AI Agents Are Running Hedge Funds Now. Here’s Exactly How.

A Hedgeweek survey published in early 2026 found that not a single hedge fund manager said they had no plans to use AI. Zero respondents. In an industry built on disagreement and contrarian positioning, the consensus is unanimous — and that alone should tell you something about where this is going.

Two years ago 86% of funds were experimenting with AI tools. Today major firms are running them operationally and some funds have eliminated human analysts entirely. That is not a trend. That is a structural shift, and it is already done.

The people treating this as a future problem are already behind.

What Is an AI Agent, Actually?

Before we get into what’s happening at funds, it’s worth being precise about the term — because it’s getting thrown around loosely.

An AI agent isn’t just a chatbot you ask questions to. An agent is an autonomous system that can perceive inputs, make decisions, call tools, take multi-step actions, and produce outputs — all without a human in the loop at each step. The critical difference between a language model and an agent is agency: the ability to act, not just respond.

In a hedge fund context, this means an agent can be handed a task — say, “monitor the semiconductor supply chain and flag any anomalies in real-time data” — and execute it continuously, across multiple data sources, 24 hours a day, with no analyst babysitting it.

That’s a fundamentally different thing than autocomplete.

The Numbers Behind the Shift

The adoption curve is steep and the performance data is starting to come in.

Over 70% of global hedge funds now use machine learning models somewhere in their trading pipeline. Around 18% rely on AI for more than half of their signal generation — that figure, sourced from multiple industry surveys, was essentially zero five years ago. In 2025, the industry crossed an estimated $5 trillion in global hedge fund assets, and more than 35% of new fund launches branded themselves as AI-driven or AI-enhanced.

The performance impact is real too. Funds incorporating generative AI into decision-making have clocked 3–5% better returns on average compared to non-AI peers, with the uplift even more pronounced in equity hedge strategies where AI excels at pattern recognition.

AI adoption has also driven operational efficiencies that cut costs by up to 20% by mid-2025, allowing funds to scale assets under management without a proportional increase in headcount. Some funds report 15% better volatility management in high-risk strategies like crypto and ESG-integrated portfolios.

The 2025 returns from the top funds paint a vivid picture.:

• Bridgewater’s flagship Pure Alpha II fund surged 34% — the highest return in its 50-year history.

• D.E. Shaw’s Oculus fund returned 28.2%.

• Point72 delivered 17.5%.

• Balyasny posted 16.7%.

• AQR’s Apex Strategy returned 19.6%.

Even the “laggards” — Citadel at 10.2% and Millennium at 10.5% — outperformed the S&P 500’s 16.4% gain. And Bridgewater’s ML-driven AIA Labs fund, which uses machine learning as the primary basis of its decision-making, raised more than $5 billion and returned 11% in 2025 — its first full year of meaningful deployment.

Multi-strategy hedge funds are arguably the most complex operating environments in finance. Each new pod, asset class, or data feed adds its own workflows, approval chains, and exception paths. The complexity grows nonlinearly.

This is precisely where agentic AI earns its keep. When designed well, these systems can interpret operational objectives, decide the sequence of steps needed, call tools across the stack — databases, APIs, compliance engines, order management systems — verify outputs, handle exceptions, and produce auditable records of what was done and why.

Here’s how the typical agent stack looks in practice:

The Market Data Agent is the system’s eyes and ears. It collects real-time data from multiple sources, standardises formats, identifies anomalies, and monitors trading volumes across markets and time zones. Think of it as the layer that ensures every other agent is working off clean, consistent information.

The Sentiment Agent processes news feeds, earnings call transcripts, analyst reports, and social signals to produce a real-time read on market mood — directional, sector-specific, and calibrated against historical precedent.

The Fundamental Agent handles the classic analyst workflow: earnings analysis, balance sheet review, competitive positioning, growth modelling. An analyst who previously tracked 20 stocks can now oversee a fleet of agents covering 200.

The Quantitative Agent runs mathematical models over price data, volume patterns, and cross-asset correlations, generating trading signals based on statistical relationships.

The Risk Agent monitors portfolio exposure continuously — setting stop-loss triggers, assessing liquidity risk, watching for concentration, flagging when the portfolio drifts outside acceptable parameters.

At the center sits the Portfolio Manager Agent — the synthesis layer. It takes conflicting signals from every other agent, weighs them, resolves disagreements, and produces a final trading decision.

Funds are also running agents across their operational and compliance layers — maintaining audit trails, managing approval workflows, flagging regulatory issues, and preparing evidence for internal audit. Not glamorous, but in a regulated industry, this is where agent adoption is most mature and most quietly impactful.

Steve Cohen’s Point72 launched its first new fund in decades in October 2024. Called Turion, it’s a dedicated AI-focused fund that maps winners and losers across the AI supply chain — chips, foundries, packaging, memory. The fund hit nearly $1.5 billion in assets within three months of launch, temporarily stopped accepting new investors in April 2025 after hitting its initial capacity target, and returned 30% through November 2025. To put that in context, the S&P 500 returned about 16% for the year.

Point72 also partners with AI platforms to process earnings calls in real time, identifying linguistic patterns and sentiment shifts that human analysts might miss. That’s a separate, earlier wave of AI adoption — the kind that augments what analysts do rather than replacing the workflow.

Balyasny, which returned 16.7% in 2025, has built one of the most sophisticated proprietary AI infrastructures in the industry. Led by former Google data scientist Charlie Flanagan, who assembled a team of over 13 researchers and engineers recruited from Google and DeepMind, the firm built its own internal tool called “BAMChatGPT.”

Currently used by 80% of the fund’s 2,000 employees, the system is hosted on Azure and connected to ten different data pipelines. Flanagan stitched together dozens of micro-agents — one flags changes in 10-K wording, another builds morning notes, another monitors for ESG controversies in alt-data before they hit production signals. The result: tasks that previously took a senior analyst two days now take thirty minutes.

The firm later launched Deep Research, a bot that searches five million documents and answers portfolio manager questions in minutes. The system demonstrates measurable performance advantages over general-purpose AI tools — which is exactly why the firm built it in-house rather than relying on commercial products.

D.E. Shaw — The Federated Model

Quant giant D.E. Shaw, whose Oculus fund returned 28.2% and Composite fund returned 18.5% in 2025, takes a deliberately different approach. Rather than building a centralised AI platform, D.E. Shaw’s architecture — an Assistants/LLM Gateway/DocLab stack — lets any desk build custom tools with “as little as ten lines of code,” while a central team enforces prompt logging and model-use policies.

It’s the most compelling proof yet that federated AI innovation can coexist with hard governance. Individual teams get flexibility and speed. The firm gets control and auditability.

Bridgewater — AIA Labs

Bridgewater, freshly under CEO Nir Bar Dea after Ray Dalio’s complete exit from the firm, has been in aggressive rebuild mode. Its Pure Alpha II fund’s 34% return in 2025 — the best in the firm’s 50-year history — came alongside the launch and scaling of AIA Labs, its machine-learning-driven fund that now manages more than $5 billion and reportedly uses models from both OpenAI and Anthropic.

The key distinction in Bridgewater’s approach: this isn’t AI for research augmentation. AIA Labs uses ML as the primary basis of its decision-making. The fund combines large language models with reasoning tools designed to understand causal relationships in markets, not just statistical correlations.

Man Group — The Alpha Assistant

The London-listed Man Group, which manages roughly $160 billion, has built what it calls the “Alpha Assistant.” Designed to shrink the idea-to-P&L cycle from weeks to hours, the system can read, reason, code, and back-test in a single loop. Early prototypes are already drafting trade rationales and surfacing anomalies in alternative data before they reach production signals.

Man Group also runs “ManGPT,” rolled out in June 2023, which is now used by roughly 40% of employees each month for tasks including research summarisation, foreign-language translation, and coding support.

Numerai — Crowdsourcing AI Intelligence

Numerai, the crowdsourced quant fund backed by JPMorgan Asset Management (which committed up to $500 million in August 2025), is redesigning its entire system to support AI agents rather than just human data scientists. Under the new framework, agents can create models, submit predictions, run validation tests, and monitor performance on their own — executing the full research cycle without human intervention. The human role is shifting from building models to designing and supervising AI research systems.

Numerai’s founder Richard Craib is characteristically blunt about what this replaces: “I’m worried about hedge fund managers having coffee at the St. Regis and buying $100 million of NVIDIA based on their vibes. There are 2,000 dimensions of data — at what point do you realise you had a lucky call, and it had nothing to do with you?”

What’s Still Human (and Why It Matters)

Agents are exceptional at processing structured data at scale, executing defined workflows, monitoring for anomalies, and synthesising quantitative signals. They are significantly worse at interpreting qualitative judgment — reading a management team in a room, weighing a strategic pivot, understanding cultural dynamics inside a company, or recognising when a pattern that looks statistically valid is actually noise.

Experts across the industry are consistent: agentic AI reduces operational friction, it doesn’t remove the need for judgment. The risk agent can set stop-losses. It can’t tell you when the stop-loss logic itself is wrong. The fundamental agent can analyse 200 balance sheets. It can’t tell you whether the CFO is being honest about guidance.

That distinction is where the career conversation gets interesting.

What This Means for Anyone Building a Finance Career

The floor is rising. AI proficiency is now a baseline requirement at top funds — not a differentiator. In the Hedgeweek survey, zero respondents said they had no plans to use AI. The industry has arrived at the point where not knowing how to work with these systems is a liability, not a neutral position.

The ceiling is also rising. The analysts who figure out how to use agent fleets as leverage — covering more names, running more scenarios, synthesising more information — can do work that would have required a team of five a few years ago. Balyasny’s example is instructive: tasks that once took a senior analyst two days now take thirty minutes. That’s not replacing the analyst. That’s multiplying their output by a factor of ten.

And the moat has shifted. The traditional analyst edge was information access: I read more, I got to the data first, I talked to more people. That edge is nearly gone. What remains scarce — and what no agent stack can replicate — is the judgment to ask better questions, design better systems, and know when the output is wrong.

The people who understand both sides of the equation — the finance fundamentals and the AI systems — are currently in short supply. That window won’t stay open forever.

That’s the arbitrage right now. Come watch the trade play out.