Google just released something that feels less like an update and more like a glimpse into the future of coding. It is called Antigravity, and it is not another AI autocomplete tool. It represents an entirely different approach to thinking about how developers and AI collaborate.
What Exactly Is Antigravity?
Antigravity is a new IDE built on top of Visual Studio Code (yes, yet another VS Code wrapper)that treats AI agents as first-class developers, not as helpers. Instead of you writing code while an AI suggests improvements, you describe what you want built, and the agent plans, codes, tests, debugs, and verifies everything across your editor, terminal, and browser simultaneously.
Google released it on November 18, 2025, alongside Gemini 3, their most capable AI model yet. The platform is free and available for Windows, macOS, and Linux with what Google calls “generous rate limits” that refresh every five hours.
The Three Surfaces That Change Everything
Antigravity has three main surfaces where you work.
The first is Editor View, which resembles a standard IDE with an agent available in the sidebar, ready to modify code and suggest changes.
The second is Manager View, which is essentially mission control for running multiple agents simultaneously across different workspaces and tasks.
The third surface is the browser itself. Agents can directly control your browser through a Chrome extension, take screenshots, record interactions, and verify that what they built actually works before handing it over to you.
Artefacts Are The Key Innovation Nobody Is Talking About
Instead of showing you hundreds of confusing tool calls and logs, Antigravity uses what Google calls “Artifacts.” These are human-readable deliverables like task lists, implementation plans, screenshots, and browser recordings that document exactly what the agent did and why.
You can add Google Docs-style comments directly on any artifact to give feedback. The agent reads your comments and incorporates them mid-task without restarting from scratch. This is not just cleaner than raw logs. It transforms how developers can actually audit and trust what AI is doing.
The Four Design Principles Behind the Magic
Google built Antigravity around four explicit tenets that separate it from other AI coding tools. Trust comes from artifacts and verification steps that let developers see what happened. Autonomy means agents can operate across editor, terminal, and browser without constantly asking permission.
Feedback is woven into the system through inline commenting on any artifact or screenshot. Self-improvement happens because agents can save code snippets and workflows they used successfully before, building an internal playbook of reusable procedures that grows smarter over time.
Why “Vibe Coding” Is Not A Gimmick
Gemini 3 powers Antigravity, and it introduced something Google calls “vibe coding.” This means describing what you want in natural language with almost no technical syntax, and the model translates it into a fully interactive application. You say “Build me a real-time flight tracking dashboard with dark mode,” and it does it.
The model handles multi-step planning, handles edge cases, creates visually polished interfaces, and debugs issues automatically. It is not perfect, but it works well enough that developers are legitimately asking if this changes what their job even is.
Multiple Models, Not Vendor Lock-in
Antigravity does not trap you into using only Google’s Gemini models. You can use Gemini 3, Anthropic’s Claude Sonnet 4.5, or OpenAI’s GPT-OSS, all within the same environment. This gives developers real model optionality instead of betting everything on a single vendor.
You can literally switch which model an agent uses mid-project if one is hitting rate limits or if you want to try a different approach. This is a breath of fresh air compared to most AI tools that lock you into one ecosystem.
Editor View vs Manager View… Understanding The Difference
Editor View is for developers who want to stay hands-on. An agent sits in a sidebar, you can chat with it, it can suggest inline code changes, and you retain full control. It feels like working with a very smart colleague reviewing your code in real-time.
Manager View is for developers comfortable with more autonomy. You launch multiple agents across separate workspaces, describe high-level tasks, and then monitor them from mission control.
They run asynchronously and coordinate with each other, like supervising a team of junior developers all working on different parts of the same project.
The Knowledge System Creates Organisational Memory
One feature flying under the radar is the Knowledge system. When an agent solves a problem or discovers an effective workflow, it can save that approach to a reusable playbook. Future agents can query this knowledge base and reuse successful procedures instead of rediscovering the same solutions repeatedly.
Over time, this becomes an organisation’s accumulated intelligence… a library of what works in your specific codebase and tech stack that gets better with every project completed.
Why This Matters More Than You Think
Antigravity is not trying to replace developers. It is trying to eliminate the parts of coding that are repetitive, mechanical, or just require someone to stitch together existing solutions. Developers become architects and quality assurance specialists instead of people manually typing boilerplate code.
For someone building a side project or prototype, this is genuinely transformative. You can sketch an idea in the afternoon and have a working application by evening. For enterprise teams, it could reshape how sprints work and what developers focus on.
The Elephant In The Room… Can You Trust The AI?
There is a legitimate concern lurking underneath all this. If an agent is writing most of your code, how do you verify it is not doing something dangerous or insecure? Google acknowledges this directly in their terms, warning that Antigravity has “certain security limitations” including data exfiltration and unintended code execution risks.
The mitigation is to run Antigravity in a sandboxed environment and crank up human review settings to require approval on everything the agent does. But this defeats the speed advantage that makes the tool interesting in the first place.
The Honest Assessment
Antigravity is a genuinely impressive step forward in agent-assisted development. It is not perfect, it has real security concerns, and it is clearly still in early stages. But the architecture is thoughtful, the execution is smooth, and it costs nothing to try.
If you are sceptical about AI agents writing production code, you are right to be sceptical.
But if you are curious about what the future of developer tools looks like, downloading Antigravity and playing with it for an afternoon will answer more questions than reading any think piece can.