In Part 1 of this series, Two Sigma leaders shared their perspectives on AI’s trajectory and the developments they’re watching most closely. The consensus: 2026 will bring significant change, but success will depend as much on research discipline and organizational capabilities as on raw technological advancement.
Here in Part 2, we turn to the practical question: how will AI actually change the way quantitative investment managers approach their work?
Q: How do you anticipate AI will change the way quantitative investment managers approach key challenges in the next year or so?
Jeff Wecker, Chief Technology Officer
I am constantly thinking about what Two Sigma is doing to leverage AI for productivity and workflow improvements across all areas, from alpha identification to legal and compliance applications.
Two Sigma continues to integrate frontier AI models and tools to enhance both individual and team productivity, not just in investment research but in every workflow where gains are possible. Everyone at Two Sigma is expected to use LLMs to accelerate their work and consider workflow improvements for their teams.
The firm is making these AI tools increasingly “Two Sigma aware,” so they understand internal platforms, research, production environments, incident management, and business processes — enabling faster, higher-quality outcomes across the organization. The challenge is to keep pushing for the most effective ways to use these tools to impact and improve the business.
Matt Greenwood, Chief AI Innovation Officer
I’ve been trying to make sense of where AI is heading—not the everything-is-awesome version or the doom-and-gloom version, but something messier and more interesting.
Beyond the operational shifts I described earlier, several deeper currents matter. The era of “bigger is better” is ending. Training costs are outpacing ROI, and the action is shifting to efficiency: new architectures, specialized accelerators, neural compression. OpenAI’s Ilya Sutskever says “scaling the right thing matters more now than ever”—a significant shift from pure scaling.
Across the field, interpretability is moving from research curiosity to capability race. Researchers are working to understand more clearly what’s happening inside models, and finding circuits that handle arithmetic and context. This may soon become an industry expectation.
And there’s the foundational unknown: can current architectures achieve generalized abstraction and counterfactual reasoning, or are they fundamentally limited?
Multimodal models are forming genuine unified representations across modalities, that is, the capability that determines whether agents can actually operate in trading environments, reading dashboards and interpreting market state.
And there’s the foundational unknown: can current architectures achieve generalized abstraction and counterfactual reasoning, or are they fundamentally limited? This is the line between powerful copilots and truly autonomous agents. Anyone claiming certainty at this stage is, I believe, overconfident.
A good Bayesian continuously updates their priors, and intellectual humility about what we still don’t know remains appropriate.
Ben Wellington, Head of Complex Feature Engines
AI has already changed the way we approach challenges. For example, one challenge we have is turning raw data into novel and predictive features. AI has turbocharged this ability for us, creating sets of insights that would have taken months to create in just days.
Jin Choi, Head of Technique Forecasting
AI will undoubtedly boost productivity for everyone. It’s like each of us getting a team of highly capable, tireless assistants that can process a vast amount of information, synthesize insights, and automate mundane tasks.
That said, I don’t believe AI will magically solve the key challenges in quantitative investing: predicting the future from past data, while avoiding overfitting to backtesting results and navigating regime changes. If anything, these challenges become more difficult with powerful AI.
With AI agents, researchers can easily generate a large number of hypotheses and backtest them, which can exacerbate the overfitting issue. And when researchers use pre-trained LLMs in forecasting pipelines, we need to be especially cautious trusting backtest results that predate the model’s knowledge cutoff. The model may implicitly “know” about major regime shifts (e.g., the pandemic, the AI boom), making it hard to evaluate its robustness to unseen regimes.
This is why skilled ML/AI researchers remain crucial, even as we get access to more powerful LLMs. We still need rigorous research discipline and careful production monitoring, just as we did when deploying our first neural networks years ago. I believe our deep institutional knowledge in safely developing and deploying ML/AI systems will continue to be a key differentiator.
Mike Schuster, Head of AI Core
It appears safe to say that many more financial firms will begin to use neural networks, reinforcement learning and LLMs in the coming years if they are not already doing so. However, it is hard to say whether that is going to be positive or negative for new players. Some firms likely will not succeed in embracing the new technology.
Excellent talent is hard to find and keep. I feel it is a key challenge for many firms to attract and retain those employees, such that they are willing to push boundaries enough within the company to really make large steps forward.
While technology keeps advancing, a lot about how we approach important challenges may not change so much. The general path is still going to be the same as it has been from before AI: Managers will still be trying to hire good people, focus (as tech companies typically do) on collaboration, company performance over individual performance, developing excellent engineering skills, understanding the latest research, etc..
Financial companies, just like all others, will keep trying to automate workflows and mundane tasks as much as possible. But for some, a key risk could be to believe the hype too much and spend too much effort in trying to solve ALL problems prematurely with GenAI (LLMs)—without being sufficiently skeptical.
The takeaway
As several colleagues note above, nobody can be totally sure what the future of AI in investment management (and beyond) will bring. What’s certain, however, is that 2026 will herald new capabilities, new challenges, and new uncertainties—and Two Sigma will continue working at the frontier.
The pace of change is accelerating, and the potential impact keeps growing. If you’re interested in being part of that work, we’d love to hear from you! Visit our Careers page to learn more.