Key takeaways from day two of the Databricks Data + AI Summit

After unveiling a number of key announcements at Databricks’ Data + AI Summit in San Francisco on Tuesday, the company added further context and messaging around its latest product and service introductions during a three hour keynote session on Wednesday.

Here are five key takeaways from day two:

1. Databricks views the cost of AI as a significant issue for the enterprise.

Across both days of the summit, Databricks executives repeatedly referred to the escalating cost of AI as organizations seek to reign in expenses surrounding token usage and agentic development. In a briefing for analysts this week, Chief Executive Ali Ghodsi pointedly noted that Databricks itself has experienced its own cost headaches, explaining that agentic AI is increasing consumption of the firm’s information technology resources.

Ghodsi returned to this theme during his day two keynote remarks today. “This is going to get extremely expensive,” Ghodsi said. “We are just scratching the surface.”

2. Omnigent highlights the need for an agentic control platform.

The just-released meta-harness Omnigent is Databricks’ statement around the need for agentic supervision. Today the company outlined the importance of agentic combination, sharing and control, features of Omnigent to enable enterprises to manage the increasingly critical software surrounding language models that are an agent’s interface with the world.

“It’s a harness of harnesses,” Databricks co-founder and Chief Technology Officer Matei Zaharia told the summit gathering. “We need a new open layer above the harness level.”

3. Developers can play a central role in AI governance.

Agent Bricks, coupled with recent enhancements for Unity AI Gateway, demonstrated Databricks’ belief that developers will play a key role in AI governance. At the same time, the company wants developers to focus on building agents, not infrastructure.

The enhancements for Unity AI Gateway offered contextual security policies, an ability to monitor costs and set budgets, along with a Smart Routing feature for finding the best model to execute a task. “Agents are the most privileged actor,” said Kasey Uhlenhuth, director of product at Databricks. “This is what AI governance should look like in the enterprise.”

4. Databricks sees a market opportunity for machine learning.

In a presentation on Wednesday, Databricks Director of Product Management Mike Del Balso offered a compelling case for machine learning as a sometimes overlooked, yet highly important enterprise resource. He also made it clear that Databricks is well-positioned to lead a machine learning initiative.

The company added more details about AI Runtime, a serverless graphics processing unit for enterprise deep learning that has been in preview for the past two months. Databricks also announced Genie Code for ML and Genie Zero Ops for ML, an agent for running machine learning systems.

“What we really want to do is automate ML,” Del Balso said. “We’re entering this era where we can scale ML across all of our businesses. Nobody has seen more ML in the real world than Databricks.”

5. The marketing world is facing a significant transformation because of agentic AI.

In collaboration with large language models, agents are becoming buyers and this is reshaping how marketing and data teams build new initiatives. Databricks’ announcement of CustomerLake, an agentic customer data platform or CDP, provided yet another example of how the marketing field is being transformed.

CustomerLake is designed to help marketers move from static, one-off campaigns to infinity campaigns, agent-driven engagement loops that analyze customer signals and act across channels.

“Humans do not buy on their own anymore,” Tasso Argyros, Databricks’ vice president of engineering, said today. “Commerce is going headless. What if we could have an agent for every customer? We think this is the future.”

Photo: Mark Albertson/SiliconANGLE