Agent task demand
—
tasks per hour before approval
Parallel agents / human gate
More agents.
Same approval.
When token throughput scales, command approvals can become the scarce resource. Run a one-hour scenario and see where useful work actually stops clearing.
Waiting for a scenario
Load the sample or enter your own operating assumptions.
Command flow into the human gate
not calculated
demand —capacity markercapacity —
Command demand
—
approval requests per hour
Approval capacity
—
one human gate
Gate utilization
—
demand divided by capacity
Queued approvals
—
after one hour
Effective throughput
—
tasks per hour cleared
The result will explain which side of the system binds first.
Read the model honestly
This is not live agent telemetry or a productivity promise. It is a steady-state capacity scenario: use measured task sizes and approval times before making an operational decision.
This is not live agent telemetry or a productivity promise. It is a steady-state capacity scenario: use measured task sizes and approval times before making an operational decision.
Why token speed can flatten out
Tokens create task demand, but every command crossing a human gate consumes approval capacity. Once that gate binds, faster generation grows the queue instead of completed work.
Change the constraint
Try fewer commands per task, faster approvals, or another approval lane. The useful question is not only “how many agents?” but “which capacity clears the work?”