Google is introducing Google Antigravity, an agentic development platform designed to redefine the Integrated Development Environment (IDE) for the era of autonomous coding. Built around the capabilities of Gemini 3, the platform shifts development from prompt-by-prompt interactions toward higher-level, agent-driven execution that can plan and complete complex software tasks.

Antigravity begins as a familiar AI-powered IDE but evolves quickly into an agent-first environment, blending browser control, asynchronous agent execution, and multi-surface autonomy. The platform is now available in public preview at no charge, offering generous rate limits for Gemini 3 Pro.

Google positions Antigravity as the “home base” for software development in the age of intelligent agents. With models like Gemini 3 now capable of running autonomously for longer periods across different surfaces, developers need an environment that reflects this new level of capability.

The goal is to support anyone—from enterprise developers to hobbyists—who has an idea and wants to “experience liftoff” by turning that idea into reality through agentic execution. Antigravity is Google’s answer to a world where users will increasingly operate at higher abstractions while AI agents handle more of the underlying complexity.

Antigravity is built around four foundational tenets—trust, autonomy, feedback, and self-improvement—that together define how users and agents collaborate throughout the entire development lifecycle. These principles shape how Antigravity organizes tasks, exposes reasoning, verifies outputs, responds to user guidance, and learns from past executions.

By grounding the platform in these pillars, Google aims to create an agent-first development environment that is reliable, transparent, and capable of supporting increasingly complex, end-to-end software workflows.

Current AI coding tools often fall into two extremes: showing every tool call with overwhelming detail or showing none of the agent’s process at all. Antigravity moves to a middle ground by grouping actions into task-level abstractions, offering users context without flooding them with raw logs.

To support trust, agents generate Artifacts—such as task lists, implementation plans, walkthroughs, screenshots, and browser recordings—giving users tangible, verifiable outputs in familiar formats. These Artifacts help users better understand what the agent is doing and how it is validating its work, not just evaluating the final work itself.

At its core, Antigravity includes a state-of-the-art, AI-powered IDE experience through its Editor view, combining tab autocompletion, inline natural-language code commands, and a fully functioning side-panel agent. This Editor environment preserves the familiarity of a traditional IDE while layering in real-time agentic intelligence, allowing users to write code, refine logic, and collaborate with an embedded agent through conversational workflows.

But Antigravity also introduces something fundamentally new: an agent-first architecture where agents aren’t confined to a single editor, terminal, or browser. Instead, they operate across all surfaces simultaneously, coordinating work in parallel and moving between tools as needed. To support this shift, Antigravity adds the Manager surface — a mission-control dashboard that flips the traditional paradigm. Rather than agents being embedded within individual tools, the tools themselves are embedded into the agent, allowing users to create (or “spawn”) multiple agents, observe their progress, and orchestrate them across many parallel workspaces from one central control point.

This intentional separation between the synchronous Editor and the asynchronous Manager creates a workflow where users can move between surfaces instantly, handing tasks off to agents without losing context. By not forcing both modes into a single interface, Antigravity supports a more fluid development rhythm, where agents work continuously in the background while users check in intermittently. This structure anticipates a future shaped by increasingly capable models like Gemini 3, where asynchronous, agent-driven development becomes the norm.

Even with improved agentic intelligence, no agent completes 100% of tasks perfectly. A major limitation of remote-only agent models is that they make iteration difficult — if a user can’t easily correct the last 20% of a task, the agent becomes more work than help.

Antigravity addresses this by making feedback a central part of the workflow. Because the agent begins with local operation, users can give fast, natural feedback across any surface or Artifact, all without interrupting the agent’s work. Users can leave Google-doc-style comments on text Artifacts, annotate screenshots with select-and-comment tools, or provide lightweight directional guidance without stopping or restarting the agent. This flexibility keeps the agent from becoming a black-or-white system where it’s either perfect or unhelpful — instead, users can guide it toward the right outcome. Antigravity automatically integrates that feedback into the ongoing execution, helping the agent resolve the remaining work more efficiently and making collaboration feel natural rather than interruptive.

This asynchronous feedback loop becomes even more powerful in team settings, where multiple stakeholders may need to review or refine an agent’s work at different times. By enabling artifact-level comments and targeted corrections, Antigravity reduces the friction traditionally associated with collaborative debugging or code reviews. This allows teams to maintain velocity while still ensuring quality and alignment across complex projects.

Learning is built directly into Antigravity’s design. Antigravity maintains a knowledge base that both retrieves past learnings and contributes new ones as agents work. This can include explicit elements—like useful code snippets or architectural notes—as well as higher-level patterns, such as the steps agents took to complete certain subtasks successfully. Over time, this enables agents to improve performance on repeated workflows.

In its public preview, Google Antigravity is launching as an accessible agentic development platform with broad compatibility and multiple model options, giving developers a practical way to explore agent-first workflows across different operating systems and tools.

Google encourages developers to explore the documentation, browse use cases, and follow upcoming feature drops across its blog and social channels.

Download Antigravity today.

Q: What models can developers use within Google Antigravity?
A: Antigravity supports Gemini 3, Claude Sonnet 4.5, and GPT-OSS, giving developers a range of model options directly inside the agentic platform.

Q: How does Antigravity build trust with users?
A: By grouping tool calls into task-level abstractions and generating Artifacts—like implementation plans and screenshots—developers can validate agent reasoning without being overwhelmed by low-level logs.

Q: What makes Google Antigravity different from today’s IDEs?
A: Antigravity introduces an agent-first approach, with synchronous development in the Editor and asynchronous orchestration in the Manager, enabling higher-level agent autonomy across multiple surfaces.

Q: How does Antigravity help users validate agent work without reviewing every tool call?
A: Antigravity uses higher-level, task-based summaries and Artifacts—such as task lists, walkthroughs, implementation plans, screenshots, and browser recordings—to show users exactly what the agent did and how it verified its own work. This gives developers the context they need to trust the result without digging through raw logs.

Google Antigravity is more than a new IDE — it marks a shift toward an era where software is increasingly built through collaboration between humans and autonomous agents. For developers, this means spending less time on mechanical tasks and more time on architecture, creativity, and decision-making. Antigravity’s agent-first approach allows engineers to offload multi-step workflows, validate work through Artifacts, and guide agents with lightweight feedback instead of micromanaging every line of code.

For non-developers, Antigravity lowers the barrier to turning ideas into working products. As agents become more capable across surfaces — editor, terminal, and browser — people who have never written code can increasingly participate in building apps, automating workflows, or prototyping new concepts. This could expand who gets to create software, bringing more voices and perspectives into the innovation cycle.

At the industry level, Antigravity signals a major strategic shift: software development is moving from tool-driven workflows to agent-orchestrated workflows. Companies will be able to scale engineering output, shorten development cycles, and redistribute effort toward higher-level design and problem-solving. As agentic platforms mature, we’ll see new roles, new expectations for product velocity, and new competitive dynamics shaped by how effectively organizations leverage autonomous systems.

In short, Antigravity represents the early stages of a broader transformation — one where software creation becomes faster, more accessible, and more collaborative than ever before.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing, image, and idea-generation support from ChatGPT, an AI assistant. However, the final perspective and editorial choices are solely Alicia Shapiro’s. Special thanks to ChatGPT for assistance with research and editorial support in crafting this article.