This page was written with AI assistance.

Five parts of diagnosis deserve one part of grip. The wave is faster than your gut, the money behind it is nation-state scale and wobbling, the subscription economy has been quietly buying up the floors of your pyramid, the attacks now run at machine speed, and the worst case runs through us, not around us. Here is the good news the first five parts earned: the way out is concrete, cheap, and already on the shelf.

The story so far: five parts of hard news, work repriced, a trillion-dollar bubble, ownership gone, machine-speed attacks, and a capability we cannot un-invent. Now the turn.

Every fear in this series has the same answer

Let me be as clear as I have been gloomy. I am not an AI doomer. I am an optimist who took five parts to earn the right to say so. AI is the most empowering tool handed to ordinary people since the web, and I think it is going to make life radically better, richer, healthier, more creative, for people who would never have afforded any of that before. The catch is a single word, and the whole series has been circling it: control.

Look back at the five worries and notice they collapse into one solution.

• Your work (part one). You are not replaced, you are amplified, once the model works for you instead of the other way around. Own the tool, keep the judgement, and the substitution that erased the horse becomes the truck the driver learned to drive.
• The bubble (part two). You do not have to bet on the trillion-dollar giants or fear their crash. The one honest thing they produced, the intelligence, is turning into a commodity you can buy once and keep. Let the bubble be their problem, not yours.
• Ownership (part three). The intelligence is the rare asset you can still buy outright, on your own hardware, and put back on the compounding side of the ledger. It is the on-ramp back onto owning that the subscription economy tried to close.
• Security (part four). A model you run yourself, can see, log and fence is a model you can actually secure. Safety is not a property of the vendor; it is a property of who holds the keys.
• The singularity (part five). The only answer to one bad actor with a copy is a million good actors with their own copies and their own guardrails. Concentration is the dangerous path; distribution is the safe one.

One word answers all five, and it is not a policy or a summit. It is control, taken personally.

The turn

AI will save us, if we take control. Not us as a nation. Each of us, for ourselves.

I want to be precise about that last line, because it is the heart of it. The rescue here is not going to be handed down by a European champion or a national programme; parts two and three are the story of what happens while you wait for one. It is retail, not wholesale. It is the nurse, the developer, the civil servant, the small vendor, each putting a model on their own machine, pointed at their own data, working their own ideas. A million people taking their own control is a stronger, safer, freer outcome than any one institution taking it on their behalf. That is not a consolation prize. It is the better ending, and it is the one actually on offer.

Intelligence becomes something you can own again

Everything in the money chapter points one direction: intelligence is commoditising. Open weights sit three points behind the frontier, and the memory footprint of a frontier-class model shrank from 350 gigabytes to 24 in eighteen months, which is the size of a consumer graphics card. Translate that out of benchmark language: a premium AI is becoming a thing you buy once, plug into your own PC, and never pay for again. Not a subscription. A possession, with a power cord.

The arithmetic is almost embarrassing.

This is not a thought experiment for us; it is the thing we build. At Conduction the owned-intelligence half is Hermiq, an AI runner that executes open-weight models on your own hardware, inside your own audit trail, with the guardrails and egress fence set by you rather than a vendor. The owned-data half is OpenRegister, the register and object layer that keeps your case files, patient letters and citizen records under your control, with the access rules, versioning and audit that make it safe to point a model at them. The pairing is the whole argument of this series in two apps: the model you own, working the data you own, on the hardware you own. Neither is a subscription you cannot cancel.

It was never the model that fires progress

Here is the thing the valuation spreadsheets keep missing. A model, local or rented, produces nothing on its own. It sits there, brilliant and inert, until a human with an itch points it at something. What fires progress is the inspiration, the creation and the drive of the people using it: the civil servant who is done retyping the same letter, the developer with an idea at two in the morning, the nurse who knows exactly which form wastes twenty minutes per patient. Intelligence just lowered the cost of acting on all that human restlessness by an order of magnitude.

The point

Progress is not fired by the model. It is fired by inspiration, creation and drive, and those were never for rent.

That is also why the commodity turn matters more here than anywhere: when the engine becomes cheap and ownable, the scarce resources become the ideas and the data, and both of those are already yours. This is the same conclusion as the two-wolves post, arrived at from the economics side: feed the one that ships.

And then, just do it

Everything up to here has been defence: own the model so nobody else can rent, watch or weaponise your mind. Now stop defending for a moment and dream, because the same setup that protects you is the most powerful creative tool a single human has ever held. Let me show you what it actually looks like when it is switched on.

Picture every person with their own Hermiq, running on their own hardware, wired into the things that are genuinely theirs: their email, their calendar, their documents, their files, the whole scattered knowledge base of a working life. Not a chatbot in a browser tab that forgets you between sessions, but a companion that already knows your projects, your writing, your half-finished ideas, and never sends a byte of any of it anywhere. Ask it to read the three hundred pages before tomorrow's meeting and it hands you the four that matter. Ask it to chase a question across everything you have ever saved and it comes back with the answer, sourced, in the time it takes to make coffee. Knowledge stops being something you google and skim and half-remember, and becomes an exact handout of precisely what you need, the moment you need it.

Then point the same thing at building. You have an idea for a tool that should exist and never did, the small application that would fix the exact thing wasting an hour of your week. Yesterday that meant a budget, a vendor, a six-month project and a business case nobody would sign. Today you describe it, and Hermiq with a builder like Buildiq assembles it around you, on your own data, by the afternoon. The distance between "someone should build this" and "I am using it" collapses into a conversation. Every good idea that used to die in the gap between having it and affording it now simply gets made.

And here is the part that should make you sit up. Think about the parts of your job you quietly hate: the administration, the status reports nobody reads, the spreadsheet you rebuild by hand every month, the form-filling and copy-paste and reconciliation that eat the day. All of it is exactly the work a model does best and a human does worst. You can robotise the entire grey layer of modern work and keep the part you actually became a nurse, a teacher, a developer or a civil servant to do. Cut the drudgery out of the job and what is left is the reason you took it.

The dream, and it is buildable

Cut the admin and the Excel and the drudgery out of your day, and what is left is the work you actually wanted.

Nike coined "just do it" for sport. We are about to borrow it for something far bigger, because we are starting the most interesting game of our lives: the one where an ordinary person, with an owned model and their own data, can research, decide and build at a speed that used to belong only to institutions. The limit stops being money, headcount or permission. The limit becomes your imagination and your nerve. You get to make your own world, and then live inside it. So make it. Just do it.

We have a standing chance, if we start now

The money chapter ended with Europe holding 19 percent of the capital and none of the mega-funds, and I stand by the diagnosis: we will not out-raise the big three, ever. But the commodity turn changes which game is being played. When frontier-class intelligence costs two thousand euros once, the contest stops being "who can borrow 122 billion" and becomes "who can integrate, host, secure and trust the thing inside real organisations". That is not a capital game. That is a craft game, and craft is the one input where Europe was never behind: the Common Ground movement, the open source fleets, the integrators who already live inside gemeenten, hospitals and MKB.

The window is real and it is now, because windows like this close the way they closed before. The last commodity wave, the PC and the web, ended with new landlords because nobody guarded the exits while the getting was good. If we spend the next two years putting rented AI subscriptions into every workflow the way we once put Office macros into every process, the ownership chapter repeats itself with intelligence in the leading role, and this time the landlord holds your second brain.

Rent the hosting. Own the keys.

Which brings me to the nuance that saves this from being a self-hosting sermon: not everybody needs a local Nextcloud, and not everybody needs an AI box humming under the desk. Most organisations should not run their own infrastructure, for the same reason they do not generate their own electricity. SaaS is a perfectly healthy arrangement, on one condition: you rent the hosting and the support, never the ownership. Full disclosure, this is the model my company sells, so judge the argument on its merits, not my invoice.

The test is one question in every procurement: if this vendor disappoints us, what does leaving cost? If the answer is "an export and a weekend", you own your stack, wherever it runs. If the answer involves the words "migration project", you are a tenant, and the rent will rise.

So here is the momentum forward, the whole series in one working method. Understand that the wave is faster than your instincts and start this quarter, not after the next evaluation. Do not fund the coronations; the giants' own competitors are making their product a commodity. Buy back the base of your pyramid: one card, one open model, pointed at your own archive, worked by your own people and their own ideas. And where you do rent, rent operations, never ownership.

Buy the card. Keep the keys. Feed the wolf that ships. The window is open now, and it will not stay open because we deserve it.

Sources

1. Interconnects (Nathan Lambert). Kimi K3: the open-weights escalation. The open frontier three points behind the closed one.
2. The Rundown. Moonshot's Kimi K3 closes the frontier gap. Near-frontier capability at a fraction of frontier prices.
3. AI Weekly. Kimi K3 puts open weights within months of frontier. The shrinking gap estimate.
4. Conduction ConNext. "This is going to hurt" (2026 talk). The deck behind this series; the VRAM trend and the break-even arithmetic are illustrative, anchored to the cited endpoints, and the subscription-versus-card chart is deliberately labeled arithmetic rather than a TCO study.