By: Vedant Raghav, Manager - Program & Project Management Data & Analytics

What if your sales lead could ask one question and, within seconds, see revenue, campaign performance, and inventory risk on the same screen? Why does that still feel hard in 2025—when data is everywhere?

Most organizations sit on a mountain of information, yet cohesive, cross-domain insights remain elusive. Traditional architectural complexities, with disparate data warehouses, unstructured data lakes, and custom-built solutions create silos. The result: slower decisions, higher costs, and lack of confidence in insights.

Start small, think big

Deploying a natural-language tool such as Databricks Genie is not a matter of “plug it into everything and hope for magic.” Real success starts with a single, high value use case that serves a clear set of users.¹

1. Define your use case
   Rather than trying to answer every question across the organization, zero in on a high-impact domain like Sales Performance Analysis, Marketing Campaign ROI, or Supply Chain Logistics. Collaborate directly with domain experts like sales managers, marketers, logistics coordinators, and ask them:“What five to ten questions do you ask every week to do your job?”Those questions become the first targets for Genie. Examples include:
   • “What were our top ten products by sales revenue last quarter in the EMEA region?”
   • “Show the conversion rate for the latest email campaign, broken down by customer segment.”
   • “Which warehouses fall below safety stock for product SKU ‘XYZ’?”

   A narrow, deep score delivers tangible value by answering the most pressing questions with precision.

2. Choose the right data
   Once your use case is set, data selection becomes a deliberate process. The mantra: less is more.
   • Limit the scope: Start with a minimal set of high-quality, trusted tables. Avoid raw or redundant data.
   • Prioritize curated data: Use cleaned, well-structured “gold” tables.
   • Enrich with context:
     • Add plain-language comments to cryptic columns (e.g., cust_stat_cd: “1 means active, 2 means inactive”).
     • Define primary and foreign key relationships to guide Genie’s joins.
     • Add synonyms (e.g., “sales” for “revenue”) to align with user language.
     • Add relevant instructions, example queries, and benchmarks.

   By starting small and building iteratively, you create a Genie that’s not just smart—but strategically useful.

   While tools like Databricks’ Genie have empowered users to “talk with their data” using natural language, a common hurdle emerges. As more datasets are attached to a single “Genie space,” its accuracy decreases. This often call for creating multiple domain-specific “Genie spaces,” leading to a fragmented user experience where individuals must hop between different interfaces to get their answers.

   The solution lies in a multi-agent approach that intelligently orchestrates these specialized Genie spaces, providing a single, convenient interface for users to access comprehensive insights across various data domains.

The challenge of siloed expertise and declining accuracy

When a single “Genie space” is tasked with understanding and querying a vast array of disparate data, its performance can suffer. Just as a human expert specializes in a particular field, an AI agent optimized for a narrow domain will perform more accurately within that domain. This has led to a common practice of creating several “Genie spaces,” each expertly tuned to specific data sets like sales, logistics, or finance.

However, this compartmentalization, while boosting individual Genie accuracy, introduces a new user experience problem. Users must know which Genie space holds the answer to their query and manually switch between them. This friction hinders rapid decision-making and limits the ability to ask complex questions that span multiple organizational functions.

The rise of the multi-Genie agent: Orchestrating data intelligence

Agent Bricks—Databricks’ innovative solution simplifies the creation of high-quality, domain-specific agents². It tackles the complexities of agent creation, evaluation, and tuning, allowing organizations to focus on the business problem rather than the underlying AI plumbing.

Databricks’ Agent Bricks: Auto-optimized multi-agent supervisors

Agent Bricks offers a multi-agent supervisor capability that directly addresses cross-domain querying challenges.

• Orchestration power: The multi-agent supervisor can stitch together multiple agents, including Genie spaces, custom LLM agents, and external tools like Model Context Protocol (MCP) servers, to tackle complex reasoning tasks. This allows organizations to arrange agents for tasks such as intent detection, document retrieval, and compliance checks, creating complete, personalized responses for advisors and clients. For example, a multi-agent supervisor can combine specialized Databricks Genie spaces (e.g., Sales Genie, Logistics Genie, Customer Support Knowledge Assistant) to answer a complex question. It can further break it down into sub-questions and delegate them to relevant agents. The supervisor can then synthesize the results into a comprehensive report including insights around market demand analysis, customer satisfaction, and cost considerations.
• Automated optimization: Agent Bricks is designed to automatically optimize AI agents using a customer’s unique data, ensuring the agents are both cost-efficient and trustworthy.
  • It generates task-specific evaluation benchmarks and custom LLM judges to ensure a reliable quality assessment. This helps overcome the limitation of generic academic benchmarks, which often fail to reflect real-world business needs.
  • It intelligently searches and combines various optimization techniques, such as prompt engineering, model fine-tuning, reward models, test-adaptive optimization, to deliver high quality, removing the need for manual trial and error.
  • Users can choose the iteration that best balances their desired level of quality with cost efficiency. This often results in solutions that are both higher in quality and lower in cost compared to traditional, manual approaches.
• Continual learning: Agent Bricks includes Agent Learning from Human Feedback (ALHF), which allows the system to continuously improve by incorporating natural language guidance from users. Over time, this improves the agent’s behavior and performance. This also means domain experts can help enhance the system directly, without needing deep technical AI expertise.

• Empowered business users: Agent Bricks democratizes access to deep, cross-functional insights for a wider audience. It eliminates the need for technical intermediaries and supports scalable, maintainable solutions. The result is a robust, modular system that can easily adapt to new data sources and business domains as the organization grows.

The future of data-driven decision making

The ability to build and deploy cross-domain querying solutions using multiple specialized Genie spaces, knowledge assistants, and MCP servers represents a fundamental shift in how businesses interact with their data. By abstracting away complexity and automating optimization, platforms like Agent Bricks—and other data intelligence platforms—enable organizations to deploy production-grade AI agents rapidly and confidently. These agents come with precise control over quality and cost, along with built-in security and governance.

Databricks apps, now generally available, further streamline the deployment of secure and governed data intelligence applications. They do this by moving the application to where the data and AI live, rather than the other way around. Databricks Apps integrate directly with the Databricks Data Intelligence Platform.

These apps can host and interact with agents, and users can embed the necessary dashboards to create a one-stop solution—accelerating insight generation with AI across the enterprise.

This innovation enables access to complex, multi-faceted insights that were once difficult, if not impossible, to obtain quickly. It marks a shift toward truly “data-intelligent” applications, where AI can reason over proprietary enterprise data with both high quality and cost effectiveness.

As Databricks continues to democratize access to data and AI, the integration of specialized AI agents promises is set to transform decision-making across every industry.

References

¹Curate an effective Genie space, Databricks, May 29, 2025: https://docs.databricks.com/aws/en/genie/best-practices

² Introducing Agent Bricks: Auto-Optimized Agents Using Your Data, Xiangrui Meng, Kasey Uhlenhuth, Hanlin Tang, Patrick Wendell and Matei Zaharia, Databricks, June 11, 2025: https://www.databricks.com/blog/introducing-agent-bricks