Databricks focuses on 'AI doers'
Databricks, which has surpassed a $5.4 billion revenue run-rate, is focusing its strategy on turning chatbots into autonomous "doers." The company is developing GPT-5-powered agents capable of executing multi-step tasks within enterprise environments, supported by new offerings like Lakebase and Genie. This shift underscores the growing importance of building AI systems that can perform actions rather than just provide information.
- The company's revenue growth is accelerating, with its run-rate increasing from approximately 50% year-over-year at $1.6B to over 65% YoY at its current $5.4B run-rate, outpacing competitors like Snowflake. Its AI products alone now account for a $1.4 billion revenue run-rate. - The AI strategy is supported by key acquisitions, including the $1.3 billion purchase of MosaicML, which provides the foundation for custom large language model training, and Einblick, which adds natural language processing capabilities for easier data interaction. - New tools are designed to tackle the core challenges of production AI. The "Mosaic AI Agent Framework" provides tools for building and deploying agents with a focus on Retrieval-Augmented Generation (RAG) for accuracy, while "Agent Bricks" automates the complex evaluation and tuning process by generating task-specific benchmarks and optimizing for both cost and quality. - This industry move towards autonomous agents has created a new operational discipline called "AgentOps," an extension of MLOps and LLMOps. This specialization focuses on managing the unique challenges of agentic systems, such as orchestrating multi-step workflows, monitoring for behavioral drift, and tracking unpredictable token-based costs from LLM API calls. - A relevant portfolio project would be to build a multi-agent financial analysis assistant. This could involve one agent retrieving real-time stock data via an API, a second agent using a vector database and RAG to analyze financial reports, and a third to synthesize the information into a summary, with the entire workflow being orchestrated and monitored by a tool like MLflow. - For ML system design interviews, preparation should now include designing for agentic AI. Be ready to architect scalable, long-running, multi-step workflows and address the key operational challenges: ensuring observability into an agent's reasoning, handling potential hallucinations, and implementing robust error recovery and cost-control mechanisms. - CEO Ali Ghodsi emphasizes that this shift is already impacting production environments, stating that AI agents now build 80% of the databases on the Databricks platform, a task previously handled by specialized data engineers. - The underlying architecture enabling these agents is "Lakebase," a fully managed, serverless PostgreSQL database. It is designed to merge transactional (OLTP) and analytical (OLAP) data, allowing AI agents to perform low-latency actions based on real-time data while simultaneously reasoning over large historical datasets without complex ETL pipelines.
