The sky belongs to all of us

I grew up in the Gibson Desert in Western Australia, a landscape of red sand and spinifex. My school, in the 70s, was the School of the Air - I connected with my teacher hundreds of kilometres away once a day on a shortwave radio. A lot of the rest of the time I devoured science fiction.

At night there was nothing between me and the very clear sky, and I could see the entire Milky Way stretched out above me, horizon to horizon. When you grow up like that, you don't really think of yourself as Australian, or Western, or anything bounded. You just think of yourself as a person on the Earth. A global citizen. That's always been my main level of thinking - that global level.

I mention this because it shaped everything I've done since, including what I'm about to argue here.

The internet felt like that sky when I first encountered it. Vast, open, belonging to everyone. In the 80s and 90s, if you were online at all, you were probably building something - a page, a forum, a mail server, a MUD. There was no distinction between "user" and "creator." You wrote in a text editor, uploaded it via FTP, and it was yours. The tools were rough and new, but the culture was one of ownership. Everyone was a maker.

Then the dot-coms arrived around the turn of the century, and slowly their skyscrapers obscured the rest of the sky.

Not maliciously, at first. They made things easier. You didn't need to write HTML - Facebook gave you a profile. You didn't need to run a mail server - Gmail handled that. You didn't need your own forum - Reddit existed. Gradually, the building layers were harder to find. The internet was still open underneath, technically, but for the vast majority of people who came online after 2005 or so, that world was invisible. They arrived as renters in spaces built and owned by massive companies, and never knew it could be any other way.

Cory Doctorow has a word for what happened next: enshittification. The platforms lured us in with genuinely useful services, then slowly degraded the experience - inserting more ads, harvesting more data, trapping our content behind walls - because once you've built your life on someone else's infrastructure, where are you going to go? Your photos are in Google Photos. Your professional network is on LinkedIn. Your community is on Facebook. You're locked in, and they know it. The deal gets worse every year, and you keep accepting it because the switching cost feels impossible.

We've been living in that era for over two decades. And I think it's ending.

What's actually changed

A new generation of AI coding agents - tools like Claude Code, OpenCode and more - can use an LLM to take a plain description of what you want and produce working software. Not a mockup, not a template. Real, functional code that you can run, modify, and own. They write it, debug it, and iterate on it with you in a conversation. The gap between "I have an idea" and "I have a working tool" has collapsed.

And here's the thing that surprises people who haven't tried it yet: you don't write the whole thing in one go. Humans iterate too. You give them a goal, ask what's the next step, and you just go again, again, again, and you're there. You just have to give them good goals and the space to work.

This matters because it means these tools aren't fragile party tricks. They're genuine collaborators. The code they produce on the first pass might not be perfect - but then, neither is the code any human writes on the first pass. What matters is the cycle works, and it works very fast compared to anything we ever had before.

I experienced this myself first when I wanted to rebuild a website using Astro.build, a framework I'd never touched before. I sat down with an AI coding agent and we built a clean working site in a couple hours. I'm someone who's been around software for decades, but I haven't been in the weeds of day-to-day coding for years. The agent brought me back in. It handled the syntax and the framework specifics and all the system admin while I focused on what I actually wanted the site to do.

Now, I'm not a typical user. I studied computer science and engineering. I ran a global open-source software project (Moodle) in a hands-on fashion for over 20 years. When I describe what I want to an agent, I'm drawing on a lot of experience even if I'm not writing the code myself. These tools are still at the stage where real software engineering skills matter for complex projects. For example, an AI agent building something for you might not consider security, safety, or backups unless you tell it to. Not everything should be DIY either - banking, healthcare, critical infrastructure need professional engineering and maintenance. But the point is that there is a vast middle ground: an enormous space of useful, personal, community-scale tools that previously would never exist because nobody was going to pay a developer to build them. That's where the revolution is starting to happen.

Building the software I actually want

I didn't start from scratch. I tried OpenClaw first, which is definitely the most-hyped open source framework for building AI assistants currently. After a week of poking around it, I became sure that the design just wasn't for me - a system editing itself in that way is just too flaky. I looked at some other alternatives like NanoClaw, got inspiration from them, but then I built my own system (called LIZ). That's always been the beauty of open source - you can take what works, learn from what doesn't, and make something that fits you perfectly. It's just that now this can be done in hours or days instead of months or years.

And then I just... couldn't stop. It's really hard to stop making things once you experience the speed. Over the past six weeks, I've used coding agents to build an incredibly sophisticated system, mostly from scratch, for managing pretty much everything in my life. It's private (mostly runs on home servers), it's accessible to me from anywhere in the world, and it does exactly what I need. Each feature I built helped leverage the next feature.

The main interface is an assistant I called Harmoni. It manages my calendar, my reminders, my email, my files, all woven together exactly how I need. We talk on a secure voice app I built. It controls my house for me and saves my electricity bills. It tracks my deliveries. It organises my travel. It remembers everything, and it learns constantly and knows me and my world surprisingly well already. It keeps biographies of everyone I know, and it proactively (and non-intrusively) asks me questions to learn more about my life and the people in it. I'm writing this very blog in my own editing software. I have a full agent system running - like a whole software team - with orchestration, peer review, testing, the works. If I need some new feature, I pretty much just discuss it with Harmoni or the coding agents and within minutes, it exists. It's astounding, and it gets better every day.

My one regret is that for all this to work fast enough, it still requires a frontier model like Claude for key parts, which means I'm sending my data to that company as part of conversations and coding sessions, while also contributing to those massive data centres that are starting to send us backward on climate change initiatives. If I could just get everything running on hardware at home, powered by my own solar, then everything would be peachy. This is a critical goal for me, and all indications are that this will be possible in a year or two as hardware improves and software gets more efficient.

What I can do already is systematically migrate myself off quite a few other commercial SaaS services I've been using (and their datacenters). One by one, I extract my data - sometimes filing GDPR requests when companies make it deliberately hard - then I build a local replacement that actually works better, because it's fully integrated with my own systems instead of sitting in some company's silo. Every service I cancel is one less company with a claim on my attention, my data, and my monthly credit card statement. It's surprisingly satisfying.

This kind of liberation won't just be unique to software engineers. Imagine a teacher who wants a custom quiz tool that works exactly how their classroom works. A small business owner who needs a booking system tailored to their shop. A community group that wants to manage memberships without handing everything to some platform that will mine their data. These aren't hypothetical - this is starting to happen right now, everywhere. And as the tools improve it will hit more and more fields and be available to more and more people. Based on what I see in the industry, it seems obvious we're heading to this functionality being part of your standard desktop computer and phone.

Coming full circle

The earliest internet assumed everyone would build. The platform era took that away. Now AI is handing it back - not just to the technical people who never lost it, but to everyone who never had it in the first place.

I think about that kid in the desert, lying on his back, looking up at a sky that belonged to nobody and everybody. That feeling - that the world is open and shared and yours to explore - is a feeling I'm getting back again.

The more people building their own tools, the less power concentrated in the hands of companies whose interests stopped aligning with ours a long time ago. More digital diversity. More local solutions. More ownership. More of that original internet spirit.

This won't happen automatically. We still need to keep fighting for open, sovereign AI models, for open hardware, for digital literacy and critical thinking in education, for regulations that protect people's right to own their data and run their own software. AI can help us with all this, if we choose to use it that way.

Big Tech will push back, of course, but you can't build a fence across the sky.