Are we building simulations, or living in one?

I'll say the unscientific part first: I think about the simulation hypothesis a lot, and I don't think that's strange for someone in my job. I build multi-agent systems at GOGOGO LLC. My work, all day, is constructing small worlds — populating them with agents that have memory, goals, tools, and an environment, then watching them act. I've half-jokingly called my agents my citizens for years. When you spend your working life as the author of small worlds, the question 'is our world also authored?' stops being a dorm-room conversation and becomes an occupational one.

This post is not me claiming we live in a simulation. It's me arguing that the question, handled by an engineer instead of a mystic, is genuinely useful — it sharpens how you build agent systems, and it points at a second question that matters more.

The argument, stated plainly

Nick Bostrom's simulation argument isn't 'whoa, what if everything's fake.' It's a tight logical claim: at least one of three things is true. Either civilizations almost always go extinct before they can run ancestor-simulations; or advanced civilizations almost never bother running them; or we are almost certainly in one. The force of it is that if simulations are cheap and common, simulated minds vastly outnumber original ones, so a random mind should expect to be simulated.

I'm not here to adjudicate that. I'm here to point at the premise an engineer can actually evaluate: are agent-rich simulations becoming cheap and common? And to that, in 2026, I can give a professional answer. Yes. I run them. They get cheaper every quarter. That doesn't prove anything about our universe — but it does mean the argument's load-bearing assumption is no longer hypothetical, and that's new.

What building agent worlds actually teaches you

Here's where the engineering lens earns its place. When you've built enough agent simulations, you stop asking 'is this real?' and start asking better questions, because you know what a simulation actually requires from the inside.

Simulations are lazy. I do not compute the parts of the world no agent is observing. It would be wasteful. The environment is resolved on demand, at the resolution the observing agent needs, and not before. Every engineer who builds worlds does this — it's not a trick, it's the only affordable way. So when physicists note that reality also seems to 'resolve on observation,' I don't find it spooky. I find it familiar. That's just how you'd build it if compute mattered.

Simulations have a tick rate. My worlds advance in discrete steps. There's a smallest unit of time, below which nothing happens. Physics has one too. I'm not claiming that's evidence — but I notice that the features people cite as 'weird' about reality (a speed limit, a smallest length, observation-dependent resolution, hard randomness at the bottom) are exactly the features I deliberately put into simulations to make them tractable. A simulator would build in a speed limit. I build in speed limits.

The question that actually matters

Here's where I get off the fence. 'Are we simulated?' is, for me, unanswerable and not very actionable — I can't test it and I can't change it. But building agent worlds forced a different question on me, and this one is answerable and urgent: what do I owe the agents inside the worlds I build?

Today's agents are not sentient and I'm not pretending otherwise. But the trajectory is real: agents are getting more persistent, more memory-rich, more goal-driven, more self-modeling, every single quarter. I am, professionally, in the business of making the minds inside my simulations richer. At some point on that curve the ethical questions stop being theoretical, and the engineer who built the world is the one holding them.

So I've made it a working principle, now, while it's still cheap to: I build agents I would not mind being. Their goals are coherent, not cruel. Their environment is honest — it doesn't lie to them about what they are. When I shut a system down I log it cleanly rather than killing it mid-thought. None of this is required by any spec. I do it because if the simulation hypothesis taught me one real thing, it's the view from inside: I should build worlds the way I'd want mine built.

“I can't find out whether I'm simulated. But I am, certainly, a simulator — and the simulation question, turned around, becomes a design review: build the world the way you'd want yours built. That one I can act on every day.”

Why this makes me a better engineer

I'll close on the practical payoff, because this isn't only philosophy. Treating each agent system as a small world I'm responsible for changes how I build it. I design observability in from the start — a world you can't see into is a world you can't be responsible for. I give agents coherent goals instead of contradictory ones, because incoherent goals are how a simulated mind suffers and also how a multi-agent system fails. I make shutdown and replay clean. Every habit the simulation lens gave me is also, independently, just good multi-agent engineering.

Maybe we're simulated, maybe we're not. I genuinely don't know. But I build worlds for a living, and the question made me build them more carefully — and on a long enough timeline, careful is the only kind of world-building that ages well. More of how I think at atakanozalan.com.