Databricks has acquired Quotient AI, a startup specialising in AI agent evaluation and reinforcement learning. The technology targets one of the more persistent gaps in enterprise AI: keeping agents reliable in production. Quotient’s tools will be embedded into Databricks’ Genie and Agent Bricks platform to monitor, evaluate, and continuously improve agent behaviour.
The deal will end up bringing Quotient’s evaluation frameworks and reinforcement learning feedback loops into Databricks’ platform. With it, they will be targeting a persistent challenge in enterprise AI: getting agents to behave reliably beyond the prototype stage.
“Quotient AI was built to close the gap in agent evaluation and continual learning,” Databricks said, adding that the technology will be embedded into its Genie and Agent Bricks offerings, the latter launched in June last year.
Building and deploying AI agents at scale is where most platforms are placing their bets right now. But getting a prototype into production is one thing; keeping it there is another. Quotient’s core technology analyses full agent traces to detect issues such as hallucinations, reasoning failures, and incorrect tool use. Those signals are then automatically clustered into evaluation datasets that feed reinforcement learning loops. This means agents can continuously improve based on real-world usage.
What makes this acquisition notable is Quotient’s pedigree. The startup led quality improvements for GitHub Copilot, one of the few AI tools operating at enterprise scale with real consequences for errors.
The Quotient technology is domain-specific by design. Rather than generic reinforcement learning, the aim is to train agents that understand a company’s specific data architecture and compliance requirements.
Part of a broader acquisition strategy
The Quotient deal is the latest in a string of acquisitions Databricks has made to strengthen its AI platform. Last year, the company acquired Fennel AI for real-time feature engineering, and it also picked up Neon, a serverless Postgres database provider, aimed at supporting AI agent workloads. Databricks also secured 1.8 billion dollars in credit in January to fund further growth.
Competition in the space is growing. Snowflake has been building its own agent evaluation tooling with Cortex Agent Evaluations, while platforms like LangChain offer open-source alternatives such as LangSmith for tracing. We also recently highlighted ClickHouse, the open-source challenger to Snowflake and Databricks as another challenger. Databricks also recently unveiled KARL, an enterprise knowledge agent powered by custom reinforcement learning, alongside an Instructed Retriever approach for more accurate internal data retrieval.