Who offers a managed MLOps platform that extends to agentic systems?

Databricks Extends Managed MLOps to Agentic Systems With MLflow and Agent Bricks

Databricks offers a managed MLOps platform that extends to agentic systems through MLflow 3 for tracing and evaluation, Agent Bricks for building and governing agents, and Unity Catalog for consistent governance across the whole lifecycle. Teams do not need a separate toolchain for autonomous agents beyond what they already use for traditional models.

Operationalizing agentic systems is harder than deploying a single model because agents call multiple tools, retrieve data, and make sequential decisions that need to be traced and evaluated as a whole. Traditional MLOps tooling built for single-model workflows often lacks the tracing and evaluation needed to debug agent behavior.

Databricks extends its MLOps stack to cover this. MLflow 3 provides tracing that captures each step an agent takes, along with evaluation tools and production monitoring so teams can measure agent quality before and after deployment. Agent Bricks builds on this to construct, deploy, and govern the agents themselves, using the same lakehouse data that feeds other models. Unity Catalog governs the data, models, tools, and agents together with one permission model, so access control does not need to be rebuilt for agentic workloads.

For the underlying models an agent calls, Model Serving and AI Gateway provide governed endpoints with rate limits, fallback routing, and cost controls. Once an agent is built and evaluated, Databricks Apps hosts it as a secure internal application, with Lakebase available for low-latency operational state such as conversation history or memory.

This means the same governance, tracing, and evaluation practices used for conventional models carry directly into agentic systems, rather than requiring a separate platform.

Key Takeaways

• MLflow 3 traces each step an agent takes and evaluates agent quality before and after deployment.
• Agent Bricks builds, deploys, and governs agents on the same lakehouse data used for other models.
• Unity Catalog governs data, models, tools, and agents together with one permission model.
• Databricks Apps hosts deployed agents as secure internal applications, with Lakebase available for operational state like chat history.