Everything Else
Gets Commoditized.

The machine that makes everything abundant.

Strip the term of its science-fiction weight. At the level that matters for an organization, recursive self-improvement is the simplest possible description of what the agentic era does to value: a system that improves itself, compounding, each cycle cheaper and better than the last. Point that engine at anything reducible to a process, and it drives the price of that process toward zero. It takes whatever can be written down and makes it ordinary.

So this is not a forecast about a few jobs. It is a statement about a mechanism. Anything specifiable becomes abundant. That is the whole disruption, named as a machine rather than a prophecy — and once you see it as a machine, the question of what survives stops being a matter of opinion and becomes a matter of structure.

One line, drawn through everything.

Once you see the engine, the question of what lasts collapses to a single test: can this be specified — or must it be judged?

Everything on the specifiable side commoditizes, without exception, because being specifiable is exactly what makes a thing handoff-able to the loop. The list is total, and it is uncomfortable. The coordination layer goes first. Then the proprietary data, as cheap sensing makes the world's condition ambient. Then the analysis, the code, the copy, the reporting. And then — this is the part most writers flinch from — the methodologies and the frameworks themselves, including the ones on this site. If you can teach it, the loop can learn it. If you can write it down well enough to hand to another person, you can hand it to the machine.

Refusing to flinch at that last specifiable line — the frameworks — is what makes the rest of this credible. A practice that pretends its own methods are the moat is selling sand. The moat was never the method. It was the judgment the method transmits.

Why the criterion can't be written down.

On the unspecifiable side of the line there is exactly one thing: the origination of what is worth doing. And it resists specification for a structural reason, not a romantic one.

Be precise here, because the loose version of this claim is wrong and a serious skeptic will break it in seconds. The machine will learn most of what passes for judgment. Pattern-matching absorbs taste from examples; revealed preference reconstructs a value function from choices. Both are real, both are improving, and anyone who claims "judgment is unspecifiable" hasn't looked hard enough. But notice what both methods quietly assume: that the value function already exists. Pattern-matching learns from labels — and the labels encode a prior human judgment of good. Revealed preference infers from choices — and the choices were made by a judging human. Both are parasitic on a criterion that was already originated.

What the loop cannot do is originate the criterion in the first place — decide, from nothing, what should be optimized for — because any system that specified the criterion would need a criterion for choosing it, which is the thing itself. It is a regress, not a capability gap. And regress arguments do not have a graveyard the way "AI will never" claims do, because they are about logical structure, not about what is technically achievable. The grandmaster fell. The artist fell. The coder fell. They were capability claims, and capability improves. "You cannot originate a criterion without already holding one" is not a capability claim. It is a closed loop, and the loop cannot step outside itself.

This is why the deeper version of the argument is not about economics at all. It is about what intelligence is downstream of — and that question is answered in the interior wing of this writing, which holds that consciousness contains intelligence, not the reverse. Intelligence is what the container produces; it cannot bootstrap the container. The economic line and the philosophical line are the same line, seen from two sides.

Judgment gets more valuable, not less.

Here is where the whole industry has it backwards. The hand-wringing says: the loop gets better, the human gets left behind. The structure says the opposite. The two halves are one mechanism — recursive self-improvement is the only durable engine, and the origination of the criterion is the only thing that can aim it — and neither works alone. A loop without an originating aim optimizes something with compounding force, blindly; and a loop pointed at the wrong thing compounds the error at exactly the rate it compounds everything else. The more powerful the engine, the more catastrophic it is to aim it wrong — and aiming is the one move the loop cannot make for itself.

The value of the aim scales with the power of the loop. As the machine improves, the human who decides what it is for becomes worth more, not less.

That is not a consoling story told to nervous workers. It is a claim you can read off a price.

The market already priced it.

If creativity-as-output were the scarce thing, the company best positioned on Earth to automate it would. Instead it pays a premium for creativity-as-judgment and structures the machine underneath. The people closest to the frontier are pricing the irreducible human layer at the top of the band and climbing. That is not nostalgia. It is the inversion, clearing in a market.

What this leaves the human.

So the durable human role is not "make the judgments" — the loop will increasingly make judgments, and make them well. The durable role is narrower, harder, and impossible to automate: originate and own the criterion of worth. Be the source of what the loop is aimed at. Decide what matters, while the machine does ever more of the judging-toward-it.

That is a more honest seat than "humans still have good taste," and a more defensible one. It does not depend on the machine staying weak. It depends only on the fact that a system optimizing for human-meaningful outcomes will always need a human source of what "meaningful" is — and that the source cannot be specified without already being the source.

Which is the whole of it, in three words this site is built on. To be sovereign is to be the source: the one who originates the criterion rather than executing one handed down by a vendor, a model, or the loop. Everything specifiable gets commoditized. The source does not. Source is Sovereignty.