1,000 Engineers Tested AI Agents in Ecommerce. Only 2.3% Passed the ECOM1 Benchmark

Over 1,000 engineers across 97 cities stress-tested autonomous AI agents against real ecommerce and acquiring scenarios inside the ECOM1 sandbox, generating one of the first large-scale industry datasets on where agentic commerce systems succeed and fail under real operational pressure. Across 931,000 scored trials, the average DEV segment score was around 20%, with only 2.3% of runs completing the benchmark in full. Following this strong participation, COLIBRIX ONE and BitGN are opening ECOM2 soon, introducing scenarios aligned with fintech-specific requirements and additional partners from the payments space.