Benchmark
Agent Startup Bench
Can an agent finish the work that customers already delegate to an AI startup?
Explore the original evidence
Explore the original evaluation, its methods, and reported findings.
Open evalCoverage
Seed surveyed startups with evidence of customer demand, then interviewed them to identify workflows. The report gives examples in education, finance, law and operations.
Methodology
Experts assess the quality of the finished work. Seed designed the evaluation and reports its own models’ results; public access to a complete task set or scoring rubric is unconfirmed.