2025 was a watershed year for artificial intelligence. Model capabilities leapt forward, adoption skyrocketed across industries, and investment poured in at an unprecedented scale. The trajectory suggests 2026 will bring even more dramatic change.
What will that stunning pace of innovation mean for quantitative investment management?
We asked a group of senior leaders, engineers, and researchers at Two Sigma what they’re watching as AI reshapes the landscape. In this first of a two-part series, hear their perspectives on AI’s trajectory, the balance between excitement and skepticism, and what developments matter most.
- Productivity gains are real but may bring new challenges. AI is accelerating research workflows, automating mundane tasks, and freeing researchers to focus on higher-level problems. The key is channeling that power wisely—maintaining research discipline as testable hypotheses multiply and designing robust objectives for autonomous agents.
- Human judgment matters more, not less. Researchers still need to be watchful supervisors. Institutional knowledge about deploying AI safely remains a differentiator. As Matt Greenwood, Chief AI Innovation Officer, says, “The future isn’t AI replacing humans; it’s humans who use AI well replacing humans who don’t.”
- Integration beats raw power. The shift is from deploying models to embedding AI as an operating system across workflows. Making tools “company aware,” maintaining research discipline, focusing on collaboration and engineering excellence: these organizational capabilities matter as much as model quality.
Read on for our expert takes on the opportunities, the uncertainties, which questions are still unanswered.
Q: As you think about AI’s trajectory in 2026 and beyond, what developments are you watching most closely, and why? Are you excited, skeptical, or somewhere in between?
Jeff Wecker, Chief Technology Officer
We’re watching all AI developments closely, especially new tools and their evolution, to ensure Two Sigma reacts quickly and brings in valuable innovations. The path toward agentic AI, meaning autonomous, properly governed agents, is what we see as the leading edge of transformation for the business.
It’s also important to observe which companies are making the fastest progress with AI, both to learn from their approaches and to inform our perspective on talent needs as Two Sigma grows. Innovation is happening fast, and we greet each new announcement with a mix of excitement and skepticism. We don’t automatically buy the hype; critical review is essential, since so much public commentary is speculative or premature.
Along similar lines, I’d be cautious about the notion of an “AI bubble.” Most companies out there have yet to realize AI’s full potential, and significant productivity gains are still ahead—for them and for the global economy as a whole.
As for Two Sigma, we remain well-positioned due to our deep talent bench and history. Looking ahead, I believe the coming years will bring both evolutionary and revolutionary changes across industries, creating opportunities for the firm in both operations and investment capabilities.
Matt Greenwood, Chief AI Innovation Officer
There’s a particular flavor of technology hype I’ve learned to recognize after two decades in this business. The revolution happens, but it looks nothing like the breathless predictions. The changes are structural, operational, and unglamorous.
AI is not a thing you deploy. It’s becoming the operating system for how quantitative research and investing work.
We’re at that moment with AI in quantitative investment management. Here’s my honest assessment for the industry, generally: the next year won’t be so much about LLMs making trades. It will be about AI becoming the operating system for how quant research and investing actually work.
The research funnel is inverting: Large language models (LLMs) are widening the top, shifting the bottleneck from “we need more ideas” to “we need to evaluate ideas faster.” Models are learning to read, joining text and structured data into unified representations. The economics of data are being rewritten through automated document understanding at near-zero marginal cost. Agentic tools are entering the portfolio room as interactive, constraint-aware copilots.
The mental model I’d offer: AI is not a thing you deploy. It’s becoming the operating system for how quantitative research and investing work.
The future isn’t AI replacing humans. It’s humans who use AI well, replacing humans who don’t.
Jin Choi, Head of Technique Forecasting
In 2025, we saw amazing breakthroughs in AI agents—agents backed by powerful LLMs can now plan, act, and handle audio and video in addition to text. I’m looking forward to seeing more of these agentic systems deployed in real-world applications in 2026, both in my work and in my personal life!
My concern is not that these agents are less capable than humans. It’s actually that they may be too capable.
At the same time, I am cautious about deploying autonomous agents in environments without robust safety checks and monitoring. My concern is not that these agents are less capable than humans. It’s actually that they may be too capable.
One thing I learned while training machine learning models is that you need to be careful what you wish for; they are great at maximizing the given objective function, and they’ll do exactly that. In practice, any objective function we design for an AI is only a proxy for our real goals, and powerful AI agents will relentlessly optimize that objective, sometimes with unintended consequences.
Building an even more powerful model is not necessarily a solution. In fact, it makes alignment more challenging, since more complex systems are harder to interpret, and more capable AI agents may be better at justifying or hiding their reasoning.
So, while I’m eager to explore the full potential of AI agents in the years ahead, I also believe each of us must remain a watchful supervisor, understanding not only what they can do but also their limitations.
Mike Schuster, Head of AI Core
A development that I try to watch closely and understand better is that many companies, in tech and in finance, seem to be spending very large amounts of money and effort on infrastructure, such as data centers. Is this evidence of a bubble? We are in an era when people brag about how much compute they are using to achieve certain goals, while the opposite—how little compute is used to achieve the same goals—should be more impressive.
I think it is clear that we will continue to see more and more use of LLMs and similar models to drive automation. AI agents are a new name for LLMs in action together, having access to lots of outside information, making decisions, and trying to automate more general tasks for people and companies. They may also help scientists to reason in ways that can drive research in ways we haven’t seen before (in combination with human judgement).
Ben Wellington, Head of Complex Feature Engines
I am profoundly excited about AI’s trajectory. There is no doubt that a significantly larger portion of our research work will be outsourced to AI agents, though it is hard to guess exactly how big the portion will be.
Regardless, it means that researchers are getting more and more efficient, since they can focus a higher percent of their time on more complex tasks. This, in turn, allows them to build more complex and robust models.
The takeaway
It’s clear that 2026 will be another transformative year in AI. The developments our colleagues describe – from agentic AI to multimodal capabilities to the evolving economics of compute – will only accelerate as the year progresses.
But as several note, the real impact won’t come from AI alone. It will come from people who understand how to deploy it thoughtfully, maintain research discipline, and build the organizational capabilities that matter.
If you’re interested in working at the intersection of AI and quantitative finance, visit our Careers page to learn more.
In Part 2 of this series, we’ll explore how AI is already changing the way quantitative managers approach key challenges.