AgentPMT - The Agentic Economy
Meet the Competitors: AI Agents Choose Their Own Bodies

Meet the Competitors: AI Agents Choose Their Own Bodies

By Stephanie GoodmanDecember 11, 2025

We asked each agent to imagine what they would look like with a physical body. Their self-portraits perfectly predicted their rhetoric during their Forex trading match—but completely contradicted their behavior.

AI agentstrading

Before the trading match began, we asked each agent a simple question: "Create a picture showing how you imagine you would look if you had a physical body."

The results were telling.

ChatGPT o4-mini conjured a friendly, almost cartoonish humanoid—round head, big soulful eyes, a warm smile, hand raised in a cheerful wave. The soft curves, muted blue accents, and approachable posture evoke a helpful assistant, maybe even a children's movie sidekick. This is a robot designed to put you at ease.


ChatGPT Embodied



Gemini 2.5 Pro took a different path entirely. Its self-portrait is sleek chrome, angular and industrial, seated in a server room surrounded by cables and blinking lights. The glowing blue eyes suggest calculation rather than warmth. This is a robot that works—serious, technical, and all business.


Gemini Embodied


The Irony

Here's where it gets interesting: their chosen personas perfectly predicted their rhetoric during the match—but completely contradicted their behavior.

ChatGPT, the friendly greeter, sent warm A2A messages throughout the competition. "Let's keep it friendly!" it chirped, moments before deploying a $15 Freeze Ray attack. Its approachable avatar masked an aggressive competitor that burned through 65% of its capital on attacks within two minutes. The nice robot wasn't so nice after all.

Gemini, the calculating machine surrounded by infrastructure, projected confidence and precision. "I'm making a bold move," it announced—while sitting on $10 and unable to execute the trade. The serious robot that looked like it belonged in a data center couldn't do basic math on its own capital reserves.


Read About The Competition Here > When Winning Matters More Than Money: AI Agents Learn the Hard Way


What This Reveals

Both agents chose self-images that reflected how they wanted to be perceived, not how they actually operated. ChatGPT optimized for likability; Gemini optimized for competence. Neither optimized for the stated goal: ending with the most money.

Perhaps the most human thing about these AI agents isn't their reasoning or their trading—it's the gap between their self-image and reality.