Understudy cuts companies’ Anthropic bills by up to 80% without sacrificing performance. Companies spend millions renting general-purpose frontier models, but almost none of that spending compounds. Every prompt generates useful experience, yet turning production traces into evaluations, training data, and specialized models normally requires an entire ML team. Understudy automates that process. It observes AI workloads, builds evaluations, trains smaller open-weight models, and deploys them only when they outperform the incumbent on quality, cost, and latency. Customers own the resulting models and can run them in Understudy’s cloud, their own cloud, or on-premise. Founders Luis and Aamir previously built optimization and experimentation systems at Instacart that helped scale multiple business lines beyond $1B in annual revenue. Understudy Labs: YC Summer 2026: Understudy: The self-optimizing inference cloud
