// writing / the_window

The Window

Build and own before superintelligence subsumes the unowned.

james loperfido · 2026-07-01 · thesis essay
// tl;dr
  • 01Superintelligence is coming. Not in 18 months. Short run overestimated, long run underestimated.
  • 02Between now and then there is a window. Inside it, humans drive every non-physical value engine on earth, and AI is the highest-leverage personal tool in the history of work.
  • 03The play is not to chase the frontier. The play is to use AI to build and own: a body of work, a brand, a customer relationship, a treasury, an identity.
  • 04Ownership requires sovereignty. Keys, custody, identity, anchored to you.
  • 05What is owned survives the transition. The unowned does not.
// 01the_claim// read it twice

Superintelligence will arrive. It will not arrive on the schedule the loudest voices are selling. In the gap between those two sentences sits the single largest personal opportunity of your working life. Call it the Window.

While the Window is open, humans drive every non-physical value-creation engine on earth. Every deal, every brand, every book, every customer relationship still routes through a human decision. And for the first time in history, that human holds a lever that multiplies judgment instead of hours: AI as a personal tool.

The play is not to predict when the Window closes. The play is to use the lever, now, to build and own something durable: a body of work, a brand, a business, an IP catalog, a customer list, a treasury, an identity. What is owned survives the transition. The unowned does not.

// 02the_timeline// amara's law

Amara's Law governs every major technology: we overestimate the effect in the short run and underestimate it in the long run. AI is not exempt. The 18-month-takeoff crowd and the it's-all-hype crowd are making the same mistake from opposite ends.

Watch the balance sheets instead of the podcasts. The current compute-everything race is subsidy-masked: tokens are priced below cost to buy market share, the way rides and DVDs-by-mail were before them. When the subsidies unwind, tokens re-price the field. What survives that re-pricing is not the biggest model. It is the work that got built and owned while compute was cheap.

Here is the part the frontier obsession misses: for roughly 80% of useful automation work, a model six months behind the frontier is enough. Often it is better: cheaper, faster, runnable on your own hardware, yours. Frontier obsession is a vanity metric. Small models plus agent loops win on cost and fidelity for most real work.

// 03the_direction_of_the_machine// who is the product

The default AI economy has a business model, and the business model has a requirement: you are the product. Your inputs train someone else's model. Your work feeds someone else's moat. Value flows up; you rent capability back, by the month, at a price you do not set.

This is not a conspiracy. It is a funding model doing what funding models do. But it means the platforms cannot serve the person who wants to own what they build. Structurally cannot. Their economics require the opposite.

And the same institutions will decide what you get access to. In June 2026 the US government blocked export of a frontier model on national-security grounds. That is the receipt: the most capable systems will be licensed, gated, and withdrawn on someone else's schedule. The answer to a gated frontier is an owned stack. Keys with you. Agent with you. Context with you.

Sovereign here means customer-side sovereignty. Not data-residency paperwork. Not a training opt-out checkbox. The keys, the agent, the context, the identity: yours, in your custody, portable, inheritable.

// 04what_to_build_and_own// the balance sheet of the window

Think of the Window as a balance sheet you are racing to fill with assets that survive re-pricing:

  • 01A body of work. Writing, code, product, IP. The compounding asset. Taste and judgment, operationalized in output no model can claim.
  • 02A brand and an audience. A named position and the people who trust it. Distribution you own beats distribution you rent.
  • 03Customer relationships. Names, context, history, consent. The list that moves with you when every platform re-prices.
  • 04A treasury. Capital in your custody, on rails you control. Self-custodied keys are the difference between owning and being owed.
  • 05An identity. Cryptographic, portable, inheritable. The anchor everything else attaches to.

Notice what is not on the list: a subscription, a following on a platform that owns the graph, a workflow that dies when a vendor changes terms. Those are liabilities wearing asset costumes.

// 05what_survives// the bet, stated plainly

The bet is not that open source wins the model race. The bet is not that any lab, incumbent, or government gets the timeline right. The bet is simpler and older: smart people using the best available tools to build something that is theirs, before the leverage flips to autonomous systems.

The owned work is the moat. The owned relationship is the moat. The owned identity is the moat. Sovereignty is not a virtue signal; it is the structure that makes the ownership durable across the transition.

The Window is open. It will not announce its closing. Build before they take it.

// next

This essay is the frame behind everything on this site. The manifesto states the principles; the education page teaches the method. New issues ship by email first.