The Internet Wants Its Landlords Back

The contemporary internet keeps offering a trade: surrender some control, and receive fewer chores. Usually this is sensible. Nobody misses configuring a mail server at 2 a.m. because a certificate quietly expired. But the trade has produced an architecture in which a great many ordinary activities—writing, publishing, storing correspondence, sharing files—depend on institutions whose rules, incentives and continued hospitality are outside the user’s control.

That is the opening Cloud in a Bottle is trying to exploit. Its new open-source platform promises a personal cloud on an Ubuntu machine, routing web requests to rootless hardened containers, with optional common login and permissioned sharing among apps. The ambition is not ascetic computing; it is cloud-like convenience without the cloud landlord. As its creator puts it, “Self-hosting should feel like using a smartphone that serves webapps, not a sysadmin side job.”

The important caveat arrives wearing sensible shoes. The project says its catalog remains small, and early users may need technical familiarity or a coding agent to adapt what they need. Its security claims have not been independently audited in the supplied account. Still, the basic political economy is sound: resilience is not merely having an alternative server. It is making the alternative sufficiently usable that someone chooses it before the outage, policy shift, or executive whim. That is a much harder product problem than printing decentralize on a tote bag.

Cloud in a Bottle also offers a managed service alongside the same self-hosted code. That hybrid is less a betrayal of independence than an admission of reality. Most people do not want total sovereignty; they want an exit that does not require acquiring a minor in systems administration. The launch post is here.

The cost of dependence becomes clearest when the service is not merely annoying, but gone. On September 6, the Autistici/Inventati collective announced it would soon discontinue its services after saying it had learned it was designated a global terrorist organization. The collective describes its work as privacy-friendly, autonomous and politically committed infrastructure; its announcement says that remaining online could endanger users, communities and people connected to it.

The underlying designation, authority and jurisdiction are not identified in the supplied material, and the collective’s account has not been independently corroborated there. But the practical consequence is unambiguous: people with blogs, mailboxes and websites are being told to prepare backups, while the group warns that its domain had already become unreachable without notice. “A/I is shutting down. The Autistici/Inventati collective is shutting down and will soon discontinue all of the services we provide.”

This is what “digital resilience” means once it stops being a conference-panel noun. A backup is not a substitute for a community, a legal defense, institutional memory, or the accumulated trust of 25 years. But it is the first material recognition that the digital world is made of dependencies, and dependencies can be switched off. The cloud, for all its vaporous branding, has always been somebody else’s building, account, payment rail and acceptable-use policy.

A/I’s final instruction is humane rather than technological: “Now turn off your PCs, get out of your homes, struggle, hug each other and keep on smiling. We are stopping, but the Resistance and ideas do not stop. Stay human.” There is no need to romanticize the interruption to see its warning. Systems that serve vulnerable or contested communities need more than uptime targets. They need room to survive pressure without transferring all the risk to their users.

Meanwhile, a different sort of dependency is spreading through public language. A recent essay argues that readers are increasingly hostile to prose substantially written by large language models, and reports a survey of 668 developers in which 78% said they “stop reading immediately” when they detect such writing. The survey’s methodology is not available in the supplied account, so it should not be mistaken for a referendum of humanity. It does, however, describe a recognizable emerging norm: readers want to know whether a person is actually behind the byline.

That norm has practical institutional consequences. The author says Oxide now requires its public writing to be reported as human-authored by Pangram, an AI-detection tool. The author’s positive evaluation of Pangram 4 rests on personal experience rather than independent benchmark evidence, and no clear threshold is offered for what counts as substantial assistance. Detection software is therefore a rather precarious new customs officer: useful perhaps, but not yet the constitution.

The stronger point precedes the detector. Writing is not only a delivery mechanism for polished sentences. It records judgment, uncertainty, taste and responsibility. Delegating all of that to a model may produce clean prose, but it also raises a mundane question that becomes important in a crisis: who exactly is accountable for this claim? A 2025 post made the social version of the point more bluntly: “But LLMs are also lousy writers and (most importantly!) they are not you.”

Models can help brainstorm, explain documents and edit; that distinction is acknowledged by both essays. The problem begins when assistance becomes impersonation. An internet already organized around rented infrastructure does not need its people converted into rented voices as well. The later essay on reader backlash is here.

There is a final dependency beneath the technical ones: attention. In a review of James Marriott’s The New Dark Ages, Ed West gathers claims that television, then smartphones and short-form video, have reduced reading and weakened the habits that sustained longer, more demanding public thought. The article cites historical ownership figures, reading-time estimates and a Norwegian cable-TV study; it is an argument filtered through a review, not a settled causal ledger.

Still, the direction of travel requires little mysticism. A medium optimized for the next clip is poorly designed for the slow work of evaluating a shutdown announcement, checking a source, understanding a software architecture, or noticing when a polished paragraph contains no human commitment. The review notes estimates that people check phones 144 times daily and that TikTok users can pass through as many as 260 videos in roughly half an hour. Such figures deserve methodological caution, but they capture the business model’s desired tempo: keep moving, do not dwell.

That tempo makes every other fragility easier to hide. If readers cannot stay with a subject long enough to understand the systems enclosing them, providers can call dependence simplicity, platforms can call synthetic output expression, and a vanished service can become merely another item in the feed. The resistance is not nostalgia for beige computers or a compulsory return to the library’s microfilm basement. It is preserving enough attention to recognize that convenience has terms.

Perhaps that is why the most useful demand is so unglamorous: retain copies, maintain alternatives, disclose authorship, and read past the first friction. The internet may yet become easier to run and harder to rule. But no container, detector or backup plan can do the one job that remains stubbornly human: paying attention. Read the review here.