At a recent Jefferies luncheon on “Beyond the Hype: Practical AI for Buy-side Decision Making,” senior leaders across finance and technology gathered to explore how AI is reshaping investment decision-making. The discussion focused on how generative AI is reshaping investor workflows, which platforms are emerging as leaders, and where adoption still hits friction.
This article highlights key insights from the keynote fireside conversation between Shawn Edwards, Chief Technology Officer at Bloomberg, and Brent Thill, Jefferies’ Tech Sector Leader for Software and Internet Research.
AI Moves from Experimentation to Daily Workflow
Major financial-data providers are moving quickly to embed Generative AI into the core of their products. Questions about how to integrate the technology into analyst workflows are top of mind — and at Bloomberg, utilization and integration into its products has been swift and transformative.
“It is enabling us to do things that we were trying to do for years,” Mr. Edwards explained. “It is an incredible, incredible technology.”
At Bloomberg, the goal has been to help increase users’ efficiency without supplanting their expertise. “(AI) doesn’t replace a good analyst,” he explained, but it can free them from time-consuming tasks so they can spend more time on the more strategic, interpretive work.
This isn’t necessarily an equalizer, Mr. Edwards said; not everyone will suddenly become a competent analyst. It’s more of an accelerant for Bloomberg Terminal users who are already good at their jobs — giving them more bandwidth and sharper capabilities.
Agentic Systems Built for Analysts
Bloomberg is now deploying agent-driven systems that synthesize research, validate facts, and deliver analyst-grade answers on demand. “You can type in your high-level thesis and questions into our system, and our agents . . . synthesize an answer,” Mr. Edwards said.
The vision is to remove the need for analysts to know which database or function on the Bloomberg Terminal holds the right information. This kind of information-management orchestration has been one of the most effective early uses of generative AI. For analysts across the industry, the system handles data routing and retrieval, eliminating the manual logistics that typically slow down research. Even more impactful than finding information is its ability to synthesize cohesive answers from the broad range of available data and documents.
Early testing with Bloomberg’s customers shows that complex workflows — such as comparing revisions across filings or pulling historical trends — can shrink from hours to minutes. Analysts can spend more of their time on client-facing work: judgment, interpretation, and storytelling.
Accuracy and Guardrails
Because generative AI is still relatively new inside financial workflows, reliability remains a central concern. Mr. Edwards stressed that Bloomberg built layers of validation to make sure the system doesn’t push unreliable or hallucinated content into production. “We had to build all these validators,” he said. “Every single agent, every single subsystem… there’s lots and lots of techniques and checks in real time.”
Mr. Edwards noted that the engineering challenge wasn’t just building the agents, but making sure all of the checks could run without adding latency that slows analysts down.
A Model-Agnostic Approach
One thing that differentiates Bloomberg’s approach is its model-agnostic infrastructure.
“Latency and cost matter, so we evaluate models continuously and shift workloads accordingly,” Mr. Edwards said. The company relies on different models, hyperscalers, and latency tradeoffs to build a resilient system.
“We’re not tied to any one model. Different tasks need different strengths, so we route to the model that performs best for that job.”
The Future of AI and Financial Data
Bloomberg’s success signals that AI has become operational infrastructure for leaders in financial data and capital markets around the globe. The hype that surrounded early adoption efforts, Mr. Edwards said, has given way to “capabilities that we’ve never had before.”
It also offers a practical model for capital-markets firms looking to adopt agentic AI. To this audience of CIOs and CTOs, Mr. Edwards offered a straightforward piece of advice: start with one specific workflow and build around it.
“Start with something very specific and understand what you’re trying to solve. Define your criteria for how you’ll evaluate results. And then when you focus on something specific, you’ll learn to quickly iterate in order to get measurable results.”
He cautioned that large language models can be unreliable unless they’re properly constrained and validated. Beginning with a narrow, well-defined use case creates the feedback loops needed to build trust and refine the system.
At a moment when firms are racing to adopt AI as quickly as possible, it’s a useful reminder of the work required to generate real, dependable value from these tools.
