A Coasean Democracy

A grocer needs a late delivery. The neighbours above the loading bay need sleep. A delivery company could use a quieter vehicle, arrive earlier, or pay for an exception, but each option costs something. Usually the disagreement ends in a complaint, a fixed rule, or an arrangement between whoever happens to know whom.

Now give the parties agents that can compare schedules and costs, carry offers back and forth, and draft an agreement for approval. A compromise that would have taken three evenings of messages might become practical. That is the attraction of a Coasean democracy: more room for people to arrange their shared lives without having every small decision settled for them.

My thesis is that cheaper agent bargaining could expand democratic agency only if people retain control over the rights, representation, and reasons that make a bargain legitimate. An agreement between programs is evidence of coordination. To call it democratic, I would want to know who was included, what they authorised, and what they can still contest.

What Coase makes visible

Ronald Coase’s 1960 paper begins with conflicts such as cattle damaging a neighbouring farmer’s crops. Avoiding the damage can also impose a cost on the cattle raiser. The question becomes which arrangements make the best use of the available resources, given the rights and the costs of changing them. His costless-bargaining benchmark helps isolate the problem; his discussion then restores the very real work of finding parties, negotiating contracts, and checking performance. The Problem of Social Cost, sections II–V.

Those are transaction costs. A beneficial agreement can fail to happen because arranging it costs more than it creates. Initial rights also decide who must seek permission and who can demand payment. Efficiency does not, by itself, tell us whether that starting position is fair.

Séb Krier’s Coasean Bargaining at Scale proposes personal advocate agents as a way to make previously impractical negotiations possible. Herbie Bradley’s Coasean economics of agent swarms explores related reductions in transaction costs within commercial arrangements. I want to test the political possibility alongside the economic one: what happens to a citizen’s effective voice when representation becomes cheap to produce but costly to supervise?

One delivery, two sides of a bargain

Start with a deliberately small model. The residents hold the right to refuse a particular evening delivery. The grocer values an exception at 120 credits. The represented residents would accept at least 40 credits for the disturbance. Assume money can compensate this particular inconvenience, both sides can make the transfer, and no protected right is being waived. These are experimental assumptions, not a price for anyone’s actual sleep.

Before negotiation costs, there are 80 credits of possible joint gain. If arranging and checking the agreement costs 20 credits, split equally between the sides, a payment of 80 leaves each side 30 credits better off than no delivery. Above 80 credits of transaction costs, this model has no positive joint gain to divide.

The payment changes who receives the gain. Reducing the cost changes how much gain exists. An advocate could help by finding a suitable time, checking who is affected, comparing a quiet vehicle with compensation, and keeping a record that everyone understands. Merely generating more counteroffers would not establish that it had reduced the total cost.

Now introduce a night-shift nurse whose window also faces the loading bay. If her interests were left out, the two positive balances in the illustration would give an incomplete account. Or suppose one resident’s agent claims an enormous disturbance cost because it has learned that being the last holdout attracts a payment. The bargaining problem now includes discovering preferences without rewarding whoever exaggerates most effectively.

Initial rights matter just as much. If the grocer already had the right to unload, residents might need to pay for an earlier delivery. That reverses the direction of the offer and changes the consequences of having little money. A system cannot make its starting rules disappear by negotiating rapidly inside them.

A shared battery is a different kind of conflict

Consider an apartment building with a battery during a power cut. One household wants to cook; another wants to work; a third depends on powered medical equipment. Agents could coordinate meal times, reschedule charging, and discover spare capacity. Those are plausible gains from exchanging information.

In my proposed rules, an essential reserve for medical needs would be protected before discretionary use was allocated. An agent could not sell that reserve because another household offered more. Residents would deliberate over how the reserve was defined and reviewed, with appropriate privacy protections. Bargaining would then allocate the flexible remainder.

This example exposes a limit in the delivery calculation. Some objections are claims about what people are owed, rather than statements of the price at which they will stop objecting. A system that turns every objection into a counteroffer has already made a political decision about the kind of claim it is willing to hear.

Habermas: a reason is more than an offer

Jürgen Habermas’s account of deliberative democracy makes public justification central to legitimate political decisions. Participants must be able to question proposals and the rules governing them. In Between Facts and Norms, he also makes room for compromise between interests that remain different, while asking what makes the bargaining conditions fair. Deliberation and bargaining have related but distinct jobs. Between Facts and Norms, discussion of bargaining and compromise.

Applied to our street, “I will pay you more” and “the proposed restriction prevents food reaching this neighbourhood” are different contributions. The first adjusts an offer within an existing frame. The second asks others to reconsider how the problem is understood. Someone might change their judgement after hearing it, even without receiving compensation.

I would therefore give the agents a public reason-giving stage before the offer stage. They would identify who is affected, state the evidence behind a proposal, preserve objections, and return unresolved questions to the people they represent. Residents could revise their priorities after seeing those arguments. The record would distinguish a changed view from a concession bought with a payment.

That is my application of Habermas, not a claim that he proposed agent government. A room full of fluent agents could produce orderly dialogue while concealing the absence of meaningful human participation. I would judge the process partly by whether residents can understand the reasons, correct their representatives, and introduce a concern that the agents had omitted.

The transaction costs of agent deliberation

Cheaper speech can create more work elsewhere. If each agent generates a hundred careful objections, someone must decide which are distinct, which rest on evidence, and which matter. Summarising them introduces another representative whose omissions need scrutiny. The cost may move from writing a contribution to checking what became of it.

I would keep separate accounts for finding participants, eliciting their preferences, running the agents, verifying claims, obtaining human approval, and enforcing the final arrangement. Compute cost and elapsed time would sit alongside human review minutes, privacy disclosures, and unresolved disputes. A negotiation that saves machine time while multiplying approval requests has not necessarily become cheaper for its participants.

Some friction is worth preserving. Time to consult a neighbour, challenge a false claim, or recognise that an apparently minor exception will recur can improve the decision. I would compare a deadline-driven negotiation with one that guarantees a review window and permits an unresolved objection to reach a human chair. Faster agreement would be one measurement, not the objective that determines every rule.

Who authorised the negotiator?

Agentic Profiles gives this experiment a way to describe its representatives. One advocate can only draft offers. Another may accept a schedule within a budget. A third can commit its resident to a recurring contract. The authority to bind someone needs to be recorded separately from how eloquently the agent argues.

I would start with revocable mandates: the questions an agent may negotiate, its budget, the commitments requiring approval, and the time its authority expires. A material change in tools or permissions would prompt a fresh profile assessment. Residents should be able to inspect a proposed agreement without reading the entire conversation that produced it.

Copies would inherit the same principal’s mandate, rather than acquire new votes. That connects to the email essay’s proposed disclosure of agenthood: tracing a delegation helps us see who is speaking for whom. It cannot decide who deserves civic standing, or ensure that a well-resourced principal has no other advantage.

Tax the entitlement, preserve the voice

Taxing Agents asks what an agent’s operator actually holds: software, records, delegated authority, or access to a scarce resource. That distinction becomes political here. A delivery-slot licence could be transferable; a resident’s vote and the mandate to represent them would stay outside the sale under my proposed rules.

Imagine the grocer’s delivery agent holds a scarce booking-channel licence under the Harberger arrangement explored in that note. Its operator declares a price, pays a recurring tax, and must transfer the licence to a qualifying buyer at that price. Another operator buys it midway through a negotiated delivery schedule. Does the buyer inherit the obligation to use the quiet vehicle? Who completes the already-promised deliveries?

The experiment would attach service obligations to the transferable licence while keeping personal records and representative mandates separate. A licence sale would not authorise its buyer to speak for the residents. I would compare continuity under those rules with a condition requiring fresh agreement after transfer, recording both disruption and the cost of renegotiation. A more active market could create work for the very people it was meant to serve.

A vote that can travel, and come back

Liquid voting lets a person vote directly or delegate their voting weight. Delegation can be transitive: if I delegate to Maya and Maya delegates onward, my vote follows that chain. Research on transitive delegation examines this feature of liquid democracy. Adding a software representative is the extension I want to explore here.

In this proposed ballot, four residents decide on one evening delivery. Maya delegates to an agent called Civic. I can follow Maya, delegate to Civic directly, or cast my own vote. The agent carries only the votes entrusted to it. It gets no additional vote for being another participant in the conversation.

Try changing Civic’s decision, then taking your vote back. The proposal uses a simple majority of all four residents, and a tie does not authorise the delivery. Delegation applies to this question alone and can be withdrawn before the ballot closes. These are chosen rules for the illustration, not features every liquid-voting system must share.

The chain also makes concentration visible. One change in an agent’s decision can move several human votes. I would want each person to see that route, know whether onward delegation is permitted, and retain a practical way to intervene. Voting can settle a decision under agreed rules; it does not replace the earlier work of hearing objections or deciding which rights a majority may override.

What would persuade me?

I would compare periodic voting, elected representation, direct agent bargaining, liquid delegation, and bargaining within protected rights, using the same simulated residents and needs. Liquid delegation means a person can vote directly or delegate a decision and later change that delegation. The runs would vary wealth, agent capability, preference uncertainty, and negotiating costs.

The difficult trials would include an omitted resident, a coordinated holdout, a principal spawning many identities, an agent accepting beyond its mandate, and a licence changing hands before a commitment is fulfilled. Each tests a particular claim about inclusion, representation, or enforcement.

I would look for agreements that residents still endorse after their consequences are explained, fewer avoidable disputes, and broader effective participation. I would also record who repeatedly loses, who cannot afford an effective representative, and whether challenges produce corrections. A Coasean democracy becomes interesting when it makes more shared arrangements possible while leaving people able to question the terms on which those arrangements are made.