The wealth has been flowing one way. That ends with coordination.

Treat the current distribution as a metastable state, not a stable one.

In quantum field theory, a false vacuum is a configuration that looks stable from the inside but is not the ground state of the system. It persists only because there is no cheap mechanism for transitioning to the true minimum. Provide one, and the transition is rapid, irreversible, and globally observable.

The post-1980 distribution of returns is exactly this. A coordination cost barrier — not a productivity differential, not a preference, not a market clearing — sustains the upward flow of margin from atomized demand to coordinated supply. Workers don't bargain because organizing 200 of them is harder than the firm's procurement team rebidding their wages once a quarter. Patients don't shop because reading three EOBs is harder than the payer's actuarial team running ten thousand denials a day. Tenants don't combine because moving is more expensive than the landlord's pricing algorithm.

The barrier is real. It is also arbitrary. It does not reflect what either side wants; it reflects what either side can currently afford to coordinate. AI agents make individual-side coordination approach zero marginal cost. When that happens, the equilibrium does not gradually adjust. It tunnels.

This is not a policy proposal. UBI proposes to redistribute the output of an unchanged extraction stack. We propose to change the stack — to make the price/wage/rent/premium/fee that emerges from any transaction itself a coordinated quantity, in the same way that supply-side prices are already coordinated by procurement, pricing algorithms, payer networks, and trade associations.

Supply is a network. Demand is a list of names.

Every transaction in modern life is between an entity with full information, full coordination, and statistical recourse — and an entity with none of those things. The asymmetry is so naturalized that it is invisible. Once you look at it directly, the wealth flow stops being mysterious.

Five places the asymmetry is already extracting

• Medical billing. A hospital's revenue cycle team runs the same denial-and-adjustment script ten thousand times a day. A patient runs it once, badly, while sick.
• Labor markets. An employer has compensation data for every role in every region. The candidate has Glassdoor, lagged eighteen months, with selection bias.
• Rent. Pricing software with cross-property data sets a metropolitan rent floor. Tenants do not know which other tenants in the building got which renewal offer.
• Insurance. An actuarial team has lifetime claim curves on every cohort. The insured has a policy document and a phone tree.
• Drug pricing. A PBM negotiates rebates across formularies for ten million covered lives at once. The patient negotiates by walking out of CVS and trying Walgreens.

Six layers, one mandate: make the demand side recursive.

The stack below is the minimum viable architecture for converting individuals into a coordinated counterparty without requiring any of them to know one another. Each layer is intentionally protocol-shaped rather than platform-shaped — the goal is not to capture coordination as a service, it is to make coordination cheap enough that capture is unnecessary.

The crucial design decision is at L5: the agent's incentives. A "smart assistant" funded by referral fees from supply is not in this stack — it is the existing extraction stack with a friendlier voice. The agent is principal-loyal or it is not in the layer. This is enforceable through both legal structure (fiduciary registration, similar to RIAs) and technical structure (the agent's negotiated outcomes are publicly auditable against the coalition's claimed clearing price).

Why this is not group buying

Group buying compresses one round of one purchase. The coordination layer runs continuously, across categories, with a memory of what each principal accepts as a clearing price. A landlord who raises rent triggers a search not just for cheaper units but for cheaper everything — utilities, internet, insurance — across the entire coalition that the agent suspects shares the same affordability constraint. The unit of coordination is not the transaction. It is the life.

What the agent is actually doing when it acts.

The semi-autonomous search at L3 is not browsing. It is a constrained optimization that runs continuously over the principal's life, the agent's coalition memberships, and the counterparty surface. A useful first formulation:

The interesting term is p*(N). Classical bargaining theory tells us that the clearing price falls as the coalition grows, up to the point where the supply side prefers to walk away. Agents do not need to negotiate cleverly; they only need to credibly threaten coordinated walk-away. The threat is credible because the agent has already pre-scored the next-best counterparty for every principal in the coalition and can switch atomically if the auction fails.

The second interesting term is Cᵢ(coord). This is the historical barrier — phone trees, signature campaigns, organizer payrolls. With protocol-level matching it falls below the variance in any individual's daily utility. The barrier disappears not because anyone subsidizes it but because nobody has to pay it.

What you can ship this year.

Each of the following is implementable with today's model capabilities, today's protocols (HTTPS, OAuth, basic agent SDKs), and today's legal infrastructure. None require breakthroughs. They require willingness to build with demand as the principal.

Ten cases. None of them require AGI. All of them are blocked by a single missing piece: an agent infrastructure that is principal-loyal by design. That is the work.

When the negotiation lives in the world, not the app.

Spatial computing is the moment the coordination layer leaves the screen. Once the agent has continuous access to what the principal is looking at, every object becomes a live offer surface — and every supply-side price becomes a quote the principal already has a better number against.

AR-native use cases

Three years out, the coordination layer is load-bearing.

By 2029 the protocol matures, the legal scaffolding is in place, the principal-loyalty norm is enforceable, and the supply side has stopped treating agent traffic as deflectable noise. The use cases below assume not new physics but compound buildout: more identity verification primitives, more A2A-fluent counterparties, more legal precedent on coordinated walk-away.

This is not gradual. It is nucleation.

The historical pattern of consumer protections is incremental, decades-long, regulator-driven. The coordination layer is none of those things. Its rollout dynamics resemble first-order phase transitions far more than ordinary policy diffusion.

The supply side will respond, of course. The first responses will be familiar: regulatory capture attempts, claims that demand coordination is "anticompetitive" (the irony of which the historical record will note), platform-level blocking of agent traffic, contractual prohibitions on agent representation. None of these survive sustained encounter with the underlying economics. Once principals have agents that pay for themselves out of captured surplus, the agents do not stop.

Why this is the speed of light, not the speed of policy

UBI requires a political coalition large enough to redirect tax flows. The coordination layer requires only a single principal somewhere willing to install an agent. The first agent that captures, say, $400/yr in pharma rebate reclaim has paid for itself, and the principal tells two friends, and the friends install agents, and each successful coalition makes the next coalition cheaper to form. The deployment curve looks like an S, not a line; the S is steepest in the middle, which is where we appear to be entering.