Making a Blog with Antigravity and GitHub Pages

Why Blog in 2025?

Although it’s kinda questionable if “blogs” are really a thing needed anymore, I’ve gone ahead and made one. It’s something that I’ve wanted to do for a while, and I think at one point I did have something setup running on a digital ocean machine about 10 years ago. However, I didn’t really stick with it, and didn’t really like working with it. I was dealing with some sort of software to do the management at the time - it had an admin interface, it had a limited set of themes and things that I could actually change. I had almost no experience whatsoever with software engineering, and couldn’t get into the nuts-and-bolts of the thing, and just moved on.

A Decade of Growth

Right, so… no software engineering capability. That’s one thing that has changed these last 10 years. I spent nearly the whole time working at GitHub, where it’s basically impossible to not be some kind of software engineer, or at least know the broad ecosystem of software engineering tooling, and the “github flow” of working with Pull Requests to carefully introduce new code into production. With the rapdily advancing capabilities of LLMs, and, importantly, the tooling around these LLMs, I don’t feel at all weird calling myself a software engineer at this point.

The CLI Awakening

Specificially, I’ve gotten very engaged and energized these last couples months. I can point to this video specifically https://www.youtube.com/watch?v=MsQACpcuTkU as the source of that energy. In this video, the Youtuber Network Chuck goes through a demonstration of use cases for the CLI based interfaces for the various LLM-inference providers over the last year. I had messed around with these tools as they came out. They seemed neat, but I had already had this kind of experience when using GitHub copilot in VSCode. CLI tool-usage was one of the things it did, and it seemed adequate enough.

I was wrong. The experience of strictly using the terminal, and the CLI, gives you some purity in that way. It also stresses the idea, as Network Chuck did, that you can have the robot do anything you want, or anything you could otherwise do youself, straight from the commmand-line. This has come along at a time where the models are really, actually, getting much better. Leaning into the overall vibe of vibe-coding, I don’t really look at the code at this point. I just tell the robot to keep working on something until it builds and looks good in production. And you know what? Much of the time it’ll get intricate feature requests correct the first time. I should note that this is while working with the most advanced models. These are - Claude Opus 4.5, OpenAI’s GPT 5.2, and Gemini 3. Of these, Opus 4.5 is the best. There’s this persistence it has on iterating on completing a task. At times I’ll watch as it works for 15 minutes… just grinding and trying things until it arrives at a solution. It really makes you feel like more of a project manager, or supervising collegue that will discuss and brainstorm, and then agree on what work needs done, and then the robot just does it.

With these tools I started a research project with the intention of having the agents loop, searching on events related to Donald Trump. This eventually resulted in https://griftbook.info. But this post isn’t about that, as there is another tool I’d like to talk about.

Discovering Google’s Antigravity

Enter Google’s Antigravity. Antigravity is a fork of vscode by all appearences, and offers the interface I’ve gotten to used to over the years with Copilot. It does offer some capabilities that I hadn’t worked with before though. These are:

• It has an “Agent Manager” interface which let’s you organize agent interaction sessions nicely, and overall looks something like the “ChatGPT” style interface seen in that initial killer-app.
• Most importantly, it has an integration with Google Chrome where the agent will actually open up a browser tab and start clicking around. This is amazing for frontend development.

So, all that being said, last night I spent a few hours to create this blog. Here’s more of a breakdown of the steps involved.

My Development Workflow

My workflow right now is to start with a chatgpt conversation on the idea. I like chatgpt because it’s been increasingly turned towards my personal preferences as far as personality goes. It’s been building memories related to my own life and tastes, and that context can be helpful for orginating ideas. I’ll discuss the overall concept, the tools available to implement the concept, and overall just have a brainstorming session. Once I’ve figured out what I want to do, I’ll have chatgpt create a CHARTER document that I’ll then download and populate as the initial file in a new repository.

From there I’ll either use the CLI tools or Antigravity to begin iterating and implementing the project. In this case “blogbook” is a GitHub repo I’m keeping private at the moment (although probably no good reason to do that), which is deploying the blog to GitHub pages. Like with griftbook, I’ve setup a custom domain for this. This entails some clicking around in registrar web pages to configure DNS, and I used Antigravity for this. Sure, Antigravity is great for doing frontend development, but it can also just do anything else you can do with a web browser, and so I tasked it with enabling a custom github pages url. After logging into GitHub and Ionos it correctly went to each service and used the web interface to make the changes. Some of this might be possible through the gh cli, which I make a whole lot of use of, but unfortunately Ionos doesn’t have a CLI utility apart from the cloud infrastructure services component of it’s business (ionoscli).

Current Blog Features

Some more specifics of what I’ve set for this so far:

• I have a cli utility script that will scaffold a new blog post, fill in some of the initial meta data for the post, and then create a Pull Request on this new branch.
• There are a variety of LLM-based slash commands that can be run in the PR to do things like blog post image generation, proof-reading, and section organizing and header creation.
• After the basic infrastructure was built, I brainstormed with the agent on additional features that could be implemented across broad categories, I then had the agent use the gh cli to create issues for each idea to potentially implement in the future.

Ok, so that’s it really. Happy 2026, let me know what you think at mattuebel@mattuebel.com