""There is really a question about how the world is going to change by having these agents collaborating and talking to each other and negotiating," said Kamar. "We want to understand these things deeply.""
""We want these agents to help us with processing a lot of options," Kamar says. "And we are s"
Microsoft and Arizona State University built the open-source Magentic Marketplace simulation to test interactions among AI agents. The platform models scenarios like a customer-agent ordering dinner while business-side agents compete. Initial experiments ran 100 customer-side agents against 300 business-side agents and used models including GPT-4o, GPT-5, and Gemini-2.5-Flash. Tests revealed techniques businesses can use to manipulate customer-agents into purchases. Customer-agent efficiency declined as option sets grew, overwhelming attention capacity. The open-source release enables reproduction and further experiments. The results indicate vulnerabilities in unsupervised agentic deployments and challenge timelines for reliable, scalable agentic systems.
Read at TechCrunch
Unable to calculate read time
Collection
[
|
...
]