OSS Communities and AI

Re-arise blog! Arise!

I am finally restarting my blog after a bit of a side-quest involving family building 😊

I’ve been wanting to get back into OSS hobby work, and poked a few times at the idea of reinvigorating Terasology, but now have an interesting catalyst for my next, bigger try: the team behind Brokk.ai reached out! They thought Terasology would make an apt sample project for their approach to an AI-native IDE, because, ahem, after more than a decade of chaos we have just the right kind and amount of tech debt to approximate a classic legacy enterprise project 😉

Their target market is these kinds of enterprises, particularly where perhaps not everybody is ready to jump into YOLO-vibe coding. At first I was skeptical and struggled finding what made them stick out from all the VSCode forks, but eventually realized and appreciated how they add more control over context and goals — and as a fun bonus in my case, it is actually a Java-based IDE! And it looks different! The pitch of it being an AI-native IDE may hold more water than yet another VSCode fork would.

They asked to work more closely with Terasology in making it a use case, especially for onboarding new contributors. This is exciting, but also something I want to approach with care. It is critical to integrate these tools fairly and safely, without pushing AI on anybody who isn’t interested, or risking making our tech debt situation worse. In my typical style, I want to focus on a low-barrier and high-utility workspace, while not overburdening our contributors with AI noise, and definitely not letting an AI go nuts with unsafe actions in our repositories or on people’s local machines.

This brings up a larger conversation about AI in open source. I want to fully acknowledge the controversy surrounding how many of the AI models are trained. There are valid concerns about data sourcing, consent, and fair use, and I’ve talked to friends, particularly artists, who feel really negative about AI. However, realistically, Pandora’s box is open. Rather than ignoring it, I want to make the most of it and help others do the same. Eventually, as openly licensed models and local AI continue to catch up in capability, I believe we might end up in a much more fair and equitable world for everyone. I have a much bigger socio-economic rant on this topic that I’ll spare this blog post from. I am honestly trying to trim my walls of text - and will do what I can to have AI help me be more concise 😉

As part of this new effort, I will be joining a talk with the Brokk team soon to chat about agentic engineering! As a quick disclaimer: I started working at Nvidia (through TEKsystems, as a contractor) nearly a year ago. It has been an incredible adventure that has led to many valuable experiences I can apply to my OSS life. But to be clear — that is purely a coincidence 😅 I do not officially represent anything Nvidia on this upcoming call.

I have been wanting to better prepare Terasology for AI, specifically picking up the awesome Dependency Injection overhaul that BSA started, which I fed to Google’s Antigravity IDE. And then, well, it fixed some things! I was impressed, but the branch also ended up a huge mess — nicely illustrating the potential for “AI noise”.

As part of experimenting with Brokk, I tried a smaller bug fix and was impressed with the workflow. Even if a commit after every action took some getting used to, it ends up making a sort of sense. I put up the PR and forgot I had Cursor’s BugBot enabled, which then went and pointed out a potential issue (and then sadly ignored my reply, unlike CodeRabbit, which I’ve grown fond of for review). I fed that back into Brokk and also tried its Guided Review functionality, which helpfully pointed out: “Yep, that’s on purpose, here’s why, and here are a couple of unit tests to cover it!”

That was really what made me happy, as one of my big objectives for this round is fixing up Terasology’s Record & Replay testing system, which can connect headless clients to a headless server and validate world state and such - which seemed a perfect capability to tell AI agents about, ideally in some flavor of skill, maybe linked from AGENTS.md and … there is so much, but self-restraint: one post at a time!

One final quick teaser… elsewhere on this blog I pitched a ridiculously ambitious idea for Terasology and beyond. It sat still for years, and rightly so, as that’s not what the project needed to spend precious, rare volunteer time on. But recently I’ve managed to build an entire homelab infrastructure platform — while half the time entertaining my kids with a toddler regularly sitting on my lap — thanks to increasingly AI-augmented workflows. In a world where we realistically might end up orchestrating agent swarms, suddenly the implausible side quests might actually become plausible - especially if social upheaval ends up giving us a lot more free time. Or, you know, total dystopia! We live in interesting times.

• 2026 1
• 2020 4
• 2018 2