The marketing and go-to-market software industry spent two years racing to ship AI agents. The next phase is quieter and more consequential: a race to own the verified data those agents stand on. On June 1, ZoomInfo and Hyland, two companies with very different roots, made the same bet on the same day. The agent is not the moat. The governed data and context underneath it is.
Two launches, one thesis
ZoomInfo made GTM.AI generally available, describing it as a headless go-to-market context layer that grounds AI agents in verified commercial data. Rather than ship another assistant, ZoomInfo exposes company search, contact discovery, real-time enrichment, and intent signals to agents running inside Claude, ChatGPT, Microsoft Copilot, Salesforce Agentforce, and HubSpot Breeze, with each call governed by the customer’s existing data entitlements. Behind it sits what the company calls the GTM Context Graph, which it says resolves 100 million companies, 500 million contacts, and billions of buying signals into one connected graph.
Hours earlier, Hyland launched its own next wave of agentic infrastructure: an Enterprise Context Engine, an Enterprise Agent Mesh to oversee how agents operate, and an Agent Lifecycle Management framework whose Agent Passport certifies each agent’s identity, capabilities, and compliance, with a Control Tower for real-time oversight. Different domain, same move. Make the data and the governance the product, and let the agents plug in.
Advertisement
300 × 250
Why the layer moved
The shift has a mechanism, and it is the Model Context Protocol. MCP gives any agent a standard way to call an external system for data and actions, which means the agent and the data no longer have to ship together. Once the interface is standardized, the durable value stops being the chat window, which every vendor now has, and moves to whatever the agent calls when it needs a fact it can trust. That is a structural change, not a feature. It is the same logic that turned payments and mapping into infrastructure that other software rents.
It is also the next layer of a build-out this publication has tracked before. The first agentic wave wired tools together so an agent in one platform could act in another, the connectivity problem. The fight now is one level down, at the data the connected agents reason over. An agent that can reach everything but trusts nothing is just a faster way to be wrong. We covered the assembly of the agentic marketing stack as vendors raced to ship agents; the June 1 launches are the same vendors conceding that the agents are only as good as the substrate beneath them.
The data vendors are becoming infrastructure
What makes the timing notable is who is moving. ZoomInfo is a data business repositioning its commercial graph as the layer agents query rather than a database humans log into. Hyland is a content-services company turning its repository into governed context for agents. The common pattern: companies that used to sell an application are now selling the verified data and the governance an agent consumes through a protocol. The CDP and the data warehouse have been heading this way too, as the standalone customer data layer dissolves into the platforms around it.
What this means for the marketing leader
The practical question is changing. For two years the buying question was which agent to adopt. The more durable question is whose data your agents will stand on, and whether you can trust it. Three implications follow. First, evaluate agents by their inputs, not their demos: an agent is only as accurate, current, and compliant as the data layer it calls. Second, treat data entitlements and governance as agent infrastructure, because an autonomous agent acting on ungoverned data is a liability that scales. Third, watch for lock-in moving down the stack. If your agents are grounded in one vendor’s graph through a proprietary context layer, that graph, not the agent, is now your switching cost.
The catch
There is a reason to stay skeptical. Verified data raises the floor on agent reliability, but it does not fix reasoning. An agent handed a perfect customer record can still draw the wrong conclusion, and “governed” is easy to claim and hard to audit. Certification schemes like an agent passport are only as strong as the body that enforces them, and right now each vendor enforces its own. The context-layer pitch is coherent and probably correct about where value accrues. Buyers should still ask the unglamorous questions: where did this data come from, how fresh is it, who can the agent act on behalf of, and what happens when it is wrong.
The agent era is not slowing down. It is just revealing what it actually runs on. The companies that spent a decade compiling and governing data are betting the next decade is theirs, because in an industry full of agents, the scarce thing is a fact you can trust. Source: ZoomInfo and Hyland.
