AI adoption is accelerating, but turning AI pilots into real production systems remains difficult. Many organizations experiment with generative AI, copilots, and agents, yet very few projects move beyond the testing phase. The challenge is the data infrastructure behind them, not models.
CData is trying to address that gap with new updates to its Connect AI platform. The company introduced expanded connectivity, new agent tooling, and additional governance controls designed to help organizations run AI systems against live business data while maintaining security and operational control.
Why Many AI Projects Stall
Spending on AI continues to grow rapidly. Yet most projects still struggle to move from proof-of-concept to production. One reason is data access. AI systems need continuous access to business systems such as ERP platforms, CRM systems, data warehouses, and operational databases. Without that access, AI tools are limited to static datasets or isolated tasks.
Many organizations still rely on custom-built integrations to connect AI tools to their internal systems. Those integrations often take significant time to build and maintain. In practice, that means AI teams spend a large portion of their time working on data pipelines instead of building AI applications.
Industry research reflects the same problem. Only a small portion of organizations report being satisfied with their current data infrastructure for AI projects, and many teams say data integration consumes a substantial share of their implementation effort.
CData’s Connect AI platform focuses on solving this infrastructure challenge. The platform is built around three core elements: connectivity, context, and control. The idea is to give AI systems direct access to business data while maintaining the security and governance requirements that production environments demand.
The latest updates extend all three areas.
Expanding Connectivity to Business Systems
Connect AI provides read and write access to more than 350 business systems without requiring organizations to replicate or move data. The new On-Premise Agent extends this connectivity to systems running inside corporate networks. This allows AI tools to interact with platforms that typically remain behind the firewall, including systems such as SAP environments, SQL Server databases, and PostgreSQL deployments.
For organizations building AI agents that operate across multiple systems, this type of connectivity can remove a common barrier: limited access to operational data. Instead of relying on copied datasets or scheduled exports, AI applications can interact directly with the systems where business data already lives.
According to
William Davis, CMO at CData, broad connectivity is one of the core requirements organizations face as they move from experimentation toward operational AI systems. He told ChannelE2E, “When we look at what separates Connect AI from other MCP providers, the answer maps directly to the three things enterprises actually need to move AI from experimentation to production: connectivity, context, and control. Most providers address one of these reasonably well. Very few address all three, and almost none do it at the architectural depth required for autonomous agents operating on live business data.”
He noted that many native MCP implementations are tied to a single platform, while enterprise environments typically span dozens or hundreds of systems.
“Connect AI provides live, read-write access to more than 350 enterprise systems, including CRM, ERP, data warehouses, project management tools, and cloud databases,” Davis said. “Most native MCP servers are single-platform by design. A CRM vendor’s MCP connects to their CRM. Connect AI works identically across hundreds of sources, which is critical for enterprises running heterogeneous environments.”
Giving AI Agents Better Context
Access to data alone is not enough. AI agents also need context about how business systems work and how data is structured. To address that, CData expanded the agent tooling inside Connect AI. The platform now includes three types of tools designed to support different levels of flexibility and control:
- Universal Tools allow agents to explore data across connected systems and handle complex or exploratory queries.
- Source Tools provide system-specific operations that give administrators tighter control over what agents can access or modify.
- Custom Tools allow organizations to define purpose-built actions tied to specific business workflows.
These tools can be grouped into toolkits, which define how an AI agent interacts with systems and data. Administrators can then assign toolkits to specific workspaces, which limit the data and operations available to each agent. This structure allows organizations to design agents that operate within defined boundaries instead of having unrestricted access to enterprise data.
Davis said managing the context exposed to AI agents is critical for both performance and governance.
“Exposing too much context creates its own risks, including increased token usage, model confusion, and unintended access to sensitive data,” he said. “Our scoped MCP architecture, through Workspaces and Toolkits, gives IT teams precise control over what each agent can see and do. Workspaces define the data boundary. Toolkits define the action boundary.”
Adding Governance and Security Controls
Running AI systems against live data introduces security and compliance concerns. The latest platform updates focus on strengthening governance controls around identity and access. Connect AI now supports SCIM 2.0 for automated identity lifecycle management, allowing organizations to synchronize user identities and permissions from existing identity systems. The platform also introduces Custom OAuth Applications, enabling companies to use their own credentials and authentication policies when connecting to external services.
Each query executed through the platform is authenticated and logged, providing audit trails for organizations that need visibility into how AI systems interact with business data.
Davis said these governance controls are often the deciding factor in whether organizations allow AI systems to interact with production environments. “Connect AI enforces per-user authentication with native source-system permissions applied dynamically at runtime, backed by full audit trails,” he said. “For enterprises moving toward autonomous agents that read, write, and act on live business data without human review, that level of control isn’t optional. It’s the condition under which AI gets approved to run in production.”
Accuracy Becomes a Critical Factor for AI Agents
As organizations move from AI assistants to more autonomous agents, reliability becomes a major concern. AI agents that read or write data across systems must interpret queries accurately and execute the correct actions. Small error rates can quickly compound in automated workflows.
CData evaluated several Model Context Protocol (MCP) providers across hundreds of real-world queries involving CRM platforms, data warehouses, project management tools, and ERP systems. The tests measured how accurately each platform returned data and handled complex queries such as relative date calculations, multi-condition filters, and write operations. Connect AI achieved 98.5% accuracy across the test set, while other providers ranged between 65% and 75%.
Davis said the difference comes down largely to architectural design.
“The accuracy gap comes down to a fundamental architectural difference,” he said. “Most MCP providers translate natural language directly into REST API calls. That works for simple lookups, but it breaks down quickly when queries involve date math, multi-condition filtering, workflow validation, or business-specific logic.”
Connect AI uses a layered architecture designed to better understand business data structures.
“Source-level semantic intelligence gives the platform a deep understanding of each system, including entity relationships, fiscal calendar definitions, workflow rules, and platform-specific conventions,” Davis said. “That semantic layer allows the system to understand what the data actually means, not just where it sits.”
He added that many failures in competing systems occur in everyday operational queries.
“In our benchmark, the failure patterns clustered around four categories: date logic errors, multi-filter combination failures, write operation validation failures, and schema mapping errors,” Davis said. “These aren’t edge cases. They’re the kinds of tasks autonomous agents need to perform reliably in production environments.”
Opportunities for MSPs and Channel Partners
The expansion of Connect AI also creates opportunities for managed service providers and channel partners that want to build AI-driven services for customers. Many organizations exploring AI deployments lack the infrastructure needed to connect models to operational systems. Partners often step in to build integrations, deploy data pipelines, and operationalize AI tools.
Davis said this gap creates an opening for service providers.
“Only a small percentage of organizations consider their data infrastructure fully ready for AI,” he said. “At the same time, many AI teams spend a significant portion of their implementation time on data integration. MSPs and channel partners are well-positioned to solve that problem for their customers.”
Partners can build vertical or use-case-specific AI solutions on top of the platform without having to build connectivity and governance frameworks from scratch.
“Connect AI handles live access to hundreds of data sources, semantic intelligence, and governance controls,” Davis said. “Partners can focus on delivering AI solutions tied to real customer workflows rather than solving the underlying data plumbing each time.”
Custom tools also create opportunities for partners to build repeatable service offerings.
“Our Custom Tools capability allows partners to define operations tailored to specific customer workflows, with optimized queries and explicit data access limits,” Davis said. “That enables a packaged service model rather than a one-off integration project.”
Some partners are also embedding Connect AI capabilities directly into their own platforms.
“We offer Connect AI Embed for partners who want to integrate these capabilities into their own products,” Davis said. “The data layer for AI is foundational, and partners who own that layer for their customers hold a durable position in those accounts.”
Why the Data Layer Matters for Production AI
Many organizations have already experimented with AI copilots or internal assistants. The next phase involves systems that can take actions across applications and automate complex workflows. That shift places new demands on the infrastructure supporting those systems. AI applications need reliable connectivity to operational systems, clear context about how data is structured, and governance controls that satisfy security and compliance requirements. Without those elements in place, AI initiatives often remain stuck in experimentation. Platforms such as Connect AI are emerging as part of the infrastructure designed to support the next stage of AI adoption, where agents interact directly with enterprise systems and real business processes.
