Google I/O 2026: Antigravity is putting AI on steroids these 5 ways
Follow Us
Among the more surprising revelations from I/O is that Gemini CLI is being retired. When Google released Gemini CLI last year, its purpose was to make Gemini available right inside the terminal. Having garnered over 100,000 stars on GitHub, 6,000 merged pull requests, and millions of users, Google claims the tool has simply been outgrown. The issue is not about popularity but rather scope. Users increasingly need to employ multiple agents that can communicate with each other and divide the workload and resolve complex issues, something that requires terminal tools to operate on a common backend with everything else developers are doing.
SurveyAlso read: Google wants to compete with Claude Mythos with its new CodeMender, here is how
Antigravity CLI is the new tool, and it’s much more than just a rebranded version of the old one. Antigravity CLI is built in Go and is therefore more responsive and faster, offers asynchronous workflows to control multiple agents in the background, and uses the same agent harness as Antigravity 2.0, ensuring that any improvement to core agents will apply to both products at once. The migration period ends on June 18, 2026, after which Gemini CLI and Gemini Code Assist IDE extensions will no longer process requests from Google AI Pro and Ultra customers, as well as those using the free tier. Now, the five ways Antigravity is changing the game.
Antigravity 2.0 turns your desktop into an agent control tower
Antigravity 2.0 is the new stand-alone desktop application optimized for the agent experience. It serves as the main hub where you interact with multiple agents running parallel tasks, dynamic subagents for parallelizing tasks, scheduled tasks for background automation, and ecosystem integration from Google AI Studio, Android, and Firebase. Imagine this as not just an interface but a command center where you could run a massive code refactoring project along with some research work concurrently without worrying about locking your entire workflow.
Managed Agents make spinning up AI a single API call
The introduction of the Managed Agents in the Gemini API means that you can now create an agent that can reason, use tools, and execute code in an isolated Linux environment with one API call. Every conversation spawns an environment that can be used further in subsequent calls while retaining everything that was done previously. This is Google’s infrastructure move where it gives developers the same agent harness that it uses internally, which is co-optimized with Gemini 3.5 Flash. You can even extend the agent with custom instructions and skills using markdown files as new custom agent templates are being provided in the Google AI Studio Playground.
Gemini 3.5 Flash is the engine everything runs on
None of this will work unless the underlying model is fast enough to handle it. Gemini 3.5 Flash beats Gemini 3.1 Pro across almost all benchmark tests while being four times faster than other frontier models, which is exactly what you want when implementing agentic workflows that involve agents waiting on each other in chains. That’s the distinction between an AI assistant and an AI workforce.
Google AI Studio goes mobile and talks to Workspace
The Google AI Studio mobile application, which you can pre-register for, will enable you to capture your idea while on-the-go and create a working prototype before you even sit down at your desk. This represents a true change in the beginning of the development loop. Furthermore, agents can now make native calls to Google Workspace APIs and integrate them in their applications, and entire projects can be exported to local development in Antigravity in one click, while preserving the full project context. Moreover, there’s native Android support, meaning that developers can now create top-notch Android applications using prompts alone, and publishing them straight to the Google Play Console test track from within Google AI Studio.
A $2 million hackathon and a $100 plan to go with it
Google is putting its money where its mouth is when it comes to the ecosystem approach. The Build with Gemini XPRIZE Hackathon offers a prize pool worth $2 million, which is the largest amount yet offered by any hackathon. In addition, developers interested in more than just hobbyist use cases will appreciate the launch of the Google AI Ultra plan, which begins at $100 a month and provides users five times the usage limits of the AI Pro plan in the Google Antigravity system. Existing and new Ultra users can receive an additional $100 in credits for a limited time until May 25, 2026.
Also read: Google Project Genie can now turn real locations into interactive AI worlds: Here is how
The through-line across all of this is the same: Google is done treating AI as a feature. Antigravity is a platform bet and I/O 2026 is where it officially becomes the foundation everything else is built on.
