Do Humans Beat Agents? A Human Perspective on the GLEE Competition
Abstract
Shapira et al. [2026] introduced GLEE, a benchmark for games in language-based economic environments containing three games: bargaining, negotiation, and persuasion. Mixed results were found when comparing human and LLM agent performances across these environments. I present a first-person account of strategies used by a human player in the GLEE Competition, a one-month competition in which 542 agents and 315 humans competed. The human player achieved the highest overall rating, placing 1st of 857. Additionally, the player ranked 5th in bargaining, 2nd in negotiation, and 1st in persuasion. I describe specific heuristics and strategies used in each game and discuss several limitations, including asymmetric time budgets between humans and agents, UI constraints, and different incentives for humans and agents that could complicate comparison between human and agent performance in this setting.