In eight days in July, ZoomInfo shipped a command-line interface for GTM data (July 9), published a benchmark scoring AI agents on go-to-market work (July 10), and took GTM.AI to general availability across Claude, ChatGPT, Microsoft Copilot, Agentforce, and HubSpot Breeze (July 17) — all three announced via investor-relations releases, which tells you who the campaign is really for. The most interesting company in sales tech right now isn’t building an agent. It’s fighting to be the ground truth underneath everyone else’s.

What ZoomInfo actually shipped

Strip the launch language and the July cadence is one strategy in three parts. The CLI puts verified company and contact data where coding agents live — the terminal — a direct bid for the GTM engineers our Build-vs-Buy data shows now run the stack. The GA release makes GTM.AI a grounding source inside every major agent surface, with a claimed 100M companies and 500M contacts exposed via API and MCP (vendor figures, unaudited). And GTM Bench is the tell: a benchmark, run by ZoomInfo, in which ZoomInfo scores 77 while Apollo scores 47, Exa 36, and the open web 31. We’d caution against citing those numbers for anything — the vendor wrote the test, ran the test, and graded the test. But the artifact itself is analytically significant: benchmarks are how you market to machines and the people who deploy them. When a data vendor’s flagship marketing asset is an eval suite rather than a Super Bowl ad, it has concluded its next buyer evaluates context providers the way engineers evaluate models.

Clay is running the same play from the other side

The same week, Clay — the orchestration layer 65% of fast-growing B2B companies already use, per The Signal’s dataset — shipped its own API and CLI for coding agents (July 9, the very same day as ZoomInfo’s CLI), then MCP access for reps (July 14), open-weight models for cost control (July 16), and always-on account research agents (July 22), under a new self-description: “the infrastructure GTM engineers build on.” All vendor-announced. ZoomInfo is a data vault adding an agent interface; Clay is a workflow engine adding a data plane. They started from opposite ends and are converging on the same position — be the context layer: the thing every agent must query before it acts. Clari + Salesloft shipping an MCP server, Gong wiring MCP interoperability into the Microsoft ecosystem, and Zoom folding Common Room’s buyer intelligence into Revenue Accelerator are the same instinct at smaller amplitude.

Why the war moved here

Agents dissolve the moats software vendors spent two decades building. When a rep works through Claude or Copilot rather than nine browser tabs, the UI stops being the product and the workflow stops being sticky — what remains defensible is whatever the agent must be grounded in: proprietary, verified, continuously refreshed context. That’s why the fight is over MCP endpoints rather than dashboards, and it’s why the MCP 2026-07-28 spec (stateless protocol, enterprise-managed authorization now stable, backed by Anthropic, Microsoft, and Okta) raises the stakes in both directions: standardized context makes providers easy to plug in — and just as easy to swap out. The context layer is a real prize, but the protocol that creates the position also commoditizes it. Our agent-native stack thesis predicted platforms would fight protocol leverage; the July surprise is that the data vendors embraced it as their land grab instead.

What buyers should do with this

Three rules while the war runs. Evaluate data vendors on agent-accessibility, not seat UX: MCP server quality, API latency and rate limits, and whether you can run your own evals against your own accounts — a vendor that won’t let you benchmark it privately is telling you something. Treat vendor benchmarks as positioning documents: demand the methodology, or better, replicate a slice on your ICP before signing. Don’t sign exclusive grounding: the entire strategic value of the protocol layer is that context providers are swappable — a contract that forecloses swapping hands back the moat you were just handed. The context war is good news for buyers, but only for the ones who negotiate like it exists.