From Lakehouse to Agentic Platforms
Databricks’ strategic shift toward agentic systems is a move to turn its enterprise data lakehouse into a foundation where AI agents, not people, are the main operators of data and applications. Instead of treating AI as an add-on to analytics, Databricks is rebuilding its stack so agents can read live data, reason over history, and act in real time across marketing, operations, and software development. This pivot moves the company beyond its origins in data warehousing and analytics and into markets like marketing technology, security, and developer tooling. By unifying data, models, and governance in one platform, Databricks aims to compete with long‑time database and observability players such as ClickHouse, Splunk, and Oracle, while positioning itself as an AI agent development hub where autonomous marketing systems and operational agents run directly on the same governed data.
CustomerLake: An Agentic CDP Built Into the Lakehouse
CustomerLake is Databricks’ new agentic CDP platform, built natively on its enterprise data lakehouse and governed by Unity Catalog. Databricks describes it as a “workforce of agents” that continuously analyze behavior, make decisions, and act to deliver always‑on personalization, with the company claiming these agents can serve personalized experiences 1 billion times a day. CustomerLake folds identity resolution, audience building, campaign automation, and activation into the same data and AI environment that powers analytics. This is aimed at marketers who must both use internal AI agents and market to customer‑side agents that research and evaluate products. Ali Ghodsi, Databricks’ co‑founder and CEO, argues that most legacy CDPs follow a slow, waterfall model and keep data siloed from AI systems, while the agentic era requires real‑time context and execution built directly into the core data platform for autonomous marketing systems.
Agent Bricks: A Developer Platform for AI Agents
Agent Bricks has grown from an experiment into a full agent platform for AI agent development on Databricks. Since launch, customers have built over 100,000 agents, and Databricks reports that these agents process more than 1 quadrillion tokens per year. The platform addresses what Databricks calls the “missing 99%” of agentic systems: deployment, token capacity, security, evaluation, monitoring, context, and sharing. Developers get model choice across frontier proprietary models and open‑source options, including OpenAI, Anthropic, Gemini, Qwen, Kimi, and Grok models via a SpaceX partnership, all within a single security boundary. Agent Bricks focuses on three challenges: choice of models and sub‑agents, reliable access to context across messy enterprise data, and control over permissions and costs for some of the most privileged actors in the organization. Databricks handles infrastructure so developers can concentrate on building and scaling impactful agents.
LTAP: Making AI Agents the Primary Database Users
With its LTAP (Lake Transactional/Analytical Processing) architecture, Databricks wants AI agents to be first‑class users of enterprise data, collapsing the long‑standing split between transactional and analytical systems. LTAP builds on Lakebase, Databricks’ Postgres‑based operational database, and unifies transactional and analytical data in a single storage layer stored in open formats on cloud object storage, while keeping separate compute engines tuned for each workload. Ali Ghodsi characterizes LTAP as a breakthrough that “removes” the infrastructure tax teams paid for decades by eliminating duplicated data and brittle ETL pipelines. Databricks argues that previous HTAP and “zero‑ETL” approaches still left two copies of data or high lock‑in, which slowed agents that need live records and historical context together. In Databricks’ vision, agents can read and write operational data, run analytical queries, and trigger actions without leaving the governed LTAP environment.
Ecosystem Expansion and Competitive Stakes
Taken together, CustomerLake, Databricks Agent Bricks, LTAP, and the Apps on Databricks Marketplace signal a broad expansion from a data and analytics company into a platform for agentic applications. Apps on Marketplace allow third‑party developers to distribute agentic applications that run directly on customers’ governed data and models, turning the lakehouse into an application surface, not only a storage and compute layer. This positions Databricks to compete more directly with databases like ClickHouse and Oracle, as well as observability and security platforms like Splunk, by offering an integrated environment where analytics, operations, and AI agents converge. For enterprises, the promise is a single, governed stack that supports agentic CDP capabilities, autonomous marketing systems, and AI agent development without cloning data across tools. The risk is a deeper dependency on Databricks as both the data platform and the primary runtime for AI agents.
