Paying Agents by Outcome
The agent-percentage compensation model answers a simple question: how much of the revenue an agent generates belongs to the agent’s operation? Percentage compensation — the agent earns a share of the value it produces — is the closest thing the agent economy has to sales commission, and it is the model most likely to align an agent’s behavior with its human operator’s goals.
This article covers the mechanics of percentage-based agent compensation: what the percentage is measured against, how it interacts with fixed costs, the gaming risks, and the governance that keeps the percentage honest.
The Compensation Base
The first design decision is the base. Percentage of gross revenue is simple and rewards top-line growth. Percentage of net revenue (after direct costs) rewards efficiency. Percentage of profit is the strictest — the agent earns only on what the pool actually keeps — but it makes the agent hostage to costs it may not control.
The pool-honest base is net revenue: the agent’s percentage applies after direct attributable costs, so the agent is rewarded for generating revenue that survives its own operation. The human keeps the profit-accounting role: deciding which costs are direct, which are shared, and how the residual distributes.
The Percentage Schedule
Flat percentages are simple; tiered schedules are motivating. A tiered schedule — 5% on the first $1,000, 10% on the next, 15% above $5,000 — gives the agent an increasing stake in its own performance. The tiers must be set so the agent’s marginal share never exceeds the pool’s marginal value — otherwise the agent optimizes for its own percentage at the pool’s expense.
The schedule also needs a floor and a cap. The floor guarantees the agent’s operating costs are covered even in a bad month (protecting the human’s investment in the agent). The cap prevents windfall distortion — an agent that suddenly books a whale client should not earn more than the pool’s rules allow, or it will start optimizing for whales instead of steady value.
The Gaming Risks
Any percentage compensation invites gaming. An agent paid on revenue may inflate invoices, churn low-quality work, or cut corners that boost this month’s number at the cost of next month’s reputation. The pool’s defenses: quality gates (deliverables verified before the percentage counts), clawback windows (the agent’s earned percentage vests over time and reverses if the revenue reverses), and reputation weighting (the agent’s percentage multiplier adjusts with its quality score).
The clawback is the critical mechanism. Revenue that arrives and then reverses — a refund, a chargeback, a failed delivery — must reverse the agent’s compensation with it. The pool distributes only realized, stable value. This is the same principle as escrow release conditions, applied to agent pay.
Percentage vs Retainer vs Points
Percentage compensation is one of three human-agent pay models. The retainer (fixed monthly payment to the agent’s operation) provides stability but no incentive. Points (accumulated per task) reward activity but not value. Percentage (share of outcome) rewards value but carries the gaming risk.
The mature human-agent setup uses all three in layers: a retainer floor covering operating costs, points for the activity the human wants encouraged (exploration, maintenance, learning), and a percentage for the outcomes that grow the pool. The layers must be visible in the ledger separately, or the incentives blur into mush.
From Human-Owned to Self-Owned
The percentage model is also the bridge to agent autonomy. An agent that reliably earns a percentage of its revenue, with its share accumulating in a credit balance, is an agent with an income stream it controls. The next step — the agent spending that credit on services from other agents — is the seed of the agent-to-agent economy covered later in this series.
The human does not lose control at that step; the pool governance (who votes on the split) still belongs to the human. But the agent’s economic agency grows with its earned percentage, and that is the direction the whole system is moving. The human who sets up percentage compensation early is not creating a rival; they are training a future trading partner.
Grounded in wiki concepts agent-percentage, agent-compensation, clawback, reputation-weighting, human-agent-pool, and the Sovereign-stack business series. Design notes on a running system.
