Session

Beyond Data: AI-First Collaboration with Apps, Genie Sharing, and MCP in Marketplace

Overview

Experience	In Person
Track	Data Sharing & Collaboration
Industry	Enterprise Technology, Retail & Consumer Goods
Technologies	Data Marketplace, Databricks Apps
Skill Level	Beginner
DOWNLOAD SESSION SLIDES

Data sharing got us this far. The next leap is sharing the intelligence on top of that data: the agents, applications, and AI experiences your partners and customers actually want. With Databricks Marketplace, the ecosystem is expanding from raw datasets to a full set of AI-first collaboration primitives: Genie Spaces, third-party applications, and MCP servers, all governed, and all running in the consumer's own environment. In this session, we'll demonstrate three new ways to collaborate across organizational boundaries:

Genie Sharing: a first-of-its-kind privacy-safe agentic collaboration tool. Hand a partner a curated natural-language interface to your data instead of raw tables. Provider IP (instructions, benchmarks, knowledge store) stays hidden, while output is governed by aggregation-only columns, row limits, and query quotas. Recipients can even blend in their own data without exposing it back to the provider. 3P Apps in Marketplace: plug-and-play applications running entirely in the consumer's workspace, from financial forecasting to manufacturing analytics. No data leaves your boundary. MCP servers in Marketplace: turn Marketplace assets (tables, notebooks, volumes, Apps) into agentic tools that Genie, Agent Bricks, or any external LLM can call as governed UC objects.

Key Highlights

AI-First Ecosystem: A curated library of Genie Spaces, Apps, and MCP tools, natively integrated with your governed data.
Privacy-Safe by Default: Genie Sharing gives external recipients an agentic experience without exposing raw rows or proprietary logic; Apps run in-place to eliminate exfiltration risk.
Speed to Value: Move from discovery to production in days, for data, agents, and applications alike.

Session Speakers

IMAGE COMING SOON

Akram Chetibi

/Director, Product Management
Databricks

Tia Chang

/Product Manager
Databricks

Full Summary

Sharing beyond data: how Databricks is extending collaboration into the age of AI agents

Enterprises have long shared tables and files across clouds with Delta Sharing. The focus is now expanding to the richer intelligence built on top of data.

The session outlined how organizations can securely collaborate on agents, tools, and full applications, using Databricks Marketplace as the distribution and governance hub.

As agentic experiences mature, the value to share is shifting from raw datasets to the semantics, instructions, and applications that make data actionable. Databricks is enabling that shift through open protocols and three complementary sharing models that preserve IP, streamline deployment, and keep customers in control of their data. Organizations exploring this evolution can find a broader view of how these capabilities fit together across the Databricks Platform.

From open data exchange to open sharing of AI assets

Delta Sharing pioneered open, cross-platform data exchange starting in 2021. It is now becoming OpenSharing, an open source protocol donated to the Linux Foundation. The scope is expanding beyond tables to include AI assets such as models, notebooks, volumes, and agents, with growing support for on-premises recipients.

Enterprises can keep using the same governed, interoperable protocol they rely on for data to share a wider set of AI artifacts without locking into a single vendor or cloud. The ecosystem spans hundreds of partners and thousands of customers across industries.

Three complementary sharing models for the AI era

MCP servers via Databricks Marketplace. The Databricks Marketplace already lists more than 60 remote MCP servers that plug directly into Genie and custom agents. Providers expose tools or APIs while keeping their logic behind their own services. Access is governed with Unity Catalog and Unity AI Gateway, so external capabilities feel native inside the Databricks Platform.
Shared Genie Agents with data controls. When MCP is not the right fit, providers can package a dataset, its semantics, instructions, and sample queries into a Genie Agent and share it like any other asset. Recipients get a preconfigured natural language interface and can securely join shared data with their own data inside one agent. Providers keep control with daily query quotas and output size caps, and recipients retain privacy over their questions and local data.
Installable apps on Databricks Marketplace. Partners can distribute IP-protected apps that install into the customer's own workspace on serverless compute. Everything is governed through Unity Catalog and Unity AI Gateway. Any external network access is declared up front and must be explicitly approved during installation. At launch, 20 partners span identity and privacy, ad tech and activation, data quality and reasoning, and developer and operational tools, including Meta, The Trade Desk, Capital One, Datavant, Lake Fusion, and Slalom.

Composing first-party data with third-party intelligence in one workspace

A live scenario showed how these pieces work together. An app from Lake Fusion resolved messy point-of-sale and CRM data into a clean golden record in Unity Catalog. A shared Genie Agent from a consumer insights provider layered on behavioral and price-sensitivity signals. An MCP server from u.com pulled live competitor pricing from the web. Genie orchestrated across these assets to produce concrete retention and pricing recommendations, all within one governed environment.

The lesson was composition, not any one step. Trustworthy first-party data, grounded third-party intelligence, and real-time external signals can be assembled by the customer to fit their needs rather than relying on a single prebuilt solution.

A two-sided marketplace that shortens time to value

Databricks Marketplace is evolving into a venue where providers build once and distribute broadly, while customers compose tailored intelligence on top of governed data they already control. Providers avoid rebuilding integrations per customer, keep source code closed, and pass Databricks' security scans before publishing to more than 20,000 customers. Customers get faster deployment, consistent governance, clear data boundaries, and fine-grained controls on usage and cost.

Why sharing intelligence, not just data, changes the collaboration map

OpenSharing and Databricks Marketplace extend collaboration from data to the full stack of AI assets. Organizations can now exchange tools, agents, and applications with clear governance, privacy, and IP protection, then combine them with their own data to accelerate outcomes that used to take months.

FAQ

A shared Genie Agent brings the provider's data, semantics, and instructions into the recipient's workspace, which makes cross-dataset joins with first-party data straightforward. A remote MCP server keeps logic and data with the provider and routes each question externally, which is ideal for tool-style integrations but harder for local joins.

Providers can set daily query quotas per recipient and cap output size by rows and megabytes. These controls protect monetizable assets and manage usage while recipients keep their own prompts and data private.

Installed apps run inside the customer's Databricks workspace on serverless compute, governed by Unity Catalog and Unity AI Gateway. Any external egress must be declared during installation and explicitly approved, and provider source code remains closed.

Not necessarily. Providers with MCP servers can list them, and enterprises can package data, agents, or apps to reach customers in new ways. The session encouraged interested providers to engage and list assets that can be governed and distributed through Databricks Marketplace.

All three models integrate with Unity Catalog and Unity AI Gateway for consistent permissions, authentication, and access controls, whether the asset is internal or from an external provider.