Adobe ATSE's Path to a Unified, Governed Data Layer for Enterprise AI at Scale

Adobe is a global software company whose products, from creative tools to digital marketing platforms are used by millions of people and businesses worldwide. With nearly 30,000 employees and operations spanning dozens of countries, keeping its internal systems connected and its data accessible is a constant operational challenge — one that falls to the company’s integration platform team.

In early 2026, Padmakar Pasala, Integration Platform Lead at Adobe, and his team took on a question that every enterprise scaling AI agents eventually hits: how do you give agents reliable, governed access to enterprise data without rebuilding the connection every time? With the pilot, an Adobe ATSE integration platform team set out to build a single platform connecting enterprise systems, enforcing governance, and giving AI agents secure, guardrailed access to live data. The plan was to prove it with a defined group of teams before scaling further.

Live, governed access—built on a shared data layer

Adobe ATSE was already familiar with CData through years of using CData Drivers. When they ran a feasibility study for the data layer initiative, CData Connect AI met all the criteria:

• Coverage across hundreds of data sources

• Live access to the data

• Access to on-premises and cloud data

• MCP-compatible tool endpoints with read and write capabilities

• Role-based access controls to govern which teams and tools can reach which systems

This enabled the team to build the pilot’s data layer on CData Connect AI, exposing their enterprise systems as MCP-compatible tool endpoints that AWS Bedrock agents could call directly. Connect AI normalizes each connected system into a structured set of tools, abstracting the underlying APIs so agents can interact with any source through a consistent interface. An internal agent called SARA which handles tasks like updating permissions and executing business workflows sits on top of the layer and calls those tools directly, with no RPA bot in the middle.

To reach SAP S/4HANA, they used Connect AI's Private Proxy Network, which bridges between AWS Bedrock and Adobe's on-premises SAP environment without exposing internal infrastructure to the public internet.

Connect AI’s toolkits feature lets the team group tools by use case. Each use case gets its own scoped toolkit, so teams discover and call only the tools their workflows require.

Role-based access controls let the team govern which tools and data sources each use case can reach so as more teams join the data layer, access stays controlled without slowing down adoption. Scoping tools per use case also makes agents more accurate and more token-efficient: they reason over a focused set of tools rather than the full catalog. SSO was a hard requirement before any team could be onboarded. The team configured Adobe’s identity provider with Connect AI before opening access to users so every session is authenticated through Adobe’s own systems, not a separate set of credentials.

Workspaces add another layer to the security story, letting the team structure access by group so that governance scales as more teams join the platform. Where source systems carry their own permission model, Connect AI can surface those boundaries to the agent layer, so access policies don’t need to be re-implemented separately.

Once the foundation was in place, expansion followed quickly. The team added GitHub (cloud and an on-premises GitCorp instance), Slack, SharePoint, Splunk, Confluence, Jira, and Wiki, each new source extending the same data layer rather than requiring a new integration project. Because all sources are exposed through the same MCP tool interface, SARA can reach any connected system using the same pattern, regardless of what's added next.

Pilot complete: foundation in place, expansion under way

The pilot set out to onboard 12 internal teams actively using the data layer and validate that agents could securely access enterprise data with the right guardrails in place. It completed faster than the team anticipated. Agents across the pilot are now reaching live data across SAP, Dynamics 365, GitHub, Slack, and more. All within boundaries enforced at the MCP layer through schema lock-in, restricting what each agent can access and ensuring it operates within defined limits. Because the foundation is shared, each team that joins inherits access to every connected system without additional integration work.

Even at pilot scale, the reliability of the data layer matters. With 12 internal teams participating in the pilot for their workflows, a gap in data access doesn’t just slow a project down. It blocks those teams from operating.

With the pilot behind them, the team is now building the next layers. An observability layer is in development, with query logs federated through CData to give the team visibility into how agents are interacting with enterprise data. Expanding governed access across more enterprise systems is the next milestone, on the path to a full MCP tools catalog covering every system Adobe relies on for business and finance operations.

CData Connect AI: the data layer AI agents run on

With CData Connect AI, hundreds of enterprise data sources become agent-ready tool endpoints—with read and write support, Cloud Gateway for on-premises connectivity, role-based access controls, workspaces, and toolkits for structured discovery. Everything a production data layer needs, without building it from scratch.