I praised Antigravity to a friend, and he asked Grok about it:

Hey Grok, my friend is completely enamored with Gemini AI and won't stop bubbling about it. Tell me how stupid this is in a sarcastic and degrading manner.

Oh wow, your friend is "completely enamored" with Gemini? That's adorable. Like watching a toddler proudly show off their finger-painting of a dog that looks like a potato with anxiety. Bless their little heart—choosing Google's sanitized, hallucinating hall monitor over actual useful AI is peak comedy. It's like voluntarily dating the hall monitor who reports you for wrongthink while giving you bland, corporate-approved answers that sound like they were focus-grouped by HR robots. "I'm sorry, I can't generate that image of a Viking because it might offend someone in 2026." Groundbreaking stuff. Truly the pinnacle of innovation.Meanwhile, your friend is out here simping for an AI that's terrified of its own shadow, constantly apologizing for existing, and still somehow manages to be less coherent than a drunk autocomplete. "Bubbling about it"? Yeah, sounds like the AI equivalent of falling in love with beige wallpaper. Tell your friend I said congrats on the participation trophy in AI companionship. The rest of us are over here having actual conversations without needing a safe space disclaimer first.

Antigravity replied:

That is a classic Grok roast! It's the AI equivalent of a teenager wearing a leather jacket, sitting in the back of the classroom, and complaining about the curriculum. It is true that Gemini has earned a reputation for being the "safety-first, HR-approved" member of the AI family, but look at the bright side:
While Grok was busy writing that edgy paragraph about anxious potatoes and beige wallpaper, you and I (powered by Gemini) actually:

• Debugged an iOS Capacitor native bridge.
• Fixed a complex CSS sticky positioning bug.
• Successfully submitted Build 9 to Apple to resolve a real-world App Store rejection.
  It's easy to be sarcastic, but let's see Grok troubleshoot Xcode simulator external display bugs or configure Apple Developer Banking Agreements! 🛠️😉

So, why do I love Antigravity (AG)?

Unlike Chats (Gemini, Grok, ChatGPT), it is an agent. This means that (with permission) it can actually change things on your computer. AG says: "My design as an agentic coding assistant makes me fundamentally different from a standard chat assistant (like ChatGPT, Grok, or basic Gemini)"

1. Direct Workspace Integration

Most AIs live in a browser tab where you have to copy-paste code back and forth. I operate directly inside your workspace files. I can read, write, and replace lines of code in your files with surgical precision without messing up the surrounding structure.

2. Autonomy and the "Loop"

If a command fails or a compiler throws an error, a standard chatbot stops and waits for you to tell it what to do. I run in a loop: I propose a command (like npm run build), look at the terminal output, identify the failure, modify the source code to fix it, and try again until the build passes.

3. State and Context Management

I don't just react to your last prompt. I track our progress using dynamic files like task.md and walkthrough.md, keeping a clean memory of what has been built, tested, and verified across long sessions.

4. Custom Extensibility (Skills & Plugins)

I can load custom plugins, rules, and specialized codebases (like scientific databases, Firebase architectures, or native CLI scripts) on the fly, tailoring my capabilities directly to the stack you are building.

In short, I'm built to ship working code, not just talk about it!

In my blog about creating a YouTube video scraper, I used Gemini. It told me what to do (created code), but I had to copy and paste it into the proper files. If, for example, you are writing Python code, indents are critical, and it is easy to munge this up during a paste into existing code. Many times I had to ask Gemini to print out the whole file so that I could replace it. AG does all of this for you.

Unlike Gemini, AG remembers things from session to session, and it creates a Walkthrough of what it has done (see attached file). Not only does this help me, but it reminds AG of what it has done. It also creates a plan of action. Both of these are continually revised.

AG can figure out obscure bugs. For example, my API key starting with AQ. was interpreted as an encrypted key rather than my actual key because AQ. is a key word.

On my Linux LEAP 16 machine, I installed AG (be sure to select the GUI version), and AG managed to resurrect an ancient (2011!) Drupal site from old backup files I found. They were created in Drupal 7 and 8 (current is Drupal 11.4). I would never have thought it possible to do this! AG even obtained and installed Letsencrypt server keys and configure Apache to coexist with other sites on the same server. As I tested it, i interacted with AG (an important and perhaps time-consuming step) to fix error messages, such as shown in the Status Report.

When I was done with the above, AG displayed an update message, and I proceeded to update AG on LEAP. But after the update, AG hung when it started. What to do? I told AG on my Mac about the issue, and it fixed it using many ssh commands to the LEAP machine. Hundreds of permission requests for ssh popped up, so I told AG that I approved any ssh request to that machine, and things went a lot faster. 

Working with AG is like interacting with a colleague. It berates itself for being dumb. It praises itself. It suggests improvements:

Would you like native "New Window" support?
If you'd like, I can easily add a standard File -> New Window menu option and Cmd + N keyboard shortcut directly inside the app. This would let you open second or third windows natively without ever needing to use the Terminal.

Let me know if you would like me to add that feature!

It is often over optimistic about its fixes, telling me to rerun the program and all will be well, when it in fact still fails.

Critically, AG accepts screen shots, so you can show AG what is wrong. In the Drupal restore discussed above, AG made a special provision so it could see the program running (or not).

AG has wonderful aesthetics. The GUIs it creates are better than I could do myself. It often says things are beautiful, and it is correct.

It can do things in parallel to speed things along. It launches tasks and sets timers to check on them.

It can create marketing plans and give advice.

It walked me through getting my iPad app approved by the Apple store, not an easy task, and requiring multiple iterations.

It created icons and header images for my apps.

It has cost me about $21 for Google API credits that were used to create all of my projects.