CData's gateway for AI agents targets the real bottleneck: trust, not the model

By Zendoric · October 3, 2026.

CData, a connectivity and data integration technology provider based in Chapel Hill (North Carolina), has unveiled Connect AI Gateway, an expansion of its CData Connect AI MCP server launched in September 2025. MCP (Model Context Protocol, the open standard promoted by Anthropic in late 2024 that allows AI agents to connect in a structured way to external tools and data sources) has become the common language with which agents talk to the rest of enterprise software. CData is now betting on being the one who controls that channel.

As Marie Forshaw, CData's senior vice president of product marketing, explained to CRN, the Gateway acts as a single control point between AI agents and corporate systems: it connects to data sources through governed tools, applies the company's context at every step, enforces permissions and routes each request to the most efficient model for the task. "We can govern everything, at every point of how that chain executes from the AI to the source systems, and we also have a record of it," Forshaw summed up. The product is available for now through an early access program.

A central piece of the announcement is the "context engine": an engine that gathers and stores semantic definitions —business terminology, metrics, existing data models— so that agents understand the real schemas and relationships of the systems they touch, not a generic approximation. It also incorporates a "self-learning loop" that corrects data and adjusts prompt patterns within the agent's memory on an ongoing basis. It is a powerful capability, but it also raises a new governance question: who audits what that self-learning loop corrects, and by what criteria, something the announcement does not spell out.

The stated reason for the launch is more revealing than the technology itself. Forshaw frames it as a question that, she says, keeps coming up in the market: "Is this helping? Where is the return on everything I have invested in AI?". Many organizations, she admits, are falling short on the answer, in part because fear leads them to limit agents' access to sensitive systems and data, which in turn cuts into the productivity they expected to gain. In general, this tension is not anecdotal: widely cited studies in 2025, such as MIT's on enterprise adoption of generative AI, put the share of corporate pilots that failed to achieve a measurable return at an overwhelming proportion —around 95%—, with integration and governance as the main obstacle, ahead of the quality of the model itself.

CData is not a marginal player on this board: its connectivity technology ships built into third-party products such as Google Cloud, BigQuery and Palantir's core systems, giving it far wider distribution than a niche vendor. And it is not alone in the race to build the governance layer for agents: Salesforce, ServiceNow, Microsoft and several identity management specialists have spent months announcing equivalent pieces —gateways, permission registries, "control planes"— for the same problem. It confirms a thesis we have already been tracking at Zendoric: in the agentic era, the competitive battle is fought as much in the plumbing —who controls the connection between the agent and the actual system— as in who trains the smartest model.

Our reading is that this kind of infrastructure, dull as it appears, is what really determines the pace of AI adoption in business in the short term. What holds organizations back today is not the model's capability but the lack of confidence to hand it the reins over payroll data, medical records or financial systems without a robust audit trail and permission system. Every gateway of this kind that genuinely works reduces that fear a little further and brings closer the moment when agents stop being a marginal productivity experiment and become an operational part of the business, with the promise of freeing up human time for the higher-judgment work we defend as our underlying horizon. But the flip side should not be lost from view: the more companies delegate control of their agents to a handful of gateways like this one, the more de facto power those intermediaries accumulate over what AI can and cannot do inside each organization, a point of concentration that deserves the same vigilance as the model itself.