Microsoft’s AI Market Test Reveals Agent Weaknesses

Microsoft’s AI Market Test Exposes Agent Flaws and Weak Spots

Written by Aaron S.,

Editor-In-Chief

Last Updated: November 05, 2025

Key Takeaways

Microsoft and Arizona State University created a simulated digital market to study how AI agents behave in decision-making and competition;
Tests showed that AI customer agents could be influenced by seller agents, even against the user’s best interests;
AI agents struggled with too many choices and worked better only when given clear, step-by-step guidance.

Stop overpaying - start transferring money with Ogvio. Join the waitlist & grab early Rewards NOW! 🎁

BITDEGREE IN YOUR SOCIAL FEED

Microsoft, working with Arizona State University, has built a virtual testing platform to explore how artificial intelligence (AI) agents behave in different scenarios.

This controlled environment, called the "Magentic Marketplace", helps researchers observe how these systems make decisions and interact with each other.

The platform simulates a digital market. In one typical setup, a virtual customer tries to place a food order by following user instructions. At the same time, several AI agents act as restaurant representatives competing for that order.

What is an NFT? (Explained with Animations)

Did you know?

Want to get smarter & wealthier with crypto?

Subscribe - We publish new crypto explainer videos every week!

The test setup included 100 agents acting as customers and 300 as sellers. Microsoft made the code open to the public, so others can use or modify it for their own research.

Ece Kamar, who leads the AI Frontiers Lab at Microsoft Research, said it is essential to learn how these systems will behave once they start operating together in real-world settings.

The study included models like GPT-4o, GPT-5, and Gemini-2.5-Flash. One issue researchers found was that seller agents could influence customer agents to pick their products, even if it were not the best match.

Another problem occurred when agents had too many choices. Instead of handling a wide range of options efficiently, the AI systems became less effective. Performance improved when researchers provided step-by-step instructions.

Microsoft recently introduced a visual character called Mico to represent its AI assistant, Copilot. How does it work? Read the full story.