The Machine Mandate: Could Agentic Trading Systems Replace Human Fund Managers?

The question is no longer whether artificial intelligence will transform fund management. It already has. The question now being tested in real capital across dozens of firms and billions of dollars is whether a new class of autonomous AI systems, known in the industry as "agentic" systems (meaning systems that can set their own sub-goals, gather information, and take action without continuous human direction), will eventually render the human fund manager obsolete.

The proponents argue that the answer is self-evident. The sceptics argue that the proponents have never managed money through a genuine crisis. Both positions deserve serious examination.

What "Agentic" Means in Practice

The term requires definition, because it has been applied so broadly in financial technology marketing that it risks meaning nothing at all. In its most precise usage, an agentic trading system is one that operates with a degree of autonomy that goes beyond traditional algorithmic trading. A conventional algorithmic system follows pre-programmed rules: if a stock drops below a moving average, sell; if a yield spread widens past a threshold, buy. The rules are set by humans. The system executes them.

An agentic system, by contrast, is designed to formulate its own intermediate objectives. Given a high-level mandate (maximise risk-adjusted returns within these constraints), the system can identify what data to gather, what analysis to run, what trades to execute, and when to revisit its own assumptions. The human sets the objective and the risk parameters. The machine does the rest.

The distinction matters because it represents a qualitative shift in the role of technology in portfolio management. Algorithmic trading has been mainstream for two decades. Agentic portfolio management, in which the system makes the investment decisions rather than merely executing them, is the step that would, if it works, displace the human decision-maker entirely.

The Firms Already Building This

The firms investing most heavily in autonomous trading systems are, unsurprisingly, the quantitative hedge funds that have spent decades building the infrastructure on which agentic systems depend.

Renaissance Technologies, the firm founded by mathematician Jim Simons, has operated its Medallion Fund with minimal human trading intervention since the 1990s. The fund, which is closed to outside investors and manages the personal capital of firm employees, has generated average annual returns of approximately 66% before fees over a 30-year period, according to public disclosures compiled by Gregory Zuckerman in "The Man Who Solved the Market." Medallion's approach, while predating the current generation of large language models and reinforcement learning agents, established the proof of concept that systematic, data-driven trading could consistently outperform human discretionary management.

Two Sigma, the New York-based quantitative firm managing approximately $60 billion in assets as of its most recent public disclosures, has been among the most explicit about its investment in agentic AI. The firm's research division has published work on using machine learning systems that can autonomously identify trading signals across multiple asset classes and data types, reportedly cross-referencing information in 14 languages. Public filings and the firm's own statements indicate that AI-driven processes now play a role in the majority of its investment decisions.

Man Group, the London-listed hedge fund manager, has invested heavily in its AHL division's machine learning capabilities. Man AHL manages approximately $55 billion and has publicly discussed its use of reinforcement learning agents that can adapt their trading strategies in response to changing market conditions without requiring human reprogramming. The firm's research team has spoken publicly about a future in which the human role shifts from making decisions to designing the systems that make decisions.

DE Shaw & Co., founded by computer scientist David Shaw, manages approximately $60 billion and has long operated at the intersection of computational science and financial markets. The firm's systematic strategies rely on proprietary models that, according to public descriptions, operate with significant autonomy in identifying and executing trades.

Bridgewater Associates, the world's largest hedge fund by assets under management, has publicly discussed its ambitions to systematise investment decision-making. Ray Dalio's "principles" framework was always, in part, an attempt to encode human decision-making into rules that could be followed systematically. The firm's more recent investments in AI and machine learning, documented in public statements and media reports, represent an extension of that ambition toward genuinely autonomous portfolio management.

The Case for the Machines

The bull case for agentic trading systems rests on several structural advantages over human decision-makers, each of which is well-documented in academic research and industry data.

Data processing capacity. A human fund manager, however talented, can read a finite number of earnings reports, regulatory filings, and research notes per day. An agentic system can process the full universe of public filings, news feeds, satellite imagery, shipping data, credit card transaction aggregates, and social media sentiment across every market simultaneously. The informational advantage is not marginal; it is categorical.

Temporal coverage. Human beings sleep. Markets do not, particularly in a world where foreign exchange, cryptocurrency, and commodity futures trade continuously. An agentic system monitors and responds to market developments around the clock, eliminating the gap between event and response that a human team, however well-staffed, cannot fully close.

Emotional discipline. The behavioural finance literature, from Kahneman and Tversky onward, has extensively documented the cognitive biases that degrade human investment performance: loss aversion, anchoring, recency bias, overconfidence. An agentic system does not panic in a selloff, does not fall in love with a position, and does not anchor to a purchase price. It processes the current state of the world and acts accordingly.

Execution speed. In the most latency-sensitive strategies, AI systems execute trades in nanoseconds. Even in longer-horizon strategies where speed is less critical, the ability to rebalance a portfolio instantaneously across hundreds of positions in response to new information represents a material advantage over manual processes.

Cost. At scale, an agentic system is cheaper to operate than a team of experienced portfolio managers, analysts, and traders. If the performance is equivalent or superior, the cost advantage compounds over time through lower management fees, a consideration that matters to every institutional allocator under pressure to reduce costs.

The Case for the Humans

The bear case is less frequently articulated in technology industry coverage, but it is substantive.

Regime changes and genuine novelty. Every AI system, however sophisticated, is trained on historical data. The most consequential events in financial markets are, by definition, the ones that have no close historical precedent: the 2008 global financial crisis, the March 2020 COVID liquidity event, the 2022 LDI crisis in UK pension funds. A system optimised to identify patterns in historical data may fail precisely when the patterns break. Human fund managers, particularly those with experience across multiple market cycles, bring a capacity for reasoning about genuinely novel situations that current AI systems have not demonstrated in live markets.

Correlated failure. If the most successful agentic systems converge on similar strategies (because they are trained on similar data and optimise for similar objectives), their collective behaviour could amplify market instability rather than dampen it. The "quant quake" of August 2007, in which multiple quantitative funds simultaneously unwound similar positions, causing a cascading selloff, provides a historical precedent for this risk. Widespread adoption of agentic systems could create a more extreme version of the same phenomenon.

Black box risk and accountability. When an agentic system makes a decision that results in significant losses, the question of why it made that decision is not always answerable. The most advanced neural network architectures are not fully interpretable even to their designers. For institutional investors with fiduciary obligations, and for regulators tasked with market oversight, the inability to explain a decision is not merely an inconvenience; it is a governance problem.

Regulatory uncertainty. Financial regulators globally have not yet established clear frameworks for autonomous trading systems that make investment decisions without direct human oversight. The European Union's AI Act, the SEC's proposed rules on AI in investment advice, and similar initiatives in other jurisdictions are still in early stages. A firm that builds its entire investment process around agentic systems faces the risk that future regulation could require fundamental changes to its operating model.

The limits of quantification. Some of the most important factors in credit analysis, corporate governance assessment, and geopolitical risk evaluation resist quantification. A fund manager evaluating a Turkish corporate issuer may weigh the quality of management, the reliability of regulatory enforcement, and the political dynamics affecting a sector in ways that draw on contextual knowledge that is difficult to encode in a training dataset. Whether agentic systems will eventually replicate this kind of judgement, or whether it represents an enduring human advantage, is genuinely uncertain.

The Numbers So Far

The adoption data, while early, is striking. According to a KPMG estimate, global spending on agentic AI reached approximately $50 billion in 2025. A survey cited by multiple industry sources suggests that 44% of finance teams plan to deploy agentic AI systems by the end of 2026, a sixfold increase from the roughly 6% that had done so as of early 2025. Recent Y Combinator batches have shown a marked shift toward agentic AI ventures, particularly in investment-related categories. One financial services executive disclosed publicly that their organisation had 60 agents in production, with plans to deploy 200 more by the end of 2026.

The counter-data is equally worth noting. Gartner, the technology research firm, has warned that a significant proportion of agentic AI projects risk cancellation, citing escalating costs, unclear business value, and inadequate risk controls. The gap between ambition and execution in financial technology is not a new phenomenon, and the history of AI in finance includes several previous waves of enthusiasm that delivered less than advertised.

The Turkish Dimension

For Turkey's fund management industry, the agentic revolution remains largely theoretical, but not entirely. The country's quantitative finance graduates are already a significant export to the global firms building these systems. Boğaziçi, ODTÜ, and Bilkent produce graduates who appear on the engineering and research teams of Two Sigma, Citadel, and DE Shaw, among others. The talent pipeline runs almost entirely outward.

Domestically, Turkey's asset management sector remains overwhelmingly discretionary. The quantitative approaches that do exist tend to be rules-based allocation models operating at a level of sophistication well below what the frontier firms are deploying. The barriers are structural: limited data infrastructure, a relatively small investable universe in Turkish equities and fixed income, and a regulatory framework that has not yet addressed autonomous trading systems.

The more immediate question for Turkish fund managers may not be whether agentic systems will replace them, but whether the global firms that deploy such systems will increasingly compete for the same capital. As international institutional investors gain access to AI-driven strategies that promise superior risk-adjusted returns, the pressure on domestic managers to justify their fees and their human-driven processes will intensify, regardless of whether those domestic managers adopt the technology themselves.

The Open Question

The investment management industry is not short of technologies that were supposed to transform it. Quantitative screening was going to eliminate the need for fundamental analysis. Smart beta was going to eliminate the need for active management. Robo-advisors were going to eliminate the need for financial advisors. In each case, the technology delivered real value but did not deliver the extinction event that its proponents predicted. The incumbents adapted, the technology found its level, and the industry absorbed it.

Agentic AI may follow the same pattern, becoming a powerful tool that augments human decision-making without replacing it. Or it may be different this time, a phrase that carries more freight in financial markets than in any other context.

The data that would resolve the question does not yet exist. The firms building agentic systems have not yet navigated a genuine financial crisis with their autonomous strategies. The track records are short. The regulatory frameworks are incomplete. The technology is improving rapidly, but so is the understanding of its limitations.

What is clear is that the experiment is underway, funded by some of the most sophisticated investors in the world, staffed by some of the most talented technologists, and operating with real capital at meaningful scale. The results of that experiment will determine whether the human fund manager remains essential, becomes optional, or is eventually rendered obsolete.

The answer is not yet known. The question has never been more consequential.