Ebrahim Aliabadi Farahani
I’m a research data scientist at Google. I received my PhD in economics from the University of Texas at Austin.
My research is in mechanism design, auctions, and the economics of marketplaces. Lately I’ve been focused on what happens when AI agents interact with one another.
Research
Do LLM Agents Respond to Commitment? A Behavioral Experiment in Strategic Communication
Working paper, November 2026
As AI agents increasingly make decisions and communicate on behalf of individuals and organizations, understanding how they respond to institutional constraints becomes important. Commitment can make strategic communication informative by restricting how a sender may revise its communication after learning private information. We study whether LLM agents respond to mechanically enforced commitment as economic theory predicts, using the sender-receiver environment of Fréchette, Lizzeri, and Perego (2022). The experiment varies commitment strength and message verifiability, allowing comparisons with theoretical predictions and existing human experimental evidence. We find substantial differences across model configurations. Claude Sonnet 4.5 exhibits the commitment-blindness pattern documented among human subjects, whereas Claude Opus 5.5 and Gemini 3.8 Flash behave much closer to theory under full commitment. Under partial commitment, even models that make individually sensible revision decisions may choose initial communication plans that fail to anticipate those decisions. Our findings highlight the importance of evaluating strategic AI agents under different institutional arrangements rather than inferring their behavior from performance in a single environment.