TLDR: Databricks has launched Agent Bricks, a new platform designed to automate and streamline the development of high-performing, domain-specific AI agents for enterprises. This innovative tool leverages methods like Test-time Adaptive Optimization (TAO) and Agent Learning from Human Feedback (ALHF) to simplify evaluation, optimize performance, and manage costs, addressing key challenges in bringing AI agents to production.
Databricks has announced the introduction of Agent Bricks, a groundbreaking new tool aimed at revolutionizing enterprise AI development. Unveiled on July 28, 2025, Agent Bricks is designed to automate the complex process of creating and deploying domain-specific AI agents, enabling businesses to achieve production-grade AI with greater speed and confidence.
The platform directly addresses the significant challenges enterprises face in bringing AI agent prototypes to full production, particularly concerning evaluation difficulties and the overwhelming number of optimization parameters. According to Databricks, many organizations currently rely on manual, inconsistent, and costly trial-and-error methods, leading to stalled projects and a lack of confidence in AI deployments.
Agent Bricks operates through a sophisticated four-step automated workflow. Users begin by declaring their task, providing a high-level natural language description of the agent’s objective, and connecting their relevant data sources. Following this, Agent Bricks automatically initiates evaluation by creating task-specific benchmarks, which can involve synthetically generating new data or building custom Large Language Model (LLM) judges for quality assessment.
A core innovation of Agent Bricks lies in its automatic optimization phase. The system intelligently searches through and combines various optimization techniques, including prompt engineering, model fine-tuning, reward models, and Test-time Adaptive Optimization (TAO), to achieve high quality. TAO is a novel model tuning method that requires only unlabeled usage data, allowing enterprises to enhance AI quality and reduce costs using their existing data. Databricks emphasizes that this automated approach often results in solutions that are both higher quality and more cost-effective compared to traditional DIY methods.
Furthermore, Agent Bricks incorporates Agent Learning from Human Feedback (ALHF), a key innovation powered by Databricks Mosaic AI Research. ALHF allows for rich context from natural language guidance to be translated into technical optimizations, such as refining retrieval algorithms, enhancing prompts, or modifying agentic patterns. This mechanism helps overcome issues like ‘prompt stuffing’ and ensures that agents are continuously refined based on human input.
Databricks states that Agent Bricks ensures agents are not only highly effective but also cost-efficient, offering users the flexibility to choose between cost-optimized or quality-optimized models. The platform also integrates with Databricks’ Unity Catalog, providing robust data governance features, access controls, and data lineage tracking to ensure compliance and enhance trust in AI agent behavior.
Also Read:
- CoreOps.AI Unveils AgentCORE: Revolutionizing Enterprise Workflows with Intelligent AI Agents
- Strategic AI Advisors Launches RemoteAgent: A New Era of Autonomous AI for Business Intelligence
Joel Minnick, Databricks’ Vice President of Marketing, highlighted the critical need for robust evaluation, stating, ‘One of the biggest things that keeps these models from getting into production is that there’s no good way to evaluate whether or not agents are going to do what you expect them to do.’ Agent Bricks aims to eliminate this guesswork, providing a systematic and reliable approach to building and optimizing intelligent agents for real-world business applications. The platform is currently available in public beta.
