Google used its annual developer conference on May 19, 2026, to push its software-building tools toward a model where AI agents, not humans, do the orchestration. The company introduced the Gemini 3.5 series, a second version of its Antigravity development platform, and a proposed open web standard called WebMCP, alongside new Android tooling and a Chrome DevTools surface aimed specifically at autonomous agents.

The announcements landed during the Developer keynote on the opening day of Google I/O 2026 in Mountain View, California, and were published in detail on the Google Developers Blog on the same date. According to the keynote recap authored by the Google I/O team, the framing for the entire programme has shifted from assistive tooling to agent orchestration. The team wrote that the company has "transitioned from AI that simply assists you, to agents that can independently navigate complex tasks across your entire workflow."

The unifying thread is straightforward. Google wants its development stack - the model, the editor, the CLI, the browser, and the runtime - configured so that an agent can take a task description, plan the work, write the code, run it, verify the output, and hand back a result that a human can audit. Each individual announcement on May 19 plugs into that flow.

Gemini 3.5 and a reworked Antigravity

The Gemini 3.5 series of models forms the foundation of the announcements. Google did not publish full benchmark tables in the keynote post itself, but third-party coverage from SiliconANGLE reported that Gemini 3.5 Flash, the new default in Antigravity, outperforms the previous Gemini 3.1 Pro on almost all coding benchmarks while running roughly four times faster than other frontier coding models. The previous-generation Gemini 3 was launched on November 18, 2025, topping the LMArena Leaderboard with a 1501 Elo score and scoring 76.2% on SWE-bench Verified, a benchmark for coding agent capability.

Antigravity itself is not a new brand. The platform was introduced alongside Gemini 3 in November 2025 as Google's agent-first development environment, pairing a familiar IDE surface with a separate manager view where agents work asynchronously. Version 2.0 separates the product from its previous IDE-only form into what Google now describes as a unified harness with two complementary surfaces.

According to Google, the new Antigravity 2.0 and an all-new Antigravity CLI give developers "two powerful surfaces for incredible productivity gains." Users can spawn what Google calls specialized subagents to handle parallel parts of a workflow. Google built in cross-platform terminal sandboxing, credential masking, and hardened Git policies to manage the security exposure that comes from giving agents access to a terminal, source control, and the file system at the same time.

The CLI is positioned for developers who prefer terminal workflows over a desktop interface but want the same agent harness available inside the main Antigravity application. That symmetry between graphical and command-line surfaces is one of the structural choices in the launch: the same agent runtime sits underneath both.

Managed agents, an SDK, and a Gemini API entry point

Two additional surfaces extend Antigravity beyond the desktop. Managed Agents in the Gemini API allow developers to call a single endpoint and receive a fully provisioned agent with a remote sandbox attached, removing the need to set up the underlying infrastructure. According to Google's keynote post, the feature delivers the power of the Antigravity agent harness via managed agents. A single API call provides a fully provisioned agent with a remote sandbox.

For teams that need to run agents on infrastructure they control, Google released the Antigravity SDK. The SDK gives programmatic control over the harness, which means a development organisation can deploy the agent loop on its own infrastructure rather than depending on Google's hosted runtime. That matters for regulated industries, data-residency constraints, or anyone who simply does not want the agent's execution context sitting on a vendor-managed sandbox.

The combined effect is to make the same agent stack available in three deployment shapes: a developer-facing desktop application, a terminal CLI, and a hosted API endpoint, with the SDK as a fourth path for self-hosting. The pattern echoes the way Google has positioned its other AI platform components, including Gemini in AI Studio's vibe coding interface, which Google first announced on October 26, 2025.

Google AI Studio gains Android, Workspace, and Cloud Run paths

Google AI Studio received a set of integrations that connect prompt-driven app building to the rest of the company's developer infrastructure. According to the keynote post, Google AI Studio now includes native Kotlin support to vibe code Android apps. The same surface gains Google Workspace integrations, a one-click deploy to Cloud Run, and support for Firebase services. Users who want to continue working on a project beyond what AI Studio supports can seamlessly export the complete project state to Google Antigravity.

That export path is significant. AI Studio is positioned as the entry point - a low-friction surface where prompts produce running applications - while Antigravity acts as the longer-form development environment. The handoff between the two avoids the historical pattern where prototype tools and production tools used incompatible artefacts, forcing developers to rebuild.

The vibe coding capability in AI Studio was introduced in October 2025, originally framed around moving from prompt to working AI app in minutes without needing to juggle API keys or wire models together. The May 19 update extends that capability into Android development specifically, with Kotlin support replacing the earlier web-only output, and adds Workspace and Cloud Run as supported deployment targets.

Android tooling rebuilt for agents

The Android portion of the keynote concentrated on a stable command-line interface and a set of open-sourced skills. According to Google, the stable Android CLI enables AI agents to tap directly into the heavy-lifting power of Android Studio. The CLI handles tasks like downloading the Android SDK, running an app on Android devices, and similar operations that previously required interactive use of the IDE.

That distinction matters more than it sounds. A command-line agent can be scripted, parallelised, and orchestrated from a central runner. An IDE-bound workflow cannot. By stabilising the Android CLI, Google has made it possible for any agent, language model, or third-party tool to use Android development capabilities programmatically rather than through user-interface automation.

Google also open-sourced what it calls Android skills - prescriptive recipes that help language models execute best practices for complex Android workflows. Two examples cited in the keynote post are migrating an existing project to Jetpack Compose and performing a Jetpack Navigation 3 migration. The skills act as a layer of expert knowledge that any LLM can use when working inside an Android codebase, reducing the gap between a model's general code-generation ability and the specific conventions of Android development.

To measure how different language models perform on those Android tasks, Google introduced Android Bench, described in the keynote post as an LLM leaderboard for Android development tasks. According to Google, the company added open-weight models such as Gemma 4 to the leaderboard during the I/O week so developers can see how other LLMs measure up.

A preview migration agent inside Android Studio rounds out the Android announcements. According to the keynote post, the agent migrates application code to a native Kotlin Android app, regardless of whether the source is React Native, a web framework, or iOS. Google said the agent analyses code and does the heavy lifting, turning migrations that would have taken weeks into hours. The claim is presented without independent benchmarks, and the feature was framed as a preview rather than a generally available product.

WebMCP: a proposed web standard for AI agents

The web announcements centred on WebMCP, which Google described as a proposed open web standard that allows developers to expose structured tools, like JavaScript functions and HTML forms, so browser-based AI agents can execute complex tasks with greater speed, reliability, and precision.

The technical motivation is documented. Browser agents today rely on screenshots, DOM parsing, and simulated clicks to accomplish tasks that the underlying website could expose directly as functions. That approach is fragile, expensive in compute, and prone to break whenever a page changes. Chrome's WebMCP framework, first introduced in an Early Preview Program on February 10, 2026, aims to replace pixel-level interaction with direct tool calls.

According to Google, the experimental WebMCP origin trial starts in Chrome 149, with support for Gemini in Chrome coming soon. Origin trials are a Chrome mechanism that lets site operators opt in to experimental APIs without the broader standardisation process being complete - in effect, a way to gather real-world implementation data before the specification is finalised.

For publishers, e-commerce operators, and any site whose value proposition involves transactions or structured interactions, WebMCP carries direct implications. Sites that expose tools natively will be easier for agents to use, faster to act on, and less prone to behavioural failure. Sites that do not may find themselves invisible to agent-based traffic, or accessible only through the slower, less reliable screenshot loop that current browser agents still depend on.

The framework arrives at a moment when bot and agent traffic has grown to a majority of all web traffic. According to research published by Imperva and cited in earlier PPC Land reporting, 51% of web traffic now originates from automated systems, much of it conducting the screenshot-and-click operations that WebMCP is designed to replace.

Modern Web Guidance, Chrome DevTools for agents, and HTML-in-Canvas

Three additional web announcements complete the developer keynote. Modern Web Guidance is a set of expert-vetted skills designed to give coding agents guidance on building performant, accessible, and secure web experiences. According to the keynote post, the preview supports over 100 use cases. The skill set integrates directly with Baseline - Google's cross-browser compatibility framework introduced at I/O 2023 - and is installable with a single click in Antigravity or via the command line through the npx modern-web-guidance install command.

Chrome DevTools for agents extends Chrome's developer tools to AI agents directly. According to Google, the capability helps agents scale their workflow with verifying, debugging, and optimizing code in real time. The supported operations include automated quality audits, real-world user experience emulation, hand-off sessions with auto-connect, and additional debugging tasks performed without manual oversight. The functionality maps Chrome DevTools' existing capabilities - originally designed for human developers - onto a programmatic interface usable by agent runtimes.

HTML-in-Canvas, available in origin trial, introduces a declarative API for embedding real DOM elements inside a canvas alongside WebGL and WebGPU content. According to the keynote post, the API allows developers to build immersive, 3D experiences that remain fully searchable, accessible, and interactable. The technical pattern matters because canvas-based applications historically lose the semantic structure of HTML once content is rasterised into pixels, breaking screen readers, search indexing, and copy-paste. HTML-in-Canvas preserves those properties while permitting the high-performance rendering that 3D and game-style web applications require.

Where this lands in Google's 2026 trajectory

The May 19 announcements are best read as a continuation of a developer-focused arc that Google has been building publicly since late 2025. The Gemini 3 launch in November 2025 introduced generative UI as a way for the model to produce complete interfaces dynamically. Subsequent releases extended that capability into Google Search through AI Mode and the standalone Gemini application. The I/O 2026 announcements push the same approach into the developer tooling that produces software, rather than the consumer-facing surfaces that consume it.

The advertising and marketing dimension of these changes is more indirect than for Google's consumer search products, but it is not absent. Google's pre-event marketing announcements in early May, including the journey-aware bidding, Smart Bidding Exploration, and demand-led pacing updates, were timed around the separately scheduled Google Marketing Live 2026 livestream on May 20. The relationship between the I/O announcements and the Marketing Live announcements is that they share underlying Gemini infrastructure but address different audiences: I/O speaks to developers, Marketing Live to advertisers.

For marketing technology teams, the practical question is what changes in the medium term. WebMCP and Modern Web Guidance both affect how AI agents interact with web inventory. If significant volumes of consumer activity migrate to agent-mediated workflows - from price comparison to booking to checkout - the underlying surfaces that perform well will be those that have exposed structured tools rather than those that have optimised purely for human visual interaction. The data infrastructure decisions that publishers and retailers make in 2026 will determine how visible they are to that traffic.

The other relevant context is the evolving relationship between Google's AI surfaces and the open web, where measurements of click-through rates, ad opportunity volume, and publisher traffic have all been moving in directions that disadvantage traditional web inventory. WebMCP and the related tooling do not resolve those tensions, but they reshape the technical layer at which the next round of integrations will occur.

According to a Wikipedia summary of the Antigravity platform's history, the original product was released alongside Gemini 3 on November 18, 2025, and was described at the time as Google's first agentic development platform. The version 2.0 announcement on May 19 separates Antigravity from its earlier IDE-only form and treats the agent harness as the primary product, with the IDE as one of several surfaces on top of it.

Sessions and remaining programme

Following the keynotes, Google will livestream additional sessions on May 19 and May 20, with more than 85 sessions, codelabs, and on-demand materials available starting May 21. According to the post, the full set of announcements and updates is published at io.google.

Google did not disclose pricing for Antigravity's $100-per-month AI Ultra tier inside the developer keynote blog post itself, although 9to5Google reported that the tier offers 5x usage limits for Antigravity workloads. The keynote also stopped short of giving specific availability dates for Modern Web Guidance beyond the early preview, and the WebMCP origin trial timeline depends on Chrome 149's rollout schedule rather than a separate Google announcement.

Timeline

Summary

Who: Google, through the Google I/O team and Google Developers, addressed an audience of developers, agency engineering teams, and marketing technology professionals. The platform changes affect users of Google AI Studio, Google Antigravity, the Gemini API, Android Studio, Chrome, and Chrome DevTools, alongside any developer working with Gemini models through third-party tools.

What: Google unveiled the Gemini 3.5 series of models, Antigravity 2.0 with a new standalone desktop application, the Antigravity CLI, an Antigravity SDK for self-hosted agent deployments, Managed Agents in the Gemini API for hosted execution, native Kotlin Android support and one-click Cloud Run deployment in Google AI Studio, a stable Android CLI, open-sourced Android skills, an Android Bench leaderboard, a preview migration agent for Android Studio, the proposed WebMCP open web standard for browser-based AI agents, Modern Web Guidance covering 100-plus use cases, Chrome DevTools for agents, and the HTML-in-Canvas declarative API.

When: The announcements were published on May 19, 2026, the opening day of Google I/O 2026. Additional sessions are livestreamed on May 19 and May 20, with the full set of more than 85 sessions, codelabs, and updates available on demand starting May 21.

Where: The keynote was held at Google's headquarters in Mountain View, California, with sessions distributed online to a global audience. The full announcement set is published at io.google and on the Google Developers Blog.

Why: The release matters for the marketing community because the same Gemini infrastructure that now powers agent-first developer tooling also underlies Google's advertising, measurement, and Search products. WebMCP and Modern Web Guidance change the technical conditions under which AI agents interact with the open web - which directly affects how publishers, retailers, and advertisers will be visible to agent-mediated traffic. The shift from assistive AI to autonomous agents has implications for how marketing automation is built, how campaigns are measured, and how the underlying inventory is exposed to the systems that will increasingly act on consumers' behalf.

Share this article
The link has been copied!