Why split one AI into many, in brief

About this track: Clawdemy uses the open-source TradingAgents framework (by TauricResearch; Yijia Xiao, Edward Sun, Di Luo, and Wei Wang; arXiv:2412.20138; Apache-2.0) as a real-world example of how an agentic AI system is structured. Our goal is to teach the architecture: how specialist agents, a research debate, a trader role, and a risk layer are coordinated by an orchestrator. We are not teaching you how to invest, trade, or pick stocks, and we make no claim that this or any AI system is profitable or predicts markets. This content is for education only and is not investment, financial, or trading advice. Rules vary by country, and this is not advice anywhere. The Trading Agents Lab capstone is a simulation you explore yourself with your own AI provider key; it does not place real trades. The framework’s own authors state it plainly: “It is not intended as financial, investment, or trading advice.”

What you’ll learn

This is the opening lesson of the track. Where the foundational agent tracks argued in the abstract about when to split work across multiple agents, this track studies a real, running multi-agent system that made the bet, so that by the end you could design one of your own.

This first lesson is the map. It meets the team conceptually, with no code yet: how the TradingAgents framework breaks one workflow into specialist roles grouped by function (analysts who gather, researchers who argue, a trader who synthesizes, risk reviewers who stress-test, and two managers who judge), and the design decision most people miss: the system runs on two model tiers and gives the more capable, more expensive model to only the two judge roles, while everything else runs lean. The throughline is transferable: a good agent team is one workflow broken into roles, with your strongest model spent only at the points where one agent decides for the rest. The lesson also shows you how to open the project in your browser and follow along, with no account or software needed.

Everything is anchored to one snapshot of the framework (a frozen version marked 7e9e7b8), and every claim is checked against that source.

Where this fits

This track sits after the foundational agent tracks. It is the applied, buildable-depth case study: the place where the concepts (tools, multi-agent design, memory, orchestration) are shown working together in a real system. Lesson 1 sets up the cast and the map; lesson 2 opens the project for the first time and shows how the analysts gather their own information with tools, and the rest of the track walks the flow through debate, synthesis, risk-gating, orchestration, memory, and a hands-on capstone.

Before you start

Prerequisites: the foundational agent tracks, especially the multi-agent systems lesson (the direct conceptual predecessor; it argues when to split, and this track shows how a real system did). You should be comfortable with the idea of an agent as a model with a focused job and tools. You do not need to install or run anything for this lesson; it is a guided overview, and the project can be read in your browser.

By the end, you’ll be able to

Describe how a real multi-agent system splits one workflow into specialist roles by function
Identify the team’s roles (analysts, researchers, trader, risk reviewers, managers) and the job each performs
Explain why the most capable model is given only to the two judge roles and what that says about spending capability
Trace how the work flows from the analysts through two judgment points to a final decision
Apply the decompose-into-functions-then-spend-capability-at-the-judges pattern to your own agent team

Time and difficulty

Read time: about 9 minutes
Practice time: about 15 minutes (a self-check, a design exercise, and flashcards)
Difficulty: deep (this track is the advanced end of the catalog; the foundational agent tracks are the on-ramp)