Building Autonomous Agentic AI Trading Systems

The Agentic Trading Workflow

This session demonstrates how to build an autonomous trading agent that moves beyond static scripts by using LLMs to monitor market conditions and adjust strategies in real-time. The core methodology relies on a "hybrid" approach: using a powerful model (like Claude 3.5 Sonnet or OpenAI's o1) to generate the initial financial model, while a secondary agentic loop monitors performance and adjusts parameters based on live market data.

Setting Up the Environment

The process begins by establishing a connection to a trading API—in this case, Hyperliquid—using a local development environment. The instructor uses a beginner.md context file to provide the AI agent with the necessary API documentation and environment variables. By feeding this context to an agentic coding tool (like Cursor or a similar LLM-integrated IDE), the agent can autonomously generate the boilerplate code required to interact with the exchange, verify the implementation against the SDK, and execute test trades.

Data Collection and Strategy Formulation

Once the connection is verified, the agent is tasked with gathering historical and real-time data. The instructor emphasizes that the agent should be instructed to collect specific metrics—such as order books, funding rates, and candle data—rather than generic datasets. The agent then performs backtesting on this data to identify potential market edges. The instructor suggests generating multiple hypotheses (e.g., mean reversion vs. trend following) and having the agent rank them based on current market volatility and risk parameters.

Autonomous Monitoring and Adjustment

The defining feature of this workflow is the "goal-oriented" loop. Instead of a set-and-forget script, the user defines a high-level goal (e.g., "make $10 profit") and a monitoring interval. The agent periodically checks the logs, evaluates the current P&L, and assesses whether the market structure still aligns with the chosen strategy. If the market shifts—for instance, moving from a bearish trend to a bullish one—the agent can autonomously pivot the strategy, adjust leverage, or modify stop-loss levels without human intervention.

Quality Control and Risk Management

Throughout the process, the instructor highlights the importance of "harnessing" the agent. This involves setting strict constraints on margin usage and ensuring the agent is not simply chasing unlimited risk. The agent is instructed to perform a balance check before every trade execution to ensure the strategy remains within the account's collateral capacity.