I’ve been digging into Databricks Agent Bricks, a no-code agent builder for data analytics within Databricks. My initial impression is that it has limited use cases and is quite expensive. I even had to reach out to their dev team through a personal connection to resolve permission and build errors.
I’m curious if anyone else is using this product or similar ones like Amazon Bedrock Knowledge Bases and Data Automation.
Here’s a summary of my findings:
Databricks Agent Bricks supports four types of agents: information extraction, custom LLM, knowledge assistant, and multi-agent supervisor. All data used to build these agents needs to pre-exist in the user’s Unity Catalog, with some agents requiring vectorized data sources.
The no-code interface makes it easy to define agent tasks in natural language and select data sources from Databricks Unity Catalog. Agent Bricks then generates agents automatically, but the generated code is not visible or downloadable.
The platform also provides automated metrics and in-depth analysis, generating metrics based on user-specified tasks and data. Users can select and/or edit metrics, and Agent Bricks evaluates all specified data and reports a detailed scoreboard.
Another key feature is automated cost and throughput optimization, which can take over an hour and $100+ but results in cheaper and faster serving of optimized agents.
Because Agent Bricks is built on the Databricks platform, it inherits robust governance and security features, including Unity Catalog for managing data and AI assets.
The strengths of Agent Bricks lie in its ease of use, speed to production, data integration, and unified platform. However, it also has limitations, such as vendor lock-in, limited use cases, lack of transparency, and being a costly and opaque beta product.
What do you think about using no-code agent builders like Agent Bricks? Have you had any experience with similar products?