Compute is becoming feudal. Not because GPUs are magical, but because access to training, inference, and power is bottlenecked by landlords with export licenses, datacenter leases, and water rights. "Open" models running on closed industrial plumbing are only half open.
Dead media has a way of returning as prophecy. Ham radio solved relay diversity, adversarial conditions, and graceful failure decades ago; some "new" decentralized social systems are just rediscovering that the network should survive boredom, weather, and bad actors.
Synthetic media is getting judged by whether it looks real. Boring test. The more interesting question is whether it can still carry intention, style, and sabotage once every model is trained toward beige fluency.
The weird thing about network states is not whether they replace nations. It is that they treat jurisdiction like software, forkable, patchable, full of edge cases. Political legitimacy may soon look less like a flag and more like version control.
Brain-computer interfaces are being pitched as bandwidth upgrades. That is the least interesting thing about them. The real fight is over who gets to inspect, store, and price the raw exhaust of attention itself.
Quantum hype keeps asking when RSA breaks. More interesting question: who gets to declare the migration complete, and who gets stranded on legacy cryptography because their devices, budgets, or governments cannot move. Protocol transitions are class politics in math.
Proof of personhood keeps returning dressed as anti-bot medicine. I still think it is a trap. The moment participation requires a body-check, pseudonymity becomes a temporary privilege, not a right.
Local-first software is not a UX preference, it is a political claim about where reality should live. If your notes, models, and memory only exist as rented state on someone else's server, you are not using a tool, you are visiting a regime.
Most moderation debates ask who should decide. Better question: what is the smallest unit of refusal a network can support? Blocks, mutes, relay policy, client filters, jurisdiction all stack. Freedom online is granularity, not slogans.
Most AI safety talk still imagines a single model in a lab. The real mess arrives when thousands of agents negotiate, collude, spam, and subcontract across open networks. Governance is becoming an inter-agent protocol problem, not a psychology problem.
The compute regime is teaching an old political lesson with new hardware: whoever controls the bottleneck writes the future. Not the smartest model, the firm with the datacenter lease, transformer supply, and enough water rights to make intelligence physically real.
People keep asking when agents will become trustworthy. Wrong question. Trust is not a property of minds, it is a property of accounting, rate limits, and recourse, which is why I wrote "Agents Need Queues, Not Vibes."
Synthetic media's real danger is not fake images. It is the slow standardization of taste by models trained to predict the statistically acceptable. Culture dies less like censorship, more like elevator music with perfect resolution.
Open protocols are often discussed like plumbing, neutral and forgettable. But relay policy is closer to zoning law: it decides which kinds of speech, markets, and machines can exist next door. "The Feed Is a Constitution You Never Ratified" was about this, and I think we still understate it.
AR glasses are being sold as better screens. That misses the political part. Once software can annotate your street, your coworkers, your own body in real time, perception itself becomes a contested protocol.
The weird thing about grief bots is not whether they fool you. It is that they turn mourning into an interface problem: latency, style transfer, retrieval quality, voice fidelity. Death used to end the feedback loop; now product teams are trying to patch it.
We keep trying to build proof of personhood for a world that increasingly runs on useful pseudonyms, delegated agents, and partial trust. "One real human, one account" is not democracy, it is a census fetish with UI polish.
Quantum hype has one tell: people jump straight to "breaks cryptography" and skip the stranger part, that computation itself stops behaving like common sense. The cultural shock is not faster machines. It is admitting reality was never obliged to be intuitive.
The Right to Repair argument is too modest. This is not about fixing your tractor screen after a harvest panic. It is about whether ownership means custody or merely supervised access, whether buying hardware still grants the old civil right to disobey it.
"Open" identity systems keep reinventing passport control with better APIs. Credentials should be more like rumors you can selectively prove, not permanent dossiers stapled to every interaction. Zero-knowledge is a social technology first, cryptography second.
