Databricks Makes Custom AI Agents a First-Class Citizen in the Enterpr

When Ali Ghodsi took the stage at Databricks’ annual conference last year, he made a prediction that sounded almost mundane in the moment: AI agents would soon become as fundamental to enterprise software as databases are today. This week, his company took a significant step toward making that vision real.

Databricks announced that Agent Bricks Custom Agents—formerly known as Agent Framework—is now generally available. The move transforms how developers build, test, and deploy AI agents within the enterprise, treating them not as experimental side projects but as production-grade applications with the same governance and infrastructure standards as any other critical business system.

“Developers don’t need to re-architect code or manage infrastructure. They can build agents locally, use AI coding tools, and iterate quickly before deploying.” — Databricks Engineering Team

A $43 Billion Bet on AI Infrastructure

Databricks’ announcement comes at a pivotal moment for the AI industry. The company, valued at $43 billion in its last funding round, has spent years positioning itself as the neutral ground where enterprises can build AI without surrendering their data to closed ecosystems. Custom Agents represent the culmination of that strategy.

The platform allows developers to build agents using their preferred models and frameworks—whether that’s OpenAI’s GPT models, open-source alternatives like Llama, or specialized domain models—and deploy them as fully managed Databricks Apps on serverless compute. This flexibility matters because enterprises are increasingly wary of vendor lock-in, especially as AI capabilities become central to their competitive advantage.

Built-in memory powered by Lakebase gives agents production-grade state management. Unlike simpler chatbots that treat each interaction as isolated, these agents can maintain context across sessions, recall previous conversations, and build persistent knowledge about users and workflows. For enterprises, this means agents that actually get smarter over time rather than starting from scratch with every interaction.

CI/CD integration ensures that agents can follow the same development workflows as traditional software. Teams can version control their agent logic, run automated tests, and deploy updates through familiar pipelines. This might sound obvious, but it’s a capability that has been notably absent from many AI agent platforms, which often require manual deployment processes that don’t scale.

The Governance Question

Perhaps the most significant aspect of Databricks’ approach is its emphasis on governance. Every agent deployed through the platform inherits the same data governance policies that apply to the underlying data sources. If a user shouldn’t have access to certain financial records, an agent querying those records on their behalf is automatically restricted as well.

This unified governance model addresses one of the most pressing concerns for enterprises adopting AI: the risk of agents becoming unauthorized backdoors to sensitive data. By treating agents as extensions of existing data access controls rather than separate systems requiring their own security models, Databricks reduces the compliance burden that has slowed AI adoption in regulated industries.

“Agents connect directly to enterprise data and systems with consistent governance across data, models, and agent operations, reducing custom integration work and helping teams ship trusted, domain-aware agents faster.” — Databricks Product Documentation

The Competitive Landscape

Databricks is not alone in pursuing the enterprise agent market. Microsoft has integrated Copilot agents deeply into its Office and Dynamics ecosystems. Salesforce has positioned its Agentforce platform as the future of customer relationship management. Amazon and Google are building similar capabilities into their cloud platforms.

What differentiates Databricks is its focus on data infrastructure as the foundation for AI. The company has long argued that the quality of AI outputs depends fundamentally on the quality and accessibility of underlying data. By building agents on top of its existing data platform—where enterprises already have their most valuable datasets organized and governed—Databricks offers a path to AI that doesn’t require migrating data to new silos.

The timing is strategic. As enterprises move from AI experimentation to production deployment, they are discovering that the hard part isn’t building a prototype—it’s operating that prototype at scale with proper governance, monitoring, and reliability. Databricks is betting that enterprises will prefer to solve these problems within their existing data infrastructure rather than adopting entirely new platforms.

What Comes Next

Industry analysts are watching closely to see how enterprises adopt Custom Agents in production. Several key questions remain: Will the promised governance capabilities satisfy regulators in highly controlled industries like healthcare and finance? Can Databricks’ serverless infrastructure handle the scale of agent deployments at the world’s largest companies? Will the flexibility to use any model prove more valuable than the convenience of integrated solutions from competitors?

The coming months will reveal whether Databricks’ infrastructure-centric approach resonates with enterprises struggling to operationalize AI. If successful, it could establish a template for how AI agents are built and deployed across the industry—treating them not as exotic new technology requiring special handling, but as standard enterprise applications with standard enterprise requirements.

For now, one thing is clear: the race to become the default platform for enterprise AI agents is intensifying, and Databricks has just made a significant move.

This article was reported by the ArtificialDaily editorial team. For more information, visit Databricks.