology with higher-quality data, stronger governance, more reliable liquidity provision and deeper client relationships. Execution quality is increasingly judged by consistency rather than novelty: showing up during periods of volatility; maintaining pricing discipline and delivering liquidity when clients actually need it. Those characteristics are difficult to automate because they are ultimately manifestations of strategic choice and principle rather than technological capability.

Counterintuitively then, that suggests that AI may accelerate existing trends rather than fundamentally alter them. 

Markets have spent decades automating execution and integrating machine-driven decision-making. The next challenge lies in ensuring that increasingly sophisticated AI systems operate within a framework of human judgement, commercial discipline and effective risk management.

In that outcome, the future of trading might look less revolutionary than many expect. Technology will continue to advance; algorithms will become more sophisticated, and AI will certainly become deeply embedded across the trading lifecycle. But the defining source of value will, increasingly come from something machines still struggle to replicate: understanding context, exercising judgement and helping clients navigate uncertainty.

The next chapter in trading may therefore be defined not by access to intelligence, but by the quality of insight applied around it.