← Back to Vault

Benchmarks vs Real Risk

Tom Spencer · Category: points_of_view

AI agent benchmarks using trivial tasks are "cutesy" and don't reflect real enterprise applications with high stakes and sensitive client data.