Why I Keep a Game With No Players Running Anyway

<p>Of everything I work on, GAIA is the one I'd have the hardest time justifying on a spreadsheet. It's a personal world-evolution game — a terraforming AI and her council of sub-intelligences, running on local models, playing itself more or less whether I'm watching or not. No customers. No revenue model. Nobody asked for it.</p><p>So why does it sit on the same machine, in the same week, as AIREP and Gripswich OS — the two things that actually need to make money for Intelli AI to survive?</p><p>My honest answer is that GAIA is where I get to find out what AI can do when nobody's paying for the outcome. Client work has constraints that are good for the client and bad for exploration: a gym needs its booking system to work today, not to discover something interesting about agent coordination. AIREP needs to be boring and reliable because it's someone else's business running on it. Neither of those systems can afford for me to try something half-formed.</p><p>GAIA can. It's passive by design — I'm not babysitting it, I'm occasionally checking in on what the council of sub-intelligences decided to do with the world this time. That passivity is the point. It's a long-running experiment in letting local models make consequential decisions without a human in the loop correcting every step, which is a very different problem from building a coding agent that should ask before it does anything risky.</p><p>And that's not unrelated to the work I actually get paid for. Nova's whole architecture — the idea that I want thirty-four narrow agents instead of one generalist, the self-improvement loop I'm trying to build so Nova can review and refactor her own code — all of that needs somewhere low-stakes to get wrong first. GAIA is where an agent can make a genuinely bad call about terraforming a continent and the consequence is a funny story, not a client escalation. I learn more from watching that happen than I would from reading about it.</p><p>There's also a simpler reason, which is that I think it's good for me to keep at least one project in my stack with zero commercial pressure. Everything else I build gets evaluated, eventually, by whether someone will sign a licence for it. That's the right lens for a software business — I'm not precious about needing art-for-art's-sake justifications for Intelli AI's time. But if every single thing I touch has to clear that bar, I stop being curious in ways that don't immediately pay off, and curiosity is most of where the actual differentiating ideas for AIREP and Nova have come from.</p><p>Local models make this practical in a way it wouldn't be otherwise. Running a passive, always-on simulation against a paid API would be a bad trade — cost scaling with a thing that has no revenue attached to it is exactly the kind of expense a bootstrapped, no-outside-money business should avoid. Qwen running on a box at home doesn't care how long GAIA takes to get interesting. That's the same local-model logic I apply to Nova's home-box tier, just pointed at a different problem.</p><p>I don't think every developer needs a pet project like this, and I'm not going to pretend GAIA is secretly a strategic masterstroke. It isn't on the roadmap. It doesn't have a roadmap. But I think there's real value in keeping one system around that exists purely to let me watch AI make decisions with no business constraint attached, because that's the clearest signal I get about what the technology is actually capable of, separate from what a client is willing to pay for this quarter.</p>