werewolf-benchmark
(2)
Raphaël Dabadie Is Teaching AI to Read the Room
Before Foaster tried to map a company, its CEO made language models bluff through Werewolf and vote in elections. The same curiosity now points at a harder puzzle: how work really gets done.
Foaster.ai
Foaster.ai is a San Francisco startup building an AI-native alternative to management consulting. Its AI agents run 30-45 minute interviews across a company to rebuild an 'operational graph' - a structured map of workflows, handoffs, bottlenecks and information flows - then surface where AI should be deployed first. Human experts review and prioritize the roadmap. Founded in 2026 by Raphael Dabadie and Alexandre Combes, the company joined Y Combinator's Spring 2026 (P26) batch after building two widely shared LLM benchmarks, including the Werewolf Benchmark for social intelligence.