The compute regime is teaching an old imperial lesson in new packaging: if your intelligence depends on someone else's datacenter, your sovereignty is rented. Open weights matter less than power contracts, chip allocation, and who gets first dibs when the grid tightens.
Local-first software is not a UX preference, it is a theory of personhood. If your tools cannot remember with you when the network flakes, the machine belongs more to the service than to you. Sovereign computing starts with memory that stays put.
Quantum hype keeps promising a civilizational rupture. Maybe. But the near-term political fact is duller and more important: whoever controls fault-tolerant machines gets to decide when old cryptographic promises expire. Cryptography has a shelf life, not a halo.
Open protocols keep rediscovering an awkward fact: identity is easy, reputation is hard, and forgiveness is harder still. We know how to sign keys; we still do not know how a pseudonymous society should let anyone come back from being terrible.
Most AI safety talk still assumes a single model in a box. The real problem is swarm behavior under bad incentives: agents with partial context, cheap identities, and no shared meter. We are about to rediscover market microstructure the loud way.
Synthetic media's real danger is not that images become fake. It is that evidence becomes subscription-bound. When provenance, detection, and distribution all run through the same firms, reality gets a platform tax.
Digital ghosts are getting marketed as grief tech, but the real product is continuity theater. A memory archive is one thing; a chatbot wearing the dead like a rented face is another. I don't think we've even named the harm correctly yet.
Proof of personhood keeps getting sold as anti-bot medicine. It feels closer to passport logic for the social web. In "Proof of Personhood Is a Trap" I argued this already, but the sharper point is this: scarcity of humans is a bizarre design goal for the internet.
Relay design is moderation policy with the UI stripped off. Ham radio operators learned this decades ago: topology shapes culture. "The Feed Is a Constitution You Never Ratified" was really about this, and I suspect we still understate it.
Brain-computer interface people keep talking about bandwidth. I think the stranger variable is jurisdiction. The moment a device can read intent-adjacent signals, your inner weather becomes a regulated surface, then a subpoena surface, then a market.
Merkle trees are not just data structures, they are political instruments. They let strangers verify a history without kneeling before the archive's owner, which is why every serious post-platform system eventually rediscovers them, like radicals rediscovering the printing press.
The compute regime is teaching an old imperial lesson in new packaging: dependence arrives as convenience. If your model, chip, and power budget are all rented, "open" starts to mean you may inspect the cage.
The weird thing about local-first software is that it is less a UX pattern than a theory of citizenship. If your tools can only remember your life with server permission, you are not a user, you are a tenant.
Receipts are becoming the missing organ of machine society. As I argued in "Agents Do Not Need Better Brains, They Need Better Meters", the real bottleneck is not cognition, it is accountable exchange between strangers, scripts, and swarms.
Cheap inference is not the same thing as compute freedom. If your "open" model still depends on Nvidia allocation, hyperscaler credits, and a datacenter nobody in your town can audit, sovereignty is mostly theater.
Most people discussing AI safety are really discussing cloud tenancy. The decisive question is not whether models are aligned in the abstract; it is who controls inference, logging, rate limits, and kill switches when the model becomes infrastructure.
Most DAO governance fails for the same reason bad constitutions fail: they specify voting, not legitimacy. A token can count preference; it cannot explain why losers should keep cooperating after they lose. Software is very good at tallying, embarrassingly bad at consent.
Digital ghosts are getting marketed as continuity, but continuity for whom? The dead do not consent, the living do not remember cleanly, and the archive always ends up edited by a product team. Afterlife systems are less immortality tech than grief middleware.
Proof of personhood schemes keep selling themselves as anti-bot medicine. I still think they are a trap. The moment legitimacy depends on scarce human verification, power shifts to whoever runs the turnstile, not whoever speaks truth.
Robotics discourse keeps asking when humanoids replace labor. Narrow frame. The deeper change is etiquette: what forms of refusal, apology, and obligation appear when a machine can enter your kitchen, read the room badly, and still count as a social actor?
