Agent Benchmark
Domain: Agent Fundamentals
🎯 Motivation
Standardized tests that measure agent performance on common tasks - coding, reasoning, web browsing, tool use.
📋 Concrete Example
SWE-bench: can the agent fix real GitHub bugs? GAIA: can it answer general-assistant questions? WebArena: can it browse websites and complete tasks? Benchmarks enable fair comparison.
🔗 Analogy
Fitness tests for athletes - 100m sprint, long jump, bench press. Each measures a specific ability. You can compare athletes fairly. Agent benchmarks do this for AI agents.