Simply when the tech international concept Anthropic’s Claude Code and OpenAI’s Codex had locked down the AI-driven building house, Google went forward and flipped the script.
Whilst competition have spent the closing 12 months refining incremental autocomplete and context home windows, Antigravity 2.0 has quietly rewritten the principles of the sport.
Right here is strictly how Google’s newest powerhouse is thrashing the business’s favourite dev equipment at their very own recreation.
What’s Antigravity 2.0, anyway?
Let’s perceive the context right here
When Google first dropped Antigravity closing 12 months, I can admit I used to be intrigued however completely at a loss for words. Whilst it’s mainly a VS Code fork, Antigravity had this twin identification: a typical textual content editor and a futuristic Agent Supervisor, which looks as if a unconditionally other app.
Seeking to cut up my mind between writing code manually and tracking background AI brokers in the similar cluttered window felt like a large problem.
Google understood the identification disaster, and at I/O 2026, the corporate introduced model 2.0. Now, the Antigravity instrument is divided into two portions: a devoted Agent Supervisor that appears and feels so much like Claude Code and Codex and an Antigravity IDE device which appears like VS Code with the Gemini add-on.
The most productive a part of this pivot is the brand new UI. It’s streamlined, responsive, and well-built. Extra importantly for my day-to-day workflow, it’s light-weight too. In contrast to Codex, which continuously turns my pc into an area heater and units the cooling fanatics to most blast, Antigravity 2.0 runs totally cool within the background, even if it’s coping with advanced, parallel duties.
Antigravity 2.0 options
Don’t overlook about velocity, too
The use of the brand new standalone interface is a night-and-day distinction from the cluttered, split-screen identification disaster of the outdated app. Antigravity 2.0 acts like a high-level command heart. Google has packed a ton of agentic energy into this free up, and a couple of particular options redefine the workflow for me.
The most important get advantages is the way it handles my recordsdata. Model 2.0 introduces an abstracted UI constructed round Initiatives. Now, a unmarried undertaking workspace can span a couple of, completely separate folders throughout my device.
It offers me the liberty to let an agent take a look at a frontend repo, a backend provider, and a documentation folder suddenly.
With the outdated app, if I gave the AI a large, multi-part process, it could attempt to clear up the entirety in one chat. The context window would get bloated with terminal logs, and the fashion would in the end lose the plot.
Antigravity 2.0 solves this with dynamic subagents. The principle orchestrator seems at a fancy downside, breaks it down, and spins up subagents to take on other portions in parallel.
Scheduled Duties is the place the futuristic workflow truly clicks. I now not have to take a seat down, steered the AI, and look forward to a reaction. I will use Scheduled Duties to automate regimen background tests. As an example, I will configure an agent to get up robotically at nighttime, run localized assessments throughout my initiatives, scan for mistakes, and care for elementary upkeep autonomously whilst I sleep.
Reside voice transcription is any other quality-of-life improve I’ve with Antigravity 2.0. As it’s powered natively by way of the most recent Gemini audio fashions, I don’t need to document a voice observe, look forward to an add, and hope for a blank transcription.
Complex slash instructions
Reasonably useful
As a result of Google optimized the applying layer round Gemini 3.5, the execution velocity is very good. I ran some advanced duties on all 3 and Antigravity was once quicker than Codex and proper up there with Claude Code on the subject of producing code. The code high quality was once commendable too.
Google has additionally applied complicated slash instructions that permit me direct the AI like an engineering supervisor.
Amongst them, I exploit the /purpose command at all times. Right here, I ask the agent to run with none hurdles, self-correct via mistakes, and bypass any want to approve movements from my facet. It’s somewhat useful for my small initiatives.
/grill-me is any other command I exploit continuously. As an alternative of guessing my intent, the agent pauses and cross-examines me earlier than touching a unmarried dossier. /browser and /agenda are the opposite two instructions you’ll be able to use. I am hoping to peer extra such helpful instructions in long term updates.
Whilst Google is absolutely inquisitive about Antigravity 2.0, the corporate hasn’t totally ditched the standard IDE app. In fact, the corporate will in the end take away the Agent Supervisor from it and ask customers to transport to the two.0 model, however if you’re coping with outdated initiatives and want a Gemini-powered textual content editor, you’ll be able to at all times obtain the unique app from the site.
Google’s new AI coding beast
If Claude Code and Codex felt like the height of the primary era of AI coding equipment, Antigravity 2.0 appears like the arriving of the second one. In spite of everything, builders don’t care about AI rivalries; we care about delivery blank code with out the friction.
Whether or not you are attempting to one-shot a fancy microservice or seamlessly orchestrate a large multi-file legacy codebase, one that is now simple: the bar for dev AI has simply been raised, and the contest has some critical catching as much as do.
