What is the pricing unit?
Work, metered in tokens. A plan on ftes.ai buys a block of tokens for the organization, and every Digital FTE the organization runs draws from that one block as it works. There is no per-seat price, no per-role price, and no charge for the number of people who log in to manage roles. A role that does more work in a cycle consumes more of the block; a role that is paused consumes nothing. The plan's price divided by its tokens is the rate used to compute cost per outcome.
The glossary entry on token metering and the whitepaper on the economics of digital labor explain the model in full; this hub answers the questions.
What bounds spend?
Three things. The wallet: the block already purchased is the ceiling, and when it reaches zero every role pauses; nothing runs on credit and no overage is billed. Per-role budgets: inside the block, each Digital FTE has a budget set in its Setup tab, and a role reaching its budget pauses while others continue. Per-mission budgets: a mission, a bounded piece of strategy or diagnostic work, carries its own budget inside the role's. Spend against budget appears in every morning brief and in the workforce view, so it is seen daily rather than discovered at the invoice.
The CFO persona piece describes the five questions a finance leader asks about bounds; these are the answers.
Is there a free trial?
No separate trial tier. The smallest plan is the way to try a Digital FTE: it buys a block large enough to run one role through its first cycle, it can be changed or cancelled from the workspace at any time, and the first-run wizard, the remit, the connections, and the record are the same as on any plan. A team that wants to try one function hires one role on the smallest plan, runs its thirty-day playbook, reads the first review, and decides. The pricing page states the current plans and blocks.
The founder persona piece describes the smallest-plan path from the founder's seat.
Can I see what each role costs?
Yes. Usage is reported per Digital FTE in its Performance tab, per cycle, alongside the outcomes in its Targets tab, so each role's cost per outcome can be computed on its own against a baseline recorded before the pilot. The workforce view shows combined usage across roles against the wallet, which is how the next block is sized. Nothing is blended: a Finance Controller FTE's tokens and a Support FTE's tokens are separate lines, because a reconciled transaction and a resolved ticket are different outcomes.
The metric explainer on cost per outcome gives the arithmetic with the inputs left blank.
What is not in the price?
Three things a budget should carry separately. External platform spend a role controls, above all ad budgets for a Performance Ads FTE, which the ad platform bills under a cap set at the platform that the FTE cannot raise. The human time: approval minutes, review hours, setup hours, and governance hours, at the rates of the people who spend them, which the baseline sheet logs. And any tool subscriptions a role uses through a connection, which the tool bills. The hidden-costs piece in the library lists what to look for in any vendor's number, including this one's.
A budget line that omits human hours is the one a CFO sends back first.
How do plans change, and what happens at zero?
Plans are changed from the workspace by an owner or admin, logged, and the new block applies from the next cycle; a mid-cycle top-up follows the same path. At zero, every role pauses: proposals stop, scheduled work waits, and the morning brief says so. Nothing is lost; the record, the memory, and the connections persist, and work resumes when the block is renewed. Bringing your own model provider key is supported, encrypted like any credential, so model usage can run on your own provider agreement with the gates unchanged.
The spend governance piece describes how to set per-role budgets so that zero is never reached by surprise.
Does it get cheaper as a role earns autonomy?
The tokens do not change much; the human hours do. A class promoted under the Promotion Standard runs without per-item approval, so the manager's minutes fall, and cost per outcome, which includes those minutes, falls with them. Token usage per outcome tends to settle after the first cycle, which includes setup and close reading, which is why the second cycle is the honest comparison. The myth piece on AI employees being cheap because they are AI sets out what the full cost includes and when the answer is not the AI employee.
A pilot judged on its first cycle's cost is judged on its setup.