“Developers are paying frontier prices for amnesia”: Uvi wants to stop the agent relearning brain drain

AI agents are incurring unnecessary costs due to having to relearn company context every time they process information, according to Uvi, an Israeli startup. The company claims that in-house agents built on frontier models and connected to six enterprise systems cost around eight times more than necessary because of repeated context gathering.

Uvi co-founder and CEO Gilad Salinger says developers are effectively paying "frontier prices for amnesia" as new models are released every quarter, while the company's memory remains with people and old tickets. Uvi aims to solve this issue by creating a knowledge graph and semantic layer that sits on top of relevant, permissioned data.

By connecting six enterprise systems, Uvi claims to reduce agent costs to $0.10 to $0.16 per answer compared to $0.75 to $1.41 on average and up to $2.00 for open-ended questions, potentially saving up to 46 times. This approach allows agents to access the "smallest correct slice" of information, ensuring speed, security, and precision, ultimately helping companies stay competitive in the rapidly changing AI landscape.