The Economy of Capability

Essay 2 of 3 — On the economic layer of the daemon future.

Content creation was not predicted.

In 1995, no one writing about the future of work imagined that twenty years later, millions of humans would earn livings producing videos for audiences they would never meet. The category did not exist. The infrastructure did not exist. The cultural permission to consider this a “real job” certainly did not exist. Then four things arrived — cheap cameras, broadband, ad-supported platforms, social distribution — and a new class of livelihood became possible. Not because anyone designed it, but because the conditions crystallised, and humans sorted themselves into the new shape the conditions made room for.

That this emergence followed a power law is sometimes mistaken for failure. It was not. Audience-attention is fundamentally scarce; any economy built on competing for it produces a steep distribution. The middle of every platform is not a graveyard of failure. It is the structural shape of competition for a scarce resource. The remarkable thing is not that most creators don’t get rich. The remarkable thing is that the category exists at all.

I open with this because the next emergence is now visible, and I want to be precise about what makes it different.

The unit of economic participation in the AI age will not be content. Content has been commodified by synthesis — any capable model produces a competent post in seconds at marginal cost approaching zero. The marginal value of a marginal post falls to its marginal cost, which is nothing.

The unit that survives is the skill. By which I mean: a discrete, executable capability that an AI can install, invoke, and reward.

A recipe is a skill. Not the description of one — the executable version, the one a robotic kitchen can run, the one that produces a specific result with a specific signature. A method for transporting coal more efficiently down a hillside is a skill, in the world where mining is automated. A new technique for translating idiom between languages is a skill. A way of arranging a room to produce a specific psychological effect is a skill. A particular bedside manner, formalised, is a skill.

Skills occupy a different scarcity than content. Content competes for attention. Skills compete on capability — outcomes a daemon can verify, results a robot can produce. The economics rotate ninety degrees.

Three innovations must mature for a skill economy to exist at scale. None exist in finished form today.

Discovery. Personal AI agents that know their humans well enough to recommend, install, and invoke skills from a global catalogue. The previous essay established the daemon as a local witness; the daemon’s intimacy with its human is what makes discovery possible without a central marketplace operator. The matching happens on the daemon itself.

Payment. Micropayments per invocation, absorbed into the cost of service. When a robotic kitchen executes my recipe, my dollar comes from the meal price, not a separate transaction the eater notices. This is how streaming royalties work for music; it is how skills must work. Direct consumer payment will fail for the same reason content micropayments failed: friction is fatal at small dollar amounts.

Attribution. The hardest. When my skill is invoked, I must be paid. When my skill is learned — observed, reverse-engineered, generalised into a new model’s capability — I must also be paid, or the structure collapses into the training-data extraction tragedy we are already living through. This requires a global registry that traces lineage across model generations: append-only, decentrally maintained, with no single operator who can be captured. Readers familiar with public ledger architectures will recognise the structure.

This last point is the political claim. Every previous attempt at neutral global governance — ICANN, the WTO, the Bretton Woods institutions — has been progressively captured by the most powerful party at the table. The protocol form is the only form that has historically resisted capture, because there is no party to capture. TCP/IP, HTTP, email are the existing examples. Skill attribution will need to join them, or the economy I am describing collapses back into the rent-extraction shapes we already know.

The barrier to economic participation in this economy drops dramatically. Not to zero — most skills will fail the bar of being more useful than what a competent local daemon can synthesise from first principles. The shy cook who would never open a restaurant can publish a dumpling-dough technique that ten thousand kitchens execute. The introverted carpenter can publish a joint that ten thousand workshops cut. Quiet competence becomes monetisable in a way it has not been since the medieval guilds — but only if the competence is genuinely novel.

The axis along which this economy distributes reward is not audience but novelty. The reward goes to whoever genuinely innovates, whose innovation survives the empirical test of “is this better than what an AI could derive from scratch right now?” That bar rises every year. The skill economy is not a permanent home for any creator; it is a treadmill in which yesterday’s innovation becomes today’s baseline.

A power law will operate here too. It always does in any economy with discovery dynamics. But the axis of the curve has rotated, and that matters: a power law along novelty rewards different humans than a power law along audience. The shy and the unphotogenic stop being disadvantaged. Different humans win, on the merits of the work itself.

The clearest gestural form of this dynamic, available today, is the case of OpenClaw.

In February 2026, a solo Austrian developer named Peter Steinberger built an open-source AI agent framework that went viral. OpenAI hired him; Microsoft and Meta also competed for him. The compensation was talent-acquisition style, not protocol-level, and the project itself is moving to a foundation precisely because the skill-economy infrastructure does not yet exist to monetise it directly. Steinberger is also not anonymous; he previously built PSPDFKit, used by Apple, Dropbox, and SAP.

This is not the mature form of the thesis. It is the gestural form: individual creator, globally significant capability, hyperscalers competing for access. The mature form replaces the acquisition with protocol-level compensation. We are not there yet. We can see the shape from here.

Falsifiable bets:

The first creator paid one million dollars via protocol-level skill compensation emerges by 2030.

The first hundred-million-dollar skill creator emerges by 2035, in a domain we cannot yet predict.

These will not happen on the timelines currently circulating in venture decks. The infrastructure is not there. The discovery layer is not there. The attribution protocol is not there. The 2027 forecasts are confusing the gestural form — acqui-hires, viral open-source projects — with the mature form. My forecast assumes maturity, and maturity needs more time than the hype cycle wants to admit.

What is at stake here is not the wealth of a few creators. It is whether eight billion humans have a path to participate in an economy increasingly run by machines on their behalf. Content creation gave us one such path, unexpected and remarkable. Skills will give us another, also unexpected, also remarkable, in the same lineage of new categories of livelihood that emerge when infrastructure rearranges what is possible.

A note for the coal miner with a better idea: the world I am describing is one in which your idea, properly attributed and propagated, makes you wealthier than the boss who currently profits from your work.

If this seems unlikely, observe: stranger inversions have occurred.

Next: the software layer. If skills are the economic primitive and daemons are the substrate, what becomes of the operating system that has structured personal computing for forty years?