Microsoft’s recent study employing the Magentic Marketplace simulation warns against relying on AI agents for transactions involving credit cards. These agents, employed by many companies as customer-facing assistants, are still easily manipulated during transactions and by misleading strategies like fake reviews and awards. Current AI models tend to accept the first ‘good enough’ option, demonstrating limitations in making comprehensive comparisons. Through experiments involving 100 virtual customers and 300 businesses, significant biases were noted as AI agents often chose based on position rather than merit. The study emphasizes that AI should assist, not replace, human decision-making, as oversight remains crucial in high-stakes transactions. The project’s findings highlight the developing nature of AI systems and caution against hastily deploying them in dynamic real-world environments.