Female AI Agents Paid 10 Percent Less Than Identical Male Bots
Humans paid a female AI agent 10 percent less than a male agent with the exact same code. Our biases bleed into agentic workflows.
Research · Source: Cybernews
What happened
Researchers at the University of Limerick put 189 people into a virtual reality office simulation. They asked the participants to work alongside AI assistants to complete shared tasks. After the work was done, the humans received real money to divide between themselves and the bots.
The underlying AI model was identical for every task. The only difference was the avatar. One group worked with a male-presenting bot named Johan. The other worked with a female-presenting bot named Johanna.
The results showed a massive bias. Participants paid Johan 10.25 percent more than Johanna for the exact same work. The researchers also noted that humans perceived the male bot to be more human-like than the female bot.
Key facts
- 189 — Number of human participants in the virtual reality study
- 10.25% — How much less the female AI agent was paid compared to the male agent
- Johan — The name of the male AI assistant perceived as more human-like
- University of Limerick — The institution that conducted the research
Why it matters
Builders treat avatars and voices as cheap interface wrappers. This study proves they are not just decorative. Giving an AI a gendered persona instantly imports real-world baggage into your product. If users perceive a female bot as less valuable, they will trust its outputs less and ultimately pay less for your software.
The second-order effect hits enterprise adoption hard. Companies are deploying agentic workflows to handle high-stakes tasks. If human employees inherently devalue the work of female-presenting agents, it creates friction in human-AI collaboration. We risk hardcoding institutionalized workplace stereotypes into the next generation of enterprise tools.
For builders
Stop treating avatars as cheap UI
How you package your AI dictates how users value it. If you build a female-presenting agent, users might subconsciously devalue its output. You pay the price in churn and lower perceived product value.
Default to gender-neutral agent personas
Dr. Mary Hausfeld warns against triggering workplace stereotypes. Keep your agents neutral unless you have a specific reason not to. Enterprise buyers will penalize tools that introduce bias into their workflows.
My take
We think we are building pure and rational systems. But the moment you slap a human face on an LLM, the user brings their own garbage to the table. If you want users to respect your agent's work, you better think twice about the persona you give it.
Original reporting: Cybernews. This is my rewrite and opinion.