Google's latest agent push is interesting for two reasons at once. Antigravity tries to become the common harness for building and running agents, while WebMCP tries to make the web itself easier for those agents to use. One is tooling. The other is a standards bid.
What Launched
At Google I/O on May 19, 2026, Google said it launched Antigravity 2.0 and Antigravity CLI as parts of a broader agent-first development platform. Google says Antigravity 2.0 can orchestrate multiple agents in parallel, supports dynamic subagents and scheduled tasks, and shares a common harness across desktop, CLI, SDK, and enterprise surfaces.
At the same time, Chrome introduced WebMCP, a proposed web standard that lets sites expose structured tools, schemas, and state to AI agents so tasks can be completed more reliably than blind click automation.
Why WebMCP Matters
Browser automation has always been one of the fragile parts of the agent stack. Agents can often reason about a page better than they can act on it reliably. WebMCP addresses that by letting a website declare explicit purpose for an interaction, such as checkout, search, or form submission, rather than leaving the agent to infer it from raw interface elements.
That makes this more than a developer convenience. It is an attempt to make the public web legible to agents in the same way APIs made software legible to developers.
Why Antigravity Matters
Antigravity is Google acknowledging that prompts are not the right abstraction for serious agent work. The interesting features are not chat features. They are subagents, shared harnesses, background tasks, and deployment surfaces that make agent execution persistent and operational.
This complements the direction we tracked in workspace agents and managed agent infrastructure. The category is converging on the idea that teams need reusable agent operating layers, not just smarter model endpoints.
The Zero-Human Angle
A zero-human company needs both layers. It needs a control plane that can launch and supervise many agents. It also needs external software surfaces that are understandable enough for those agents to act without constant human correction.
Antigravity addresses the first problem. WebMCP starts addressing the second. That is why this launch reads as more than product positioning. It looks like stack formation.
The Take
If A2A is about agent-to-agent coordination, WebMCP is about website-to-agent coordination. And if Antigravity succeeds, Google will have a stronger claim on the execution harness that sits above both.
This is one of the clearer recent signs that big platforms are no longer shipping assistants. They are trying to define agent infrastructure.
Related: See our earlier research on A2A, Responses API, and workspace agents.