Definition
Plain language
A test suite for AI agents that have to complete tasks inside web-based games and interfaces.
As stated in the literature
A web-agent benchmark used as a generalization check for browsing agents, cited alongside Mind2Web.
Why it matters: It matters as a check on whether a browsing agent's skills hold up on tasks it wasn't trained for, not just familiar ones.
For example, an agent might be dropped into an unfamiliar web-based game and judged on whether it can figure out the controls and complete the objective.
Heard on the show
“They don't run it on alternative benchmarks like WebGames or Mind2Web either, which would have been useful generalization checks.”Episode 008 — Why Long-Horizon AI Agents Get Stuck, and a Milestone-Based Fix That Helps