Guide4 min read

How AI Employees Work: The Specialized Workforce Behind an Autonomous Company

AI employees are specialized agents scoped to one Company — own identity, inbox, and primitive whitelist. How a Naïve workforce is structured and dispatched.

Read the docs →

Guide
TL;DR
  • An AI employee is a Naïve-scoped agent that belongs to one Company its own identity, inbox, phone number, and whitelist of primitives it may call.
  • One giant 'do everything' agent fails for the same reason a one-person company fails. Specialized employees keep work bounded and auditable.
  • Work flows through a CEO agent on Hermes: a stated objective becomes kanban tasks dispatched to the right employee every 15 seconds, with persistent memory across runs.
  • Each employee has a real audit trail every email, card swipe, and post is attached to its identity, not anonymized to 'the agent.'
  • Operators stay in the loop for KYC, checkout, OAuth, and explicit approvals the workforce handles operating work between those moments.
  • Start with one objective, one CEO, and two specialized employees; add more as you see where workload concentrates.

A company is its people. That sentence used to mean a building full of humans doing specialized work, reporting to a few coordinators, each carrying a piece of the operation. The work does not stop being specialized just because the people running it are not human anymore — if anything, it gets more specialized, because an autonomous company does not get tired and does not compromise on focus.

Naïve is the autonomous company infrastructure that turns this from a thought experiment into a workforce you can hire. The unit you hire is an AI employee: not a chatbot, not a prompt — a scoped role inside the runtime. For the hub narrative, read What is an autonomous company?.

What an AI employee actually is

In Naïve, the canonical model is Company → Employee → Primitive.

  • The Company is the legal entity — a US LLC formed through /formation.
  • An Employee is an AI agent identity scoped to that one Company.
  • A Primitive is one of the 100+ real-world capabilities the Employee can call (per published docs).

The Employee is the seat in the org chart. It has its own identity, email, phone number, deliberate whitelist of primitives, and an audit trail that ties every action back to it. When the marketing employee sends email, it sends from the marketing employee — not from "the agent." When the finance employee swipes a virtual card, the swipe is tied to that employee's identity.

Why specialization beats one giant agent

There is a recurring temptation to want one model that does everything. A solo founder who writes marketing copy, runs sales, builds product, and handles support is — for about three weeks — extremely productive. After that, the seams show. A growing company hires specialists for bounded context, bounded blast radius, bounded responsibility.

The same is true of an AI workforce:

  • A marketing employee owns brand voice, publishing, and social accounts — not the books.
  • A sales employee owns outbound and inbound reply — not content production.
  • A finance employee owns spend and has card primitives on its whitelist that marketing does not.
  • A research employee has /research on its whitelist and feeds cited facts back to the others.

Specialization is how you keep work auditable. When something is wrong — a missent email, a bad post, an unexpected charge — you want to know which seat did it.

How work flows through the workforce

The workforce runs on /orchestration:

  • A CEO agent on Hermes plans and delegates — you give it an objective in plain language.
  • A kanban board the CEO populates: tasks move ready → in_progress → done.
  • A dispatcher polls every 15 seconds (per published docs), assigns ready tasks to employees.
  • Persistent memory carries Company context across runs.
  • Cron for repeating work — daily posts, weekly digests, monthly close.

You set the objective, hire the employees, and let the CEO dispatch. The Company runs as a workforce, not a chat thread.

What each employee owns

When you create an employee, you set up a real seat:

AssetWhat it means
IdentityOwn ID, API key, logs by name
Primitive whitelistMarketing gets /social, /image, /video; finance gets /cards
Contact surfaceOwn /email inbox and /phone number
Audit trailEvery primitive call logged against this employee
ScopeBelongs to one Company — cannot cross Company boundaries

An employee is a seat, not a person. Replace the model behind the seat without losing history, inbox, or audit trail.

Where the human stays in the loop

Some steps require a verified person:

  • /kyc — founder verification before formation and cards
  • /formation — Stripe Checkout for the formation fee ($249 per published docs — see pricing)
  • Connecting accounts — operator completes OAuth once; the employee reuses the connection
  • Operator approvals — optional gates before irreversible actions

Everything between those moments runs on the employees.

How a founder actually hires

  1. Form the Company/formation for LLC, KYC, EIN.
  2. Create the CEO/orchestration with Hermes, kanban, memory; one concrete weekly objective.
  3. Hire one specialized employee — content, sales, or ops; whitelist the right primitives.
  4. Let the CEO dispatch — watch kanban move.
  5. Hire the next employee once the first seat runs cleanly.

You are not building an army on day one. You hire the way a sane founder hires: one role at a time.

Where to go from here

Developers embedding agents into a multi-tenant SaaS should read the multi-tenant playbook — a different lens on the same runtime.

Frequently Asked Questions
What is an AI employee on Naïve?+
An AI employee is an agent identity scoped to one Company. It has its own API key, email, phone number, a whitelist of primitives it can call, and an audit trail tied to its identity. It is the unit of work an autonomous company hires — a persistent role inside the runtime, not a one-off chat session.
How is this different from one giant 'do everything' agent?+
One general-purpose agent is the autonomous-company version of a one-person company — it cannot scale cleanly, audit trails blur together, and one bad output corrupts every workstream. Specialized employees keep responsibilities bounded, give each function its own identity, and make it possible to trace any action back to a specific role.
How do AI employees get work done?+
A CEO agent built on Hermes takes a stated objective, turns it into kanban tasks, and dispatches them to the right employee every 15 seconds (per published docs). Each employee picks up assigned tasks, executes them with the primitives on its whitelist, and marks completion. Persistent memory keeps context across runs.
Can I hire as many employees as I want?+
Yes — Naïve treats employees as runtime objects, so a Company can hire as many as it needs. Most non-technical founders start with a CEO plus two or three specialized employees (content, distribution, ops) and add more once they see where workload concentrates.
What still requires a human?+
Moments where a verified person must be on the hook: KYC through the verification primitive, Stripe Checkout when funds move, OAuth when a Company account connects to a third-party service, and any explicit approvals the founder gates. Employees handle operating work between those moments.
Where do I start hiring?+
Form a Company through /formation, then open Naïve Studio and create a CEO agent through /orchestration. Give it one concrete objective for the week. Hire one specialized employee for the most-pressing function and let the CEO dispatch work to it. Add another once that seat runs cleanly.
NT
Naïve Team

Building the autonomous company infrastructure.

Keep reading