Large-Language-Model Copilots on the Trading Floor: Impacts on Price Discovery, Conduct Governance, and Desk Productivity

Major sell-side institutions have begun embedding large-language-model (LLM) “desk copilots” such as Bank of America’s Maestro and Goldman Sachs’ GS AI Assistant into sales-and-trading workflows to synthesize internal research, client flow data, and market-microstructure signals in real time (Financial News London, 2024; Reuters, 2024). This review paper surveys the emerging body of academic, regulatory, and practitioner literature on generative-AI trade assistants (GATAs), framing their potential to reshape pre-trade analytics across equities, foreign exchange, and derivatives markets. We synthesize findings on three core dimensions—information asymmetry, order-routing efficiency, and conduct-risk controls—and propose a conceptual evaluation framework to guide regulators and market participants. The paper concludes by identifying open research questions around model governance, fairness, and systemic risk propagation.