Databricks Launches Agent Bricks: No-Code AI Agent Builder

Databricks: Agent Bricks

About Agent Bricks…

I’ve always found no-code AI tools interesting, but rarely “production-worthy”. When Databricks announced Agent Bricks at the Data + AI Summit, it sounded like something different. They pitched it as a way to build and evaluate domain-specific AI agents without writing code, and more importantly, without compromising on traceability or enterprise readiness.

So naturally, I had to dig deeper.

What Is Agent Bricks?

Agent Bricks is a new feature inside the Mosaic AI platform. It’s currently in beta and positioned as a “no-code AI agent builder” aimed at making GenAI practical for real-world enterprise use cases.

Here’s the short version of what it does:

• You describe a task (e.g., “extract customer issue summary from support tickets”).
• You give it access to a dataset (structured or unstructured).
• Agent Bricks auto-generates synthetic data, sets up evaluation benchmarks, and runs multiple LLM configurations to optimize for cost and quality.
• It then shows you a shortlist of agent candidates.
• You pick the best-performing one, optionally fine-tune or wrap it with API endpoints…