Executive Summary

Monte Carlo is a featured launch partner for Databricks Agent Bricks, delivering end-to-end data + AI observability for enterprise AI agents built on the Lakehouse. The integration connects to Unity Catalog to observe every table and pipeline agents depend on, ingests agent traces via OpenTelemetry into Delta Live Tables, and gives teams one platform to detect, triage, and resolve reliability issues across both data and agents. Monte Carlo customers resolve data incidents 80% faster, and Databricks named Monte Carlo its 2025 Data Governance Partner of the Year.

---

Best of Breed Agent Observability

Databricks recently announced major updates to Agent Bricks, their governed enterprise agent platform. Agent Bricks gives teams a fast path from idea to production by providing one place to build, evaluate, and deploy AI agents on their trusted Lakehouse data.

Importantly, Databricks built Agent Bricks to be open and multi-vendor. In their post they point out that, “Agents need to understand what data means, operate under the right identity and permissions, and work across models without locking teams into a single vendor.”

Those words aren’t hollow. Agent Bricks is extensible across multiple models, frameworks, and partners. Enterprises can pick the right building blocks for each use case without getting boxed into a single vendor’s stack. Monte Carlo is proud to be one of those featured launch partners. 

As the only data + AI observability launch partner, we play a critical role in ensuring the reliability across the structured and unstructured pipelines that agents depend on as well as the reliability of the agents themselves. 

We integrate with Agent Bricks to help teams resolve data incidents 80% faster while catching agent reliability issues in production before they negatively impact the business or trust. That track record is part of why we work with organizations like NASDAQ, Fox, and American Airlines and why Databricks named Monte Carlo its 2025 Governance Partner of the Year.

Here’s how the integration works, and how Monte Carlo and Agent Bricks ensure teams ship production-ready, reliable AI agents.

How Monte Carlo integrates with Agent Bricks

The integration between Monte Carlo and Databricks is seamless, making it easy to observe your entire data and AI environment both inside and outside the Lakehouse. 

Even better, the Monte Carlo platform can be accessed via the UI or from inside AI coding agents like Claude Code and Cursor using our MCP server and mc-agent-toolkit.

Data reliability across the Lakehouse and beyond

Monte Carlo connects to Unity Catalog and continuously observes every table, view, and pipeline. This includes data validations and anomaly detection for table level metrics such as freshness, volume, schema as well as field-level values. Monte Carlo even goes beyond the data level by observing and drawing correlations to the system and code levels from Databricks query history, lineage, workflows and more. 

Most enterprises don’t live in a single warehouse. A Delta table may be hydrated from Kafka, joined to Snowflake data, transformed by dbt, and surfaced in Tableau. Monte Carlo observes the full estate — Snowflake, BigQuery, Redshift, Athena, and more on the warehouse side; dbt, Airflow, and Fivetran on the orchestration side; Tableau, Looker, and Power BI on the consumption side. That cross-ecosystem coverage is what makes reliability actually reliable, because data incidents almost never start where they end.

Agent observability across all frameworks and platforms

Monte Carlo ingests agent traces from Agent Bricks through our open-source, OpenTelemetry-based SDK. A Delta Live Tables pipeline takes those spans from object storage and loads them into trace tables inside your Lakehouse and alongside your trusted data context.

This works today for any agent you can instrument with OpenTelemetry. This allows teams to have a consolidated space to observe their agents whether it’s built on the Agent Bricks Framework, wraps a Genie space, built natively within another platform, or it’s coded in a personal notebook.

And we’re actively building even deeper, more native integrations with Databricks’ managed agent surfaces, so connecting an agent to Monte Carlo becomes as simple as connecting a data source.

The single pane of glass, end-to-end advantage

Because Monte Carlo observes both the data and the agents — inside Databricks and across the rest of your estate — there is one coherent context to answer the only question that actually matters in production: “Why did this agent get it wrong?” 

Whether the root cause is within the data, a tool call, or a model hallucination, it surfaces in one place rather than three different tools.

Best-in-class data + AI reliability workflows

Agent observability is broader than “watch the traces.” Agent Bricks and Monte Carlo accelerate all critical data + AI reliability workflows to help teams detect, triage, and resolve agent reliability incidents.

Detect

For an agent to be reliable in production, you need visibility across four dimensions: the context it’s grounded in, the performance it delivers, the behavior it exhibits step by step, and the outputs it produces. 

Monte Carlo ensures teams have coverage across all dimensions across the full surface area. For example:

• Context. Freshness, volume, data drift, and common dimensions of data quality. An agent can’t be more reliable than the data it retrieves.
• Performance. Latency, token consumption, error rate and throughput across your agent fleet. Alerts let you catch a cost spike or a slowdown.
• Behavior. Trajectory is a dimension of agent reliability that Monte Carlo uniquely monitors. Define the paths and tool sequences an agent should follow, then get alerted when it deviates — a step running out of order, a tool skipped, an unexpected branch triggered. Live on Databricks today, alongside Snowflake, BigQuery, and Athena.
• Outputs. Run LLM-as-judge and rule-based checks on agent responses, catching hallucinations, grounding failures, off-policy responses, and format violations. Agent Bricks gives you powerful synthetic evaluation pre-launch; Monte Carlo gives you continuous evaluation against real production traffic.

Not sure where to get started? Monte Carlo will help you build a custom evaluation monitor from a short description in natural language.

Triage

Beyond a single pane of glass, maintaining reliable data + AI systems in production requires teams to prioritize incident management and response. Too many regressions sit unnoticed within dashboards, or worse, noticed and unacted upon.

A regression or performance issue without an alert, owner, and SLA is a problem.

Monte Carlo brings first-class incident management to Agent Bricks. Every alert is routed to the right owner with a clear escalation path and tracked through to resolution. This makes it easy for teams to work across Slack, PagerDuty, Microsoft Teams, ServiceNow, JIRA or wherever else they like to collaborate on incidents. 

Incidents are tracked to resolution with status, ownership, and comment history, not lost in a channel backlog. And when an upstream table fires an alert at the same time an agent starts producing degraded outputs, Monte Carlo surfaces the connection automatically rather than making an on-call engineer stitch it together at 2am.

Resolve

Once an alert is routed, the question becomes why. Monte Carlo’s Troubleshooting Agent brings automated root cause analysis to both sides of the integration.

For data incidents, it correlates the failing asset with recent query changes, pipeline job failures, upstream schema changes, and pull requests that touched relevant code — returning a ranked root cause in seconds instead of the hour-plus it typically takes a human analyst. 

For agent trace incidents, it not only reviews the data quality, it also inspects the trajectory, the prompts and responses at each step, and any tool call inputs and outputs. Monte Carlo is the only provider that automatically identifies the root cause of reliability issues across your data + AI system. 

Measure

In the age of agents, the unit of measurement changes. You’re not grading a table or a monitor; you’re measuring a system: a data product, an agent, and the pipelines that connect them. That’s the unit of trust that actually matters to a business stakeholder.

Monte Carlo has data quality scores, but more importantly we help you measure reliability at the data + AI product level. 

Looking ahead

Building reliable AI agents is a shared mission. Databricks gives enterprises a best-in-class platform to build, govern, and deploy agents quickly with Agent Bricks. Monte Carlo complements that with the observability and incident management layer that keeps those agents reliable once they’re in production.

We’re proud to have been named Databricks’ 2025 Governance Partner of the Year, and we’re not standing still. We’re actively building deeper, more native integrations with Databricks’ managed agent surfaces. Connecting the agents you ship on Agent Bricks to Monte Carlo will get simpler, faster, and more powerful over time.

If you’re building on Agent Bricks and want agents your business can trust, we’d love to show you how Monte Carlo fits. Get in touch for a demo, or find us through Databricks Partner Connect.