AI employee vs. human hire.
The best teams run both: an AI employee takes the repetitive part of the desk while humans move up the value chain.
Swipe to compare →
| Dimension | Human hire | AI employee |
|---|---|---|
| Time to productive | 2–8 weeks | 1 day |
| Availability | Business hours | 24/7 |
| Loaded monthly cost | $6,062 modelled | From $149 |
| Turnover risk | High | None |
| Scalability | Hire again | Duplicate in minutes |
| Judgment & creativity | Excellent | Escalates to humans |
The monthly cost row is a model: a 40-hour desk at a loaded $35 an hour across 4.33 weeks a month. The calculator runs it on your own numbers; $149 is the published Starter price.
Frequently asked questions
For a desk that is repetitive, measurable and already documented, an AI employee starts at $149 a month and is productive in a day, so the decision is cheap to reverse. For work that turns on judgment, negotiation or a relationship, hire the person. Most teams end up running both on the same desk: the AI employee takes the routine share, the person takes the exceptions.
One AI employee is $149 a month on Starter and $120 a month each at five on Scale. A full-time desk models at $6,062 a month at a loaded $35 an hour across 4.33 weeks, and the routine share of it — the part an AI employee takes — models at $4,243. The human figures are model assumptions you can change on the ROI calculator; the price is the published list.
You pick the role, brief it in plain language, and point it at the material it works from. Most roles go live in a day, and Starter opens with a 14-day pilot so you can read every draft before anything is sent. No code and no engineers are involved.
Anything that has to be decided rather than drafted. It moves no money, sends no outbound message on its own, and escalates any action its rulebook has never seen. Judgment, creative direction and the relationship part of a deal stay with a person, and the role is built to hand those over with the thread attached.
That is the usual arrangement. The AI employee works the queue — reading, scoring, drafting, logging — and the person reviews, sends and handles the exceptions. Every run lands in a journal with its inputs and its outcome, so the split is visible rather than assumed.