HSBC Exec Predicts AI Agents Trading Securities

Michael Roberts, head of HSBC's corporate and institutional bank, said on Thursday, October 15, that banks may soon allow AI agents to trade in financial markets instead of human operators. The statement was made in an interview with Bloomberg TV during the IMF annual meetings in Bangkok. According to Roberts, the next two to four years will fundamentally change how the bank views the "pace, speed, and scope" of its trading activities.

The executive pointed to debt securities as the market closest to this transition and associated the change with quantum computing, which, according to him, is capable of analyzing very complex market movements. He noted that while more algorithmic trading will be conducted by agents, humans must continue to be "in the loop," meaning they will supervise decision-making.

The Technical Foundation of the Prediction

His remarks are not without basis. HSBC reported a test with IBM in which a hybrid approach, combining quantum and classical computing, improved the prediction of execution orders for securities by up to 34% compared to conventional methods, according to the bank. The corporate fixed-income market, traded largely over-the-counter and with low liquidity in many securities, is precisely where predicting whether an order will be executed, and at what price, is valuable.

There is a significant gap between improving a prediction in a testing environment and giving an autonomous agent the decision-making power to buy and sell. Roberts did not elaborate on volumes, desks, or internal timelines, and Bloomberg noted that he himself expressed doubts about the feasibility. His statement is an executive projection, not a plan formally approved by the bank.

The Background: 20,000 Jobs

The weight of his statement is in the context of personnel. Bloomberg reported this year that HSBC might cut around 10% of its global workforce as it introduces AI into the business, equivalent to approximately 20,000 positions. The bank has not formally announced this number, but Roberts' forecast indicates where automation is expected to reach after operations and service areas: the trading desk itself, which has for decades been the most protected sector of investment banking.

HSBC is not alone. Morgan Stanley analysts estimated in May that European banks might eliminate between 10% and 20% of jobs in the next five years due to AI, largely through voluntary departures and retirements.

Roberts also signaled caution on the other side of the balance sheet. In a report published by Bloomberg on the same day, he stated that AI-related business deals are becoming harder to close, with banks and investors becoming more selective when financing these projects. The same executive predicting AI agents trading securities recognizes that credit for AI infrastructure is becoming harder to secure.

Where the Effect Reaches

In the UK, the group's headquarters, the transition affects the London trading desks and regulation. The FCA requires banks to demonstrate control over trading algorithms, and an autonomous agent tests the limits of this oversight: who is responsible for an order that no human typed?

In Hong Kong, the group's Asian hub, the dollar bond market for Asian issuers is precisely the type of fragmented environment where predicting execution provides a competitive edge. There are the desks that will first feel the replacement of junior operators with agents.

In India, the scenario is different. HSBC maintains large global service centers in the country, processing back-office, risk control, and operational support for desks around the world. If the agent executes and records the operation end-to-end, some of this checking and reconciliation work will disappear, not just the position of the operator in London.

In Brazil, where HSBC sold its retail operation to Bradesco in 2016, the effect comes through competitive comparison. Local investment banks operating in private credit, a market also characterized by low liquidity, will be measured by the same global institutional clients demanding faster and cheaper execution.

Roberts' caveat about keeping humans in the loop is what regulators will first press for. It also defines the actual scale of the cuts: a supervised agent still requires the senior operator overseeing it, while the career ladder that trained them—the junior analyst learning by executing small orders—will be the one without rungs.