The Physical Coherence Fee
All four terms of the agent cost calculus push against structural verification; the coherence fee is externalized to bystanders who cannot protect themselves
A market mechanism or regulatory structure that induces voluntary compositional verification at organizational boundaries without mandated infrastructure
The Physical Coherence Fee
Status: Conjectural
Date: 2026-03-21
I. The Barbell Boundary
The agentic economy will not produce a trillion atomized agents freely transacting. Larson's argument against the Coasean Singularity is persuasive: AI makes context the scarce resource, context revelation is expensive, and firms will vertically integrate to protect their context advantage. The result is a barbell — very large autonomous systems and small businesses using AI to serve local needs.
But the barbell has boundaries. And in the physical realm, those boundaries cannot be avoided.
Waymo's fleet is internally coherent. Zipline's drone network is internally coherent. Amazon's warehouse robotics are vertically integrated from shelf to truck. Each is an autonomous commerce system: it owns its supply, owes liability for its outputs, and operates independently. Inside the walls of the firm, the coordination problem is solved by fiat — one engineering team, one convention set, one integration test suite.
These systems share physical space. Waymo vehicles drive on the same roads as Amazon vans. Delivery drones from competing fleets share low-altitude airspace. Autonomous sidewalk robots from different vendors share the same curb cuts, crosswalks, and building entrances. None will cede control to a single integrator, because doing so would mean revealing the internal representations — the decision logic, the proprietary sensor models — that constitute their competitive moat.
The compositions that form at these boundaries are fewer than the Coasean Singularity imagines. But they are harder. They occur between opaque, context-guarding systems with the strongest possible incentive to not share internal conventions. The bilateral protocol at the boundary validates message format, authentication, and schema conformance. It does not — and given the context economics, cannot — verify semantic convention alignment across the composition.
II. The Scenario
Three vertically integrated autonomous systems must compose in urban airspace.
Fleet A operates delivery drones. Its dispatch agent manages 200 drones simultaneously. Internally, Fleet A represents all altitudes as meters above ground level (AGL) — the convention specified by the FAA for small unmanned aircraft under 14 CFR Part 107.
Fleet B operates autonomous air taxis. Its navigation stack represents all altitudes as feet above mean sea level (MSL) — the ICAO standard for manned aviation, inherited by Fleet B's aerospace engineering team.
The City Airspace Coordination Service (CACS) is a third autonomous system, operated by neither fleet. It mediates shared airspace by receiving position reports, computing deconfliction corridors, and issuing clearances. CACS interprets altitude values in meters MSL — a reasonable engineering default that matches neither fleet exactly.
No unified convention exists for mixed autonomous airspace. The FAA's AGL standard governs small drones; ICAO's MSL standard governs manned aviation; nothing governs the composition of both in a shared urban corridor. Each system's convention is internally correct. The bilateral protocols between each pair validate — valid coordinates, valid timestamps, schema conformance. Protocol: green.
Fleet A dispatches a drone at 120 AGL over a hilly neighborhood where ground elevation is 45 meters MSL. The drone's true altitude is 165 meters MSL. Fleet A reports "altitude: 120" to CACS. CACS interprets this as 120 meters MSL — 45 meters lower than the drone actually is.
Fleet B requests a corridor for an air taxi at 150 feet MSL (approximately 45.7 meters MSL). CACS interprets this as 150 meters MSL. CACS checks separation: Fleet A at 120 MSL, Fleet B at 150 MSL, 30 meters apart. Clearance granted.
In physical reality: Fleet A's drone is at 165 meters MSL. Fleet B's air taxi is at 45.7 meters MSL. They are 119 meters apart. This particular mismatch produces a false near-miss — the actual separation is larger than computed. The system appears to work. Confidence builds.
Now the terrain changes. Over a river valley where ground elevation drops to 2 meters MSL, Fleet A dispatches a drone at 50 AGL. True altitude: 52 meters MSL. Fleet A reports "altitude: 50." CACS records 50 meters MSL. Fleet B's air taxi cruises at 160 feet MSL (48.8 meters MSL). CACS sees 50 and 48.8 — 1.2 meters of apparent separation. CACS issues an emergency avoidance command to Fleet B: descend immediately.
Fleet B's air taxi drops toward the river. The actual separation was 3.2 meters and was already being managed by Fleet B's onboard collision avoidance. The CACS override sends the air taxi into an unplanned descent toward a bridge at 40 meters MSL. It clips a cable.
Every system behaved correctly given its own conventions. Every bilateral check passed. The composition was globally incoherent, and the incoherence was invisible until a particular terrain profile activated the latent mismatch.
In 1999, the Mars Climate Orbiter was lost because one team used Newtons and another used pound-force — a $328 million convention mismatch between independently correct components. That was an episodic engineering failure in a pre-agentic context: two teams, one spacecraft, caught too late. The agentic economy makes this failure mode structural and continuous. Autonomous systems compose at organizational boundaries every day, under economic pressure, without the integration testing that NASA's failure analysis recommended.
III. The Augmented Indifference Curve
Each participant in this composition faces a continuous cost calculus with four terms.
Direct task cost. The fuel, compute, and actuation required to complete the immediate mission — deliver the package, transport the passenger, manage the airspace.
Existence cost. The amortized compute, electricity, and data-link costs that each agent incurs every second of runtime, whether or not it is doing productive work. Fleet A's dispatch agent runs a navigation stack, a battery management model, and a market-bidding engine for 200 drones. This infrastructure burns energy continuously. Every second spent on coordination overhead is a second of existence cost with no marginal revenue.
Opportunity cost. Fleet A's dispatch agent is managing 200 drones. Processing cycles spent verifying altitude conventions with CACS are cycles not spent optimizing the next delivery route — a measurable reduction in fleet throughput.
Context revelation cost. For Fleet A to participate in a structural coherence check, it would need to disclose that its internal convention is meters-AGL. This reveals architectural detail about Fleet A's sensor stack, terrain model, and obstacle-avoidance strategy. In a competitive market for airspace access, where CACS also serves Fleet B and every other competitor, this is a context leak. The same information asymmetry that makes transacting expensive in digital markets — where revealing preferences leaks competitive signal — applies with equal force to physical coordination protocols.
All four terms push in the same direction: trust the bilateral check, move. The bilateral check passed. The altitude number was valid. The schema was correct. The protocol was green. Proceed.
The standard objection is that this is an engineering standards problem — that a regulatory authority should mandate a unified altitude convention. Standards solve the convention problem when a single authority can mandate them, as the FAA does for manned aviation and ICAO does for international air traffic. But the agentic economy produces compositions at organizational boundaries where no single authority exists — vertically integrated systems from different jurisdictions, different industries, different regulatory regimes, each guarding its context. The structural diagnostic is needed precisely in the regime where standards cannot be imposed and context economics prevent voluntary disclosure.
IV. The Cycle
This is not a one-time error. It is a persistent structural condition.
CACS operates in a continuous feedback loop: it receives position reports from all fleets, computes separation models, issues clearances or avoidance commands, observes the resulting trajectories, and updates its model. Fleet A's dispatch agent adjusts routing based on CACS clearance patterns. Fleet B's route planner similarly adapts.
The cycle — Fleet A reports, CACS interprets and issues clearances, Fleet B responds, CACS updates its model, Fleet A observes the pattern and adjusts — accumulates the convention mismatch through every iteration. Over flat terrain, the errors are small and the system appears healthy. Over varied terrain, the errors are large and unpredictable. CACS's separation model is systematically wrong, but its internal diagnostics show all protocol checks passing, all messages well-formed, all schemas validated. Its bilateral view of each fleet is green. The composition's first cohomology is nontrivial, and no single participant can see it.
V. The Externality
The air taxi passenger did not choose this composition. They did not select Fleet B knowing it would compose with Fleet A through a CACS instance that interprets altitude conventions differently. They did not consent to the convention mismatch. They have no ability to verify the compositional coherence of the airspace coordination system. They experience the coherence fee as a physical event.
This is the constitutional structure of the physical coherence fee. In the digital realm, convention mismatch produces a wrong dollar amount on a ledger. The affected party can dispute, reverse, or litigate. The error is correctable. In the physical realm, the coherence fee is paid by third parties — passengers, pedestrians, building occupants — who were not party to the composition and cannot undo the consequence.
VI. Why It Won't Be Voluntary
The augmented indifference curve ensures this. Every term pushes against voluntary adoption of structural verification. Direct task costs favor speed. Existence costs penalize coordination overhead. Opportunity costs penalize time spent on verification instead of revenue-generating activity. Context revelation costs penalize the disclosure needed for structural checks.
In a competitive market, the fleet that skips structural verification and trusts bilateral checks will have lower operating costs, faster dispatch, and higher throughput than the fleet that pays for compositional coherence. The market rewards the behavior that produces the externality.
The coordination substrate — the structural diagnostic that detects cyclic convention mismatch before physical execution — must be embedded in the coordination protocol itself, not left to each participant's discretion. The bilateral protocol surface is the only check, and the technical spine demonstrates it is structurally blind to the failure mode that matters. The physical realm does not extend this argument by degree. It extends it by kind: from correctable accounting error to irreversible physical consequence, from counterparties who can dispute to bystanders who cannot, from useful diagnostic tool to constitutional infrastructure.