Effectively adding AI trading technology to your risk conversations requires more than just a single agent operating as your AI Assistant. An effective AI team is made up of a variety of specialized agents that work together, designed, controlled and orchestrated by their human leader to tackle the multi-step and multi-role activities that are part of real-life trading and portfolio management.
From tools to skills
Beacon by CWAN’s Workflow Studio is the access point to a growing set of AI tools, skills, and agents that you can use out of the box, customize, or add to, to suit your work environment and trading edge.
Beacon’s embedded intelligence service and trader AI applications include a number of components that help you scale from individual tools to high-performance AI teams.
Tool are essentially Python code functions with the appropriate decoration, input, and output to be called by an agent.
To add a new tool, Coding Agent can be used to create one without writing any code. For example, ask AI Assistant or Coding Agent to “write an AI tool for pricing options using Black Scholes, including Greeks and volatility”. The resulting tool provides three outputs: pricing of call and put options, calculation of major Greeks, and implied volatility.
Skills are sets of tools, instructions, and rules that give the agent specific capabilities, such as risk and P&L analysis or scenario management.
Orchestrate agents into teams and workflows
Agents are the typical interface point for AI interactions and are powered by multiple skills and knowledge bases that can be mixed and matched for the desired workflows.
For example, you ask the assistant agent to get the risk on your book. First, the assistant identifies Trader Agent as the most qualified to respond to your request. Trader Agent calls report_runners to identify relevant reports and the parameters needed to call them and then selects get_risk to answer the question. If needed parameters are missing from the original query, the agent prompts you to provide them. The agent could also perform additional yield curve scenario analysis, such as a parallel shift, steepening, flattening, or more exotic shape changes. These scenarios can be remembered, customized, and shared among team members.
Agents can work together in teams, passing work from one to another when asking for more detailed into, publishing reports, or sending info via email. Workflows can be setup to run autonomously at a scheduled time or when a trigger event happens.
How do we monitor, track, and audit AI usage
Token Monitor is an administrative application for Beacon’s embedded intelligence that provides comprehensive info to help you monitor, track, and audit AI usage across the organization. The embedded AI database logs tokens used as a timeseries with user info, date, time, and resource type.
The Token Monitor home screen displays summary data for the selected date range, including the aggregate input and output number of tokens per user. Multiple filters and sorting capabilities enable admins to quickly find the info they need, monitor usage against daily and system limits, or drill down into detailed data per call record. The resulting data tables can be directly exported in spreadsheet format or plotted on an interactive chart to visualize trends.
Connect to other data sources and tools
CWAN’s easy integration philosophy extends to AI resources. Beacon embedded intelligence can use Model Context Protocol (MCP), an open-source intermediary that AI models use to securely communicate with external AI data sources, tools, and agents. Internal tools and resources can also be made available to external agents via the same mechanism. In addition, Beacon intelligence can bypass MCP with direct connections that offer better performance, simplicity, and support for additional features, such as table and chart artifacts.
Don’t be distracted by the hype
AI is moving along the technology hype cycle, with many proposed uses passing through the peak of inflated expectations. But amid the positive and negative hyperbole, some usage models are quietly delivering tangible benefits. Specialized agents are one of those success stories, providing live, real-time assistance and execution as your trading and quantitative strategy AI.
The future doesn’t belong to the smartest model or agent. It belongs to the framework that securely enables you to build and orchestrate models and agents that interact with each other and the broader world, multiplying and accelerating your advantages.
If you are already a Beacon client, please reach out to your CS team to get started today.
If you are interested in learning more about Beacon Intelligence, connect with us here.