There is a version of this Google I/O that reads like a product launch. New CLI, new desktop app, new pricing tier. But that framing doesn’t get to the bottom of it. What Google announced is a structural bet — on where software development is headed, and on what the web itself is for.
Two announcements define that bet. And if you’re making platform decisions today, you need to understand both.
Antigravity 2.0: The IDE Is No Longer the Unit of Work
The original Antigravity, launched in November 2025, was an AI IDE. Version 2.0, announced at I/O 2026, is something else: a standalone agent orchestration platform — desktop app, CLI, SDK, Managed Agents API, and an enterprise deployment path through the Gemini Enterprise Agent Platform.
The architectural shift is the story. Google is moving its development tools from IDE-centric assistance toward multi-agent workflow management as the primary abstraction. The IDE — the text editor with autocomplete — is no longer the center of gravity. The agent harness is.
In practice: the new desktop app lets you orchestrate multiple agents simultaneously, design custom sub-agent workflows, and schedule tasks that run automatically in the background. In the live demo at I/O, Google tasked Antigravity 2.0 with building an operating system from scratch — and the system deployed 93 sub-agents working in parallel. That number is worth more as a signal than as a benchmark: this is a platform designed for tasks measured in hours, not seconds.
For developers who prefer the terminal, Antigravity CLI is available globally starting today, sharing the same agent harness as the Antigravity 2.0 desktop app — which means all future improvements to the agent core apply automatically wherever you use it.
The retirement timeline is concrete: on June 18, 2026, Gemini CLI and Gemini Code Assist extensions stop serving requests for users on Google AI Pro, AI Ultra, and the free tier. Enterprise customers with Gemini Code Assist Standard or Enterprise licenses retain access — but the direction is unambiguous. Antigravity is the future of Google’s development surface.
One relevant detail for teams working with multiple models: the original Antigravity already supported Gemini 3 Pro, Anthropic’s Claude Sonnet 4.5, and OpenAI’s GPT-OSS in the same harness. Version 2.0 maintains that model-agnostic approach.
WebMCP: What If the Web Could Describe Itself to Agents?
The second announcement is more speculative in timeline, more consequential in implications.
WebMCP is a proposed open web standard that allows websites to expose structured tools — JavaScript functions and HTML forms — to browser-based agents. By defining those tools, sites tell agents exactly how and where to interact, rather than leaving them to interpret the DOM or simulate clicks.
WebMCP is being developed jointly by Google’s Chrome team and Microsoft’s Edge team, with editors from both organizations working on the W3C spec under the W3C Web Machine Learning Community Group. The co-authorship matters: this is not a Google proprietary extension. It’s a standards-track effort with cross-vendor backing from day one.
The problem it solves is one that any developer working with browser agents has already encountered. Today, when an agent tries to complete a task on a website — book a flight, submit a form, navigate a checkout — it guesses. It reads screenshots. It clicks based on visual heuristics. WebMCP replaces that guessing with a structured contract: a clear menu of actions the site offers to AI agents, with defined parameters and expected outcomes.
The practical model: MCP (Anthropic’s Model Context Protocol) connects agents with backend tools and services. WebMCP connects agents with the frontend logic of your website — the part that already exists, already has user context, and doesn’t require rebuilding anything on the backend. They’re complementary layers, not competing standards.
The WebMCP origin trial launches in Chrome 149, with Gemini in Chrome adding support for WebMCP APIs soon. Among early adopters already testing it: Booking.com, Shopify, Expedia, Intuit, and Redfin.
An honest caveat: WebMCP still has no signals from Firefox or Safari. The interoperability story is yet to be written. A W3C Community Group Report is not a ratified standard. This is early — but the governance structure and industry participation suggest it’s early in the right direction.
What This Means for Platform Decisions
These two announcements are not independent. They describe a coherent vision: agents that can build software autonomously (Antigravity 2.0) operating on a web that was instrumented to receive them (WebMCP). One addresses the development side, the other the deployment target.
For engineering leadership in Latin America, there are three things worth monitoring closely:
The CLI migration deadline is real. If your team uses Gemini CLI, June 18 is a hard cutoff for non-enterprise licenses. Evaluate Antigravity CLI now, not in three weeks.
WebMCP is where the frontend conversation is headed. If your product has a web interface that agents will interact with — and increasingly, every product does — understanding how to expose tools via WebMCP will be a frontend competency in 12 months. The companies testing it today are defining the interaction patterns everyone else will follow.
The model-agnostic surface is the safe bet. Antigravity supports Claude, GPT, and Gemini in the same harness. WebMCP is explicitly designed to work with any agent, regardless of provider. Both decisions reduce vendor lock-in at the infrastructure level — which is the right posture for teams that can’t afford to rebuild their stack every time the model landscape shifts.
The era of the IDE didn’t end yesterday. But Google just made a very public bet on what replaces it.