– Strengthened governance, monitoring, and expansion capabilities to support more stable deployment of AI agents for enterprises
Data and AI company Databricks has unveiled new tools to help enterprises move AI agents beyond pilot phases and into large-scale operational environments, with the aim of helping enterprises more confidently and effectively leverage AI for high-value use cases.
Today, 85% of enterprises worldwide are using generative AI, but even the most advanced models struggle to deliver business-specific, systematic results due to a lack of understanding of enterprise-specific data. The new tools announced today are designed to help enterprises deploy AI agents into high-value, business-critical applications, while ensuring accuracy, governance, and ease of use. Key features include:
Centralized governance for AI models: Supports managing all AI models (open source and commercial models) in one place, and enables leveraging custom large-scale language model (LLM) providers through the Mosaic AI Gateway, providing consistent governance, monitoring, and integration capabilities across models.
Easy Integration with Existing Application Workflows : The AI/BI Genie conversational API suite enables developers to integrate natural language-based chatbots directly into custom applications or popular productivity tools like Microsoft Teams, SharePoint, and Slack. Additionally, the Genie API allows users to programmatically submit prompts and get the same insights as they would in the Genie UI. Additionally, the API is stateful, so it can maintain the context of a conversation across multiple follow-up questions within a conversation thread.
Streamlining Human-in-the-Loop Workflows: The upgraded Agent Evaluation Review App enables domain experts to more easily evaluate the performance of AI agents, provide customized feedback, send traces for labeling, and customize evaluation criteria. Experts can efficiently collect systematic feedback without Excel spreadsheets or separate customized applications, contributing to continuous improvement of AI performance and systematic improvement of accuracy. In particular, it reduces the significant time and effort required to evaluate the performance of AI agents in real-world operating environments.
Batch Inference Without Provisioning : Model selection, governance, and evaluation are essential for building high-quality AI agents, but simplifying the user experience is also important for smooth technology proliferation. The newly added capabilities enable batch inference with a single SQL query in Mosaic AI, and seamlessly integrate unstructured data without separate infrastructure setup.
“Batch AI and AI functions simplify our AI workflows. They enable us to integrate large-scale AI inference with simple SQL queries without managing infrastructure. They integrate directly into our data pipeline, reducing costs and reducing configuration overhead. Since implementing the solution, we have seen a dramatic increase in development velocity when combining traditional ETL and data pipelining with AI inference workloads,” said Ian Cadieu, CTO of Altana.
“Many enterprises still struggle to deploy AI agents for high-value use cases due to concerns about accuracy, governance, and security,” said Craig Wiley, senior director of AI/ML products at Databricks. “For these enterprises, the biggest obstacle isn’t the technology itself, but rather their confidence in whether they can trust AI.” “These new tools tackle this problem head-on, enabling enterprises to move beyond pilots to deploying trusted AI agents in full-scale production environments.”
Currently, Mosaic AI Gateway, Genie Conversational API Suite, Agent Rating Review App, and Batch AI are available in Public Preview versions.
You must be logged in to post a comment.