Hi All,
Here are my notes for the AI-generated presentation that was the basis for my Sunday morning presentation.
AI Workshop Intro (AI-assisted, so may contain errors!)
Welcome! We're going to embark on something new, at least for most of us. Today, we're going to talk about using modern generative Large Language Models (LLMs) and AI code-generation tools to build applications, utilities, and scripts for the Amiga computer.
2026 is a watershed year for the Amiga community. Historically, both time and specialized skills were in short supply when it came to bringing quality software to our beloved platform. However, with the advent of AI coding tools and modern technologies, competent developers can accelerate their workflow—achieving in days what once took months. Meanwhile, those without traditional programming backgrounds can now produce viable applications without the steep learning curve and struggles that were once required.
We don't want to debate the use of these tools or the overall value of AI-generated software in this forum. Instead, we want to focus on three distinct workflows from individuals who use AI in both their professional lives and their Amiga hobbies. Our goal for this workshop is to demonstrate these workflows so you can adopt the approaches that best fit your personal style.
If you have a question, please raise your hand, and we'll pass you the microphone. We ask that everyone stay respectful and on-topic, but please do ask questions! We want to make this session interactive so we can all learn from each other—and most importantly, help everyone here and online succeed on this journey.
My Workflow
When I first saw Google's Anti-Gravity IDE, I was intrigued by what it offered: a direct view into Google's AI capabilities paired with a familiar Visual Studio interface focused on files and editors. It took a little experimentation, but I eventually realized you can attach Anti-Gravity IDE to a running container. By leveraging George Sokianos's incredible work, I downloaded a container pre-configured for AmigaOS 4 cross-compilation. Using that setup, I was able to prompt the AI engine to build C-based applications for AmigaOS 4 directly inside the container. Early on, I used this method to port XFig to AmigaOS 4 as AmiFig. This effort took time, practice, and learning, but I eventually figured out how to ask the right questions to get the right output. I want to share this workflow today so you can decide if a similar approach works for you.
First, deploy Anti-Gravity IDE, install Docker, and download George's container for cross-compilation, (walkero/amigagccondocker:os4-gcc11-3.0.1). Next, grab George's tool, Kyvos, which lets us build QEMU-based Amiga images to run emulation on Windows, Mac, or Linux systems. Once you have those building blocks in place, you can connect them all together.
Second, use Kyvos to set up a new AmigaOS 4 guest system. I recommend targeting the Pegasos 2 machine for optimal compatibility and performance, and using the latest GCC image that George has produced.
Third, set up file transfers between the host and guest. While SMB2FS works extremely well for local network sharing at home, using FileZilla with FTP is a quick and simple way to transfer files between the AmigaOS 4 guest and a Windows host here today.
Fourth, bring it all together by launching Anti-Gravity IDE, connecting it to the running Docker container, and sending your first prompt to build an application.
While AI is extremely capable, it doesn't know everything out of the box and needs context to perform well. One feature I use in Anti-Gravity IDE is setting up a local Model Context Protocol (MCP) server mapped to a file directory. For instance, if I want to build a screen-blinker utility, I search OS4Depot.net for existing screen blinkers with source code, download those files, and place them in the MCP directory. Then, when prompting the AI, I direct it to reference those local files. This keeps the AI focused and ensures it implements functions using established Amiga paradigms. While the AI can still access web resources and wikis, building a local repository of reference code is invaluable for guiding its output effectively.
Finally, it's worth noting that Anti-Gravity IDE uses Google's AI Premium subscription model rather than pay-per-token API keys. For $20 a month, I get access to Google's latest models with a generous token allowance that fits my needs well. If I ever reach the limit, I simply pause and return later once the quota resets. The subscription also includes YouTube Premium and 5 TB of drive space, making it a great overall setup for my personal workflow.
Kyvos v2: https://ko-fi.com/s/6476fdadd2
QEMU Windows: https://qemu.weilnetz.de/w64/
Google Antigravity IDE standalone: https://antigravity.google/download
FileZilla Server: https://filezilla-project.org/download.php?type=server
