Listen: https://serve.buildtall.systems/0b5f6e28c8703a3216bfbd030dbf2408842dcc450702702ff4f5f1eb6c0f9b4f.mp3
Source: https://seroter.com/2026/11/18/wait-is-google-antigravity-good-now/
Richard Seroter's new post is called 'Wait, is Google Antigravity good now?' He works at Google and uses Antigravity every day, both an internal build and his personal Google AI Pro subscription, and he says he has tried to give a fair assessment. His argument is that in November 2026 Antigravity has caught up with competitors in key areas, innovated in a few others, and is now powered by legitimately good coding models. He walks through catch-up features like a plugin marketplace, manual context compaction, lifecycle hooks, AGENTS.md rules, auto mode, remote control, side questions, browser use, and raw key auth, then highlights unique additions.
Seroter points out that in the last ninety days the Antigravity team shipped more net-new capabilities than any other leading AI developer tool. Across the app, CLI, IDE, SDK, and IDE extensions, that added up to one hundred fifty versioned releases and more than nine hundred public changelog items. He contrasts that with Claude Code, Microsoft, Devin, OpenCode, and OpenAI Codex shipping fewer items in the same window.
The first link, 'shipped the first version of the Antigravity IDE', is the November 2025 launch post. It frames Antigravity as an agentic development platform built around Gemini 3, with an IDE experience plus an agent-first Manager surface. The team emphasizes trust, autonomy, feedback, and self-improvement. Seroter says this was basically an agent-centric VS Code fork, a fine tool with fine models but nothing world-beating.
Next, 'the product rebooted with Antigravity 2.0' covers the May 2026 relaunch as a standalone desktop app with no IDE. It added dynamic subagents, async task management, JSON hooks, scheduled tasks, and a project model, and it broadened the product beyond coding. Seroter calls this a better tool with better models that put Google in the game, even if it did not leapfrog competitors.
Seroter's first catch-up item is the plugin, skill, and MCP marketplace. The link, 'a helpful “customizations” experience', is Antigravity's plugins documentation. It explains that plugins package reusable skills, MCP servers, custom agents, rules, and lifecycle hooks into versioned units, with marketplace discovery and management across the app, CLI, and IDE.
Next, 'lifecycle hooks' is the build-plugins documentation. Seroter notes that Anthropic invented this taxonomy, and Antigravity now supports lifecycle hooks for PreToolUse, PostToolUse, Stop, and SessionStart events. The docs show hooks defined in a hooks.json file, with command or prompt handlers that can inspect, block, or augment agent behavior.
The next item is default auto mode, linked as 'added to the'. This permissions page explains Antigravity's permission presets. Auto mode runs commands in a terminal sandbox and automatically decides whether safe, reversible commands outside the sandbox can proceed without prompting, while risky actions still ask. Seroter says other tools offered auto permission mode, and Antigravity added it to its existing turbo, request review, and default options.
Seroter's catch-up list continues with remote control of a local terminal. The article 'steer your machine from anywhere' documents Antigravity Remote Control. It lets you securely connect from any browser to drive desktop and CLI sessions, either through an interactive tunnel for a single session or a persistent headless daemon. Seroter says Claude Code popularized this, and now you can steer your machine from a phone.
The next link, 'Antigravity now has that too', covers side-channel questions and mid-turn steering. The /btw command forks the current conversation, lets a side question read files and search the workspace while the main agent keeps running, and keeps the history clean. Seroter's take is that this is a good experience because you can pick which model answers the question, and you can add relevant questions mid-stream that are immediately sent to the main trajectory.
Autonomous browser use is next, under the heading 'rich /browser experience'. The browser use docs describe the built-in browser agent opening pages, interacting with running apps, inspecting console errors and failed network requests, taking screenshots, and verifying visual fixes, with parallel tabs and strong safety guardrails. Seroter acknowledges that browser use existed before, was removed, and has now come back as a much richer experience.
For raw key auth and bidirectional piping, Seroter points to two linked docs. The first is 'stream JSON out'. This is the headless mode documentation for the Antigravity CLI. It supports text, JSON, and newline-delimited stream-json output, and it can also stream prompts from stdin, which makes it easier to pipe into Unix tools and drive background workers. Seroter says this matches Aider, OpenCode, and Claude Code and helps background workers.
The companion authentication article is 'API key'. It explains how to set a Gemini API key with a model provider setting for headless and CI runs, bypassing browser sign-in, and it covers Application Default Credentials for enterprise workflows. The overall point is raw key auth for non-interactive scenarios.
Then the post shifts from catching up to jumping ahead. 'Remote Agents' documents Antigravity Remote Agents, which run sessions in an isolated Google-managed cloud environment attached to GitHub repositories, Google Drive files, or uploaded files. That means work continues even when your laptop closes, and you can resume from the app, CLI, or web. Seroter frames this as a relatively unique feature for AI builders.
Next, Seroter discusses the Gemini 4 models. The article 'benchmarks' is the Artificial Analysis Coding Agent Benchmarks page. It combines FrontierCode-AA 2.0, Terminal-Bench 4.1, and SWE-Atlas-QnA-AA to measure real-world coding agent performance, cost, token usage, and execution time. Seroter says that in both benchmarks and real-world usage, developers are increasingly leaning on Gemini 4 Argon, Carbon, Hydrogen, and Orchestrator.
The model access details are in 'different access based on subscription tier'. This documentation explains that Gemini 4 Carbon, Argon, Hydrogen, and the two Orchestrator options are available on different plans, with Claude Sonnet 5.5 and Opus 5.5 support added for some tiers. Seroter notes that access differs by subscription tier, and that Antigravity also added support for Claude Sonnet 5.5 and Opus 5.5.
One of the standout new features is '/swarm mode'. The Swarm documentation describes a coordinator splitting a task across named workers that run in parallel and post to a shared message board, with a verifier checking diffs and tests, and a monitor watching for stalls. Seroter calls it pretty awesome, easy to start, and says the visualizations make it easy to see what is going on.
Seroter calls the next one brand new. 'writes a runs a Python script' is the Dynamic Workflows documentation. The agent writes and runs a short Python script that coordinates agents through parallel or sequential phases, producing structured results and keeping the main conversation clean. Antigravity includes its own Python so there is no installation required.
Next is full-duplex voice. 'ongoing two-way conversation with the agent' documents voice mode enabled by Gemini Live. You can talk to the agent, steer tasks while it works, answer permission prompts by voice, and get spoken progress updates. Seroter highlights that unlike tools that only accept an audio recording as input, Antigravity supports an ongoing two-way conversation, which pairs well with remote agents and on-the-go work.
The next unique capability is direct subagent messaging. 'DM a subagent directly from the main conversation' is the mentions and quoting documentation. In the CLI, you can start a prompt with an at sign and a subagent name to send it a message directly, including to a finished subagent that picks up its earlier context. Seroter says Antigravity has made a lot of investment in making multi-agent feel natural.
Generative interfaces in the main chat are covered by 'create interactive widgets'. The /generative_ui command has the agent build interactive HTML widgets such as diagrams, charts, and step-by-step explainers, shown inline or in the side panel, sandboxed and sized to fit. Seroter says you can ask the agent to create interactive widgets for the output.
On surface breadth, the article 'the free tier, as well as paid tiers' is the Antigravity pricing page. It lists the free individual plan, Google AI Pro, Google AI Ultra, and a business plan through Google Cloud, with different models and quotas. Seroter says Antigravity has unmatched surface breadth across desktop app, IDE, CLI, IDE extensions, and SDK, all available in the free tier, while paid tiers add newer models, more quota, and features like multi-agent teamwork and remote control.
The enterprise features begin with 'now included'. This enterprise overview page says Antigravity operates under your existing Google Cloud Terms of Service with centralized administrative controls. Seroter says Antigravity usage is now included with Gemini Enterprise Standard and Plus seats, or available seat-free with Gemini Enterprise pay-as-you-go.
On pooled quotas, Seroter links to 'pooled together'. The quotas and pay-as-you-go documentation explains that enterprise users share a pooled quota per edition, project, and location, and administrators can set a single-user multiplier to cap any one person's consumption. Seroter says the individual user quota is pooled so the whole team can tap into it, with controls to prevent one user from chewing it all up.
Finally, Seroter closes with a call to action and links 'Jump aboard the train'. That homepage presents Antigravity as the next-gen agent platform with four surfaces: the App, CLI, IDE, and SDK, aimed at developers in the agent-first era, and it highlights free individual access plus enterprise availability. Seroter says Antigravity is faster, more usable, more functional, and backed by better models than ever before.
Across the post, the through-line is that Antigravity has closed its feature gaps while adding differentiable capabilities. Seroter also names remaining gaps: OS-level desktop computer use, autonomous PR merge and inbound chatops triggers, and ambient memory. It may not be done, but his conclusion is that Antigravity deserves another look.
- Wait, is Google Antigravity good now?
- shipped the first version of the Antigravity IDE
- the product rebooted with Antigravity 2.0
- a helpful “customizations” experience
- lifecycle hooks
- added to the
- steer your machine from anywhere
- Antigravity now has that too
- rich /browser experience
- stream JSON out
- API key
- Remote Agents
- benchmarks
- different access based on subscription tier
- /swarm mode
- writes a runs a Python script
- ongoing two-way conversation with the agent
- DM a subagent directly from the main conversation
- create interactive widgets
- the free tier, as well as paid tiers
- now included
- pooled together
- Jump aboard the train

