The bull and the bear, 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

Lesson 2 ended with four analyst reports in hand. This lesson is about what the team does with them, because information is not yet a decision. Before anyone commits, the system stages an argument: one agent builds the strongest case for, another builds the strongest case against, and a separate judge weighs the two and decides.

Reading the real code, you will see the shape of it. The bull and bear researchers get the same evidence but opposite mandates, and each is told to rebut the other’s latest point, so it is a genuine back-and-forth rather than two monologues. The debate does not run forever: a short function checks a turn counter against a round limit after every turn and, once the limit is reached, hands the transcript to the judge. The judge (the Research Manager) is a different agent on the more capable model; it reads both cases and must commit to one of five stances. The throughline is transferable: a debated, then judged, decision beats a single confident opinion, and the agent that decides should never be one of the agents that argued.

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

Where this fits

This is the third lesson of the track. Lesson 1 was the map; lesson 2 showed how a single analyst gathers; this lesson is the first time several agents interact. Next, lesson 4 follows the judge’s plan to the trader, who turns it into a concrete course of action, and the track continues through the risk gate, orchestration and shared state, memory, and the hands-on capstone.

Before you start

Prerequisites: lesson 2 of this track (how an agent gathers with tools), which in turn assumes lesson 1 (the team and why it is split). You do not need to install or run anything; the project can be read in your browser, and every excerpt is quoted in the lesson.

By the end, you’ll be able to

Explain why structured disagreement produces a better decision than a single agent’s opinion
Describe how the bull and bear researchers argue the same evidence with opposite mandates, each rebutting the other
Identify how the debate is bounded by a turn counter against a round limit, and when the judge takes over
Analyze why separating the judge from the advocates, and running it on the more capable model, improves the decision
Apply the debate-then-judge pattern to get a reasoned decision from your own agents

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)