Databricks Unveils Mosaic Agent Bricks for Scalable AI Agents

Most AI agent pilots never hit production — not because the technology isn’t ready, but because evaluation remains manual, slow and inconsistent.

Automated Agent Optimization with Mosaic Agent Bricks

At the Data + AI Summit, Databricks launched Mosaic Agent Bricks, extending its 2024 AI Agent Framework. The platform automates end-to-end optimization so teams can trust agents in production.

• TAO integration for test-time adaptive optimization without labeled data
• Domain-specific synthetic data generation to mirror enterprise inputs
• Task-aware benchmarks and custom LLM judges built automatically
• Automated quality-to-cost tuning without manual tweaks
• Agent Learning from Human Feedback to replace ‘prompt stuffing’

Four Tailored Agent Configurations

• Information Extraction — converts PDFs, emails and docs into structured records (e.g., retail orgs pulling product specs from supplier files)
• Knowledge Assistant — delivers cited answers from enterprise data (e.g., technicians querying maintenance manuals instantly)
• Custom LLM — text transforms like summarization or classification (e.g., healthcare teams auto-summarizing patient notes)
• Multi-Agent Supervisor — orchestrates agents for complex workflows (e.g., finance firms coordinating intent detection, retrieval and compliance)

Built on a Unified Data and Governance Foundation

Mosaic Agent Bricks sits atop Databricks’ Lakeflow for unified ingestion, transformation and orchestration. Governed by Unity Catalog’s access control and lineage, agents respect enterprise policies out of the box.

Strategic Impact for Enterprise Leaders

With automated evaluation and tuning, teams can shift from costly trial-and-error to focusing on high-value use cases and data prep. Agent Learning from Human Feedback ensures steerable, production-ready agents that align with business goals.

By closing the agentic AI evaluation gap, Mosaic Agent Bricks transforms enterprise rollout timelines. Organizations can now deploy, iterate and govern AI agents at scale with confidence.