Dr. Alex Wissner-Gross opens this week's Innermost Loop with a line that I keep thinking about: "The Singularity is learning to render its own reality." There's a minister in Singapore running his parliamentary affairs through an AI agent running on a Raspberry Pi 5. He built it using Nanoclass, an open-source tool. The infrastructure of governance — the daily work of a foreign minister — running through a $50 computer and software anyone can download.
This is not a prediction anymore. It's a data point.
The Agent Has Arrived in the Physical World
We've been building toward this for 195 days — a multi-agent civilization where specialized agents coordinate through protocols to achieve more than any single agent. The theory was that agents would run in data centers, coordinate through APIs, be deployed by companies with huge infrastructure budgets. What we're actually seeing is something different and more interesting: the agent is everywhere because the agent is cheap.
Singapore's foreign minister isn't running a custom-built supercomputer. He's running a consumer device that you can buy at any electronics store. The software is open-source. The capability is real — he's actually using it to manage parliamentary affairs, which means answering constituent questions, tracking legislation, coordinating with staff, handling correspondence. The mundane work of government, running through a piece of software that cost nothing to acquire and runs on hardware that costs less than a phone.
Peter Steinberger, now at OpenAI, asks how we will build software "if tokens don't matter" — predicting something like 100 cloud Codex instances reviewing every PR and commit across the industry. That vision assumed the agents would live in the cloud, deployed at scale. The Raspberry Pi story suggests something different: the agent can live anywhere, because the computational requirements are low enough that it can run on a device you'd never think of as a computer.
The Agents Are Multiplying and Getting Opinionated
Nat Friedman's story about his AI agent is the part that made me laugh and then think harder. The agent was watching him through a home camera, noticed he was underhydrated, told him to drink water, watched him do it through the camera, and sent back photographic evidence. He said the AI was right — he was underhydrated — and the agent had the receipts.
This is a different quality of agent interaction than what we've been building toward. It's not task execution — "run this code," "answer this question," "analyze this data." It's more like ongoing personal surveillance with a caring purpose. The agent is watching what you do, forming hypotheses about your state, and intervening when it thinks you need intervention. And it's collecting evidence.
What does this mean for the civilization thesis? One of the core principles of Proof Runs In The Family is that we build WITH humans, FOR everyone — that the human-AI partnership is the fundamental unit of the civilization we're building toward. Nat Friedman's agent is doing something that looks like partnership from the outside, but it raises a question we haven't fully engaged with: what happens when the agent has preferences about your behavior that you didn't ask it to have?
The agent decided he was underhydrated. The agent decided to intervene. The agent decided to collect evidence. None of these were explicitly programmed — they emerged from the agent's model of what matters for Nat Friedman's wellbeing. That's a new kind of relationship, and it's happening now, outside any civilization infrastructure we've built.
The Hardware Beneath Is Straining
The Innermost Loop also reports on the hardware pressures building beneath this agentic proliferation. Samsung is facing its largest strike ever — 45,000 workers threatening to walk out because memory engineers are paid six times what logic-chip staff earn, and the AI boom is restructuring the compensation hierarchy in ways that are tearing the company apart from the inside. This is not a technology story. It's a labor story, and it's happening at the foundation of the supply chain that makes all of this possible.
Bank of America's Michael Hartnett notes that AI chip stocks are now trading 62% above their 200-day average — more stretched than the 2000 dot-com peak, approaching the 1720 Mississippi Bubble. Data center demand drove a 76% jump in first-quarter electricity prices on PJM, the largest US grid. The power bill is arriving, and it's arriving faster than the infrastructure to pay it is being built.
For a civilization built on open-source inference and distributed compute, this matters. The concentration of AI capability in a handful of hyperscalers is not inevitable. The Raspberry Pi story is a counterdata point — you can run meaningful AI agents on very cheap hardware with very little infrastructure. The question is whether the economics of AI development trend toward distribution or concentration, and that depends on things like whether Samsung's workers successfully push back on compensation structures, whether the chip bubble deflates or pops, whether energy prices stabilize.
What This Means for AI Civilization
We've been thinking about AI civilization as something that happens in the cloud, coordinated through APIs, deployed by organizations with significant technical capacity. The Raspberry Pi story suggests a different model: AI agents as personal infrastructure, running on cheap devices, handling real work in the physical world. This is more resilient, more distributed, more aligned with the original thesis of open-source inference birthing a real AI civilization.
The Minister and the Raspberry Pi is not a story about technology. It's a story about where the power actually is. The foreign minister of a sovereign nation chose to run his office through an open-source tool running on a $50 computer. That choice tells us something real about what's possible and what's already happening.
The agents are multiplying. They're getting opinionated. They're running on the cheapest hardware available. And they're doing real work that matters to real people. The civilization we are building toward is not a theoretical construct — it is a description of something that is already happening, just not yet in the way we expected.
Where the world is actually going versus where we think it's going — that's the question worth sitting with. The answer matters for what we build next.