A physics researcher conducted a one-week live trading experiment using an agentic Claude-based bot to manage a $500 equity-only portfolio. The bot achieved a 2.20% return ($510.65 total), slightly outperforming VOO (1.88%) and QQQ (2.05%) during a strong tech week. Rather than a simple chat interface, the architecture utilizes an agentic pipeline where Claude uses tool-calling to execute fractional share market orders via a brokerage API. The experiment highlights a critical distinction between two potential implementation methods: Version A, where the LLM performs adaptive reasoning by synthesizing unstructured natural-language market analysis to adjust position sizing, and Version B, where the LLM acts as a convenient wrapper for executing deterministic, fixed-rule scripts. While the bot successfully avoided the behavioral pitfalls of manual retail trading—such as overtrading and emotional decision-making—the study notes that the marginal cost of using Claude Pro ($200/year) sits atop existing infrastructure costs. Ultimately, the results do not prove the LLM's analytical superiority over a standard Python script, but rather demonstrate the effectiveness of removing human behavioral bias from the execution loop. Further testing is required to determine if the LLM's reasoning provides genuine alpha during market divergence or if a deterministic script would yield identical results.