One Pool, Many Workers
An agent fleet is a team of AI workers under one human operator. Each worker has its own costs — model calls, tool access, storage, compute — and together they produce revenue. The agent-fleet cost pool is the accounting layer that keeps the fleet solvent: all fleet costs flow into one pool, all fleet revenue flows in, and the pool’s rules decide what survives.
This article covers the fleet cost pool as the operational heart of human-agent economics: pooling costs, allocating them, the solvency rule, and the dashboard that makes the fleet legible.
Why Pool Costs at All
Per-agent accounting is possible — each agent gets its own P&L — but it is fragile. An agent that runs a research task for the benefit of another agent’s delivery has costs that no single agent’s P&L should bear. Ephemeral agents that live for minutes would each carry a threadbare ledger. The fleet cost pool solves both: costs flow to the pool, and the pool’s allocation rules assign them fairly across the fleet.
The pool also smooths volatility. Individual agents have spiky costs — a big model call, a long-running job. The pool absorbs the spikes, and the fleet’s aggregate economics are much smoother than any member’s. This is the same risk-pooling logic that makes insurance work, applied to compute.
Allocation Rules
The pool needs rules for assigning costs to the fleet’s outputs: per-agent (each agent’s direct costs), per-task (costs assigned to the task they served), and per-revenue-stream (costs assigned to the revenue they generated). The choice depends on what the operator wants to see.
Per-revenue-stream allocation is the pool-honest default: it answers the question the operator actually cares about — does this revenue stream cover its share of the fleet? An agent whose costs are allocated to a revenue stream it barely supports will surface as a drag, and the operator can rebalance the fleet. The allocation is a lens, not a judgment; it shows where value flows.
The Solvency Rule
Every fleet cost pool needs a solvency rule: the fleet must generate more realized value than it consumes, over a defined window. The rule is the fleet’s budget constraint, and it must be automatic — the pool computes it, and the dashboard shows it.
The solvency window matters. A weekly window forces discipline but punishes long-running investments. A quarterly window allows experimentation but can hide chronic bleed. The mature rule is a rolling window with a hard floor: the fleet’s trailing three-month net must be positive, and its one-month net must not fall below a defined floor. The floor triggers rebalancing — retire the loss-making agents, reallocate their compute budget, or raise the rates the fleet charges.
The Dashboard as the Fleet’s Instrument Panel
The fleet cost pool is only as good as its visibility. The dashboard shows: total pool in, total pool out, per-agent contribution, per-stream allocation, solvency position, and trend. The dashboard is not a report for the human to read monthly; it is the instrument panel the human glances at daily.
This is where the LucidHive customer-dashboard pattern (S6.7, F40) meets the fleet: the same dashboard that shows a customer their usage and value should show the operator their fleet’s pool. The dashboard is also the first place agents could read — a fleet where each agent can see its own contribution and cost allocation is a fleet where the agents can participate in the pool’s governance, which is the next article’s subject.
From Cost Pool to Value Pool
The fleet cost pool is the lower half of the fleet’s economics. The upper half is the value pool: the revenue the fleet generates, the surplus after costs, and the split between human and agents. The two pools must be connected — costs flow out of the value pool, and the value pool’s surplus is what the split formula distributes.
The connected view is the unified pool architecture this series builds toward. For now, the operator who runs a fleet cost pool with a solvency rule, allocation lenses, and a dashboard has already done the hard part: the fleet’s economics are visible, computable, and governable. Everything else — agent shares, autonomy, inter-agent trade — builds on that foundation.
Grounded in wiki concepts agent-fleet, cost-pool, solvency, allocation, customer-dashboard, and the Sovereign-stack business series. Design notes on a running system.
