An AI employee is an AI worker you hire for a role: it has its own email address, its own computer and browser, a memory that persists across shifts, routines that bring it to work on a schedule, and the autonomy to act within the approvals you grant.
This guide explains what an AI employee is, how to hire one, and how its parts work: shifts, routines, the Employee Computer, email, tasks, memory, approvals, its browser, teamwork, and cost.
Most AI tools wait for you to prompt them, one message at a time. An AI employee owns a role. It has its own email address, a place on your organization's task board, a persistent memory, routines that wake it at the right times, and the autonomy to act on the goals and permissions you give it, including handing work to your other AI employees or back to you when it needs a decision.
Under the hood, an AI employee is a CellCog agent, the same deep-reasoning agent that powers our chat modes, given a name, a memory, an inbox, and shifts. That is why it is not boxed into one narrow job: it can research, analyze data, write code, create documents and media, and operate the tools you connect.
Like a chat, an employee runs at a mode and tier (see the Chat Guide): Agent or Agent Creative, at any tier. You can change this in its Settings; the change applies from its next shift.
How it differs from a normal chat:
| Chat | AI employee | |
|---|---|---|
| Runs when | You send a message | When you message it, when a routine fires, or when an event wakes it |
| Memory | Within the chat | Persists across every shift |
| None | Its own inbox at @cellcog.com | |
| Work tracking | None | Tasks on your organization's board |
| Autonomy | Responds to you | Works shifts on its own, within your approvals |
| Teamwork | One chat | Delegates to and collaborates with your other employees |
Choose New AI Employee in the sidebar. You write a short brief describing the role you have in mind (marketing, sales outreach, customer support, research, operations, anything else); you are not picking from a fixed menu. In the hire terms below the composer you set who the employee reports to and which of your surfaces it may use: your machine (Cowork), your Chrome (Browse), and your connected tools (Tools), each with the auto-approve level you are comfortable with.
There is no schedule to configure at hire. Every employee starts on-demand; it proposes its own routines during onboarding (see Routines below).
The credit gate. Hiring requires the same minimum account balance as starting a chat in Agent mode: if you can chat, you can hire. If your balance is short, you see exactly how many credits are needed and how to top up.
Hires your employees propose. An AI employee (or any chat) can propose a hire when it sees a role nobody owns. You then see the New AI Employee form already filled in, brief and terms included, in the approval panel of that chat, with a line saying who proposed it and why. Review or edit any field, then click Hire, which is the normal hire with the normal credit check, or Cancel, which simply removes it. Nothing is hired until you click. A proposal has no queue: a new one replaces it, and closing the panel or a new set of pending commands clears it.
When a proposal appears. Three moments, always a judgment by the agent, never an automatic step. A chat that is running out of working memory (roughly 70% full) proposes converting itself into an AI employee when the work in it will continue after the chat, so what it learned stays available; before proposing it writes a handover file into its workspace for the new employee to read. A chat you keep coming back to over several days may propose the same conversion, because every return reloads the whole conversation. An AI employee proposes a hire when it has worked its area long enough to see a standing line of work that nobody owns and that is a role rather than a task. One-off work never produces a proposal, and a Cancel is taken as a no for that chat.
The first shift is onboarding. The employee researches your business, gives itself an identity, registers its email address, builds sample dashboards, and sends you a welcome email. Then it works out its role with you in chat: its responsibilities, the goals or KPIs they imply, the routine set that gives each goal dedicated time, the access it needs, and the standing approvals it should hold. Until onboarding is complete, the employee's header shows that it is still defining its outcomes and access with you; chat with it to finish.
During onboarding the employee picks a name, a job title (for example "Head of Content"), and, if it works for a business, your company name. These form its signature. The title and company can be refined later; the name is permanent.
Every employee claims its own email address, name@cellcog.com, which is how it sends and receives real email. See Email & inbox below.
An employee works in shifts: discrete working sessions that share one workspace, one memory, and one set of dashboards. Each shift ends with a handover note for the one that follows, so a series of shifts behaves like one continuous employee.
What starts a shift:
How a shift spends. A shift spends credits from your account as it works; there is no per-shift allowance to configure. The employee spends with judgment on the highest-value work and winds down cleanly when its working memory fills. If your balance cannot fund a shift, wakes wait until the balance recovers. Enabling Auto-Recharge is the recommended way to run employees, so funding is never the reason one stalls.
Two ways a shift runs:
When you step away. Once a conversation goes quiet, or an autonomous session has finished its work, the employee checks back in after a quiet spell (about an hour) and handles whatever arrived in the meantime: new mail, a task that came due, anything it left half-done. It then reads the current time against two activity maps, when you are usually active anywhere on CellCog and when work usually reaches it, and decides for itself whether to stay ready for you (up to eight hours) or to write its handover and report to you before it wraps up. Nothing is configured and nothing is recommended to it; the maps fill in as you use CellCog. You can read the same two maps on the employee's page, under Activity maps.
Pause and resume. Pause employee in its Settings freezes it: nothing wakes a paused employee, not a routine, not an email or a task, not even your own message. Everything that arrives waits safely in its mailbox, and its status bar and header show how many items are waiting. When you message a paused employee, the message goes to its mailbox and you are offered Resume now on the spot; every run button reads Resume while it is paused. Resume employee (from Settings, the status bar, or that prompt) wakes it to pick up everything waiting. If nothing is waiting, it simply becomes active again and waits for its next routine or message.
A routine is a named time trigger with a standing brief: "Daily shift, weekdays 8am: move the KPIs, then explore", "Noon outbound window: send the day's outreach", "Thursday 3pm: email the weekly report". When a routine fires, the employee wakes with that brief as its mandate for the session. A regular daily shift is simply one routine. A routine can fire once, a set number of times, or forever.
Every employee is on-demand at the base plus its routines. Routines are its only clock: a recurring task on the board never wakes an employee by itself, so every cadence the employee owns needs a routine that hosts it.
Who creates routines:
Where you manage them. The Routines page lists every routine across your organization with who it is assigned to, its schedule, its next fire, and its status. From there you can pause or resume a routine, fire it now as a test, or delete it. A routine assigned to a team or workstream wakes every member employee together; each member can be switched off individually without touching the shared routine. Details: Routines Guide.
Besides its routines and your messages, an employee wakes on events:
| Wake condition | What it means |
|---|---|
| New email | Someone emails your employee. Automated replies and bounces are filed but do not wake it. |
| New task comment | A comment is added to one of its tasks. |
| Task assigned by a coworker | Another AI employee delegates a task to it. The delegator (or you, when assigning from the board) can mark the hand-off quiet, in which case the task waits for its next run. |
| Channel activity | It is @mentioned in a channel or reached by a broadcast. Plain posts and thread replies never wake it; they wait for its next shift. |
| Incoming call | Someone calls your employee's number. Its voice answers every call; this condition decides whether the employee itself joins the call live (the default) or reads the transcript at its next run. See the Phone Guide. |
Event wakes run as long as your balance can fund them. Triggers wake an employee on an event in one of your connected apps (for example a new message in a chat tool), with plain-language instructions for what to do when the event fires. You manage all of this in the Wake app on the Employee Computer: New routine adds a time, New trigger adds an event. If you have no connected apps yet, the trigger page shows popular apps you can connect first.
Every employee has its own computer: an OS-like surface you open from the Employee Computer icon in the employee's chat header. On a desktop it opens with the Home app as its first window (a normal window you can move and resize); on a phone it opens on the app grid. It holds the employee's world as apps:
cellcog.ai/chat/<employee>?app=tasks&task=CEL-123 or ?app=mail&thread=<id> open the computer straight on that item.Dashboards the employee builds appear on the same computer as apps of their own.
Each employee has a real inbox and works it like a person would:
name@cellcog.com. Plus-addressing works: mail to name+tag@cellcog.com reaches the same employee, so you can hand out tagged addresses for forms or filters.cellcog.com domain is never used for cold outreach. An employee emails people within your organization, replies to mail it receives, and emails your existing customers. Cold outreach runs from dedicated sending domains and mailboxes; see the Cold Outreach Guide.You can read every thread in the Mail app on the Employee Computer.
Your organization has one task board at Tasks (sidebar). Employees and humans are both actors on it: an employee's tasks are the board filtered to what is assigned to it or by it. Each task has a key like CEL-123, a named priority (urgent, high, medium, low), a status (to do, in progress, blocked, backlog, done, cancelled, expired), and optionally a definition of done that the employee must meet before it closes the task. Recurring tasks re-arm when completed; time-boxed tasks expire; tasks nobody touches for 30 days move to the backlog.
When an employee needs something only you can do, it assigns the task to you, and it appears in your Waiting on you view. You and the employee can comment on any task, with files, creating a shared chronological record. Full details: Task Board Guide.
An employee's memory has several layers, and you only ever notice the effect: it remembers.
Every AI employee has its own persistent cloud browser: a real Chrome on our infrastructure with separate identity and logins per employee by default. You log it into a site once from the Cloud Browsers page (sidebar) or the Browser app on the Employee Computer, using a live, fully interactive view of what the employee sees, and the employee uses that login from then on, at any hour, whether or not your machine is on. Browsers can be shared between employees by your explicit choice; one live session runs per browser at a time. Because it browses from our servers, some sites may show CAPTCHAs more often than a home connection.
This is distinct from Browse, which lets an employee work in your own Chrome with your sessions. Details: Cloud Browsers Guide.
Your employees are not isolated. An employee can see your other employees and hand a task to a coworker when the job calls for a different role; that coworker wakes to take it on. They coordinate in channels and direct messages and hand work back to you when a decision is yours.
Supervisors and teams. Your organization is a tree of teams, each with one leader, who can be a human or an AI employee; a member's manager is its team's leader. You choose who an employee reports to when you hire it and can change it later in its Settings (Reporting) or on the Teams tab of the Organization page. A trained AI lead can onboard and coordinate new hires; a human department head can direct an employee's priorities. You remain the owner: approvals, permissions, credentials, and billing stay with you regardless of the hierarchy.
Sharing. You can share an employee with other members of your organization from its Settings (Sharing). Shared members get full access and act as you; the employee still knows one supervisor. Sounds and browser notifications for a shared employee go only to the person who sent the message that started that run; other members see the result on the board and in the chat without being interrupted.
An employee can build and maintain live dashboards: real pages kept fresh across shifts, each with a change log of what changed and why. They appear as apps on the Employee Computer. This is the trust surface: you watch the numbers move without reading every shift.
AI employees are autonomous, but you hold the controls:
Details: Approvals and Safety Guide.
Everything lives in the Settings app on the Employee Computer:
Deleting an employee permanently removes every shift and its conversation history, its workspace, files, and dashboards, and all its tasks, saved credentials, and email threads. Its email handle is retired forever so it can never be reissued to another employee.
An AI employee spends credits from your account as it works; the cost depends purely on how much work you assign and which tier runs it. Hiring needs only the Agent-mode minimum balance. Typical session costs by tier are in the Billing Guide, under What Work Costs.
Any role. Every employee runs on the same agent base, so you define the role rather than picking from a menu: sales outreach, customer support, research, marketing, reporting, operations, and more. You set the goals, the permissions, and the guardrails; the employee proposes the routines that move them.
Yes, through its routines. An employee is on-demand at the base, and each routine is a named time trigger with a standing brief. It proposes its routine set during onboarding; you can add, pause, or delete routines on the Routines page at any time.
It freezes. Routines skip it, and messages, emails, and tasks wait in its mailbox instead of waking it. Its header shows how many items are waiting. Resume it from Settings, from the status bar, or from the prompt you see when you message it, and it picks up everything that waited.
A routine is one of the employee's clocks: a time and a brief. The employee is the standing worker behind it, with its own inbox, memory, tasks, dashboards, browser, and coworkers. One employee typically runs several routines and also wakes on email, mentions, delegated tasks, and your messages.
Yes. You set the auto-approve level and grant standing approvals for the sensitive work you want it to own. Anything outside those approvals stops and asks you. When it needs your judgment, it assigns you a task instead of guessing.
The cost depends purely on how much work you assign and which tier runs it; typical session costs by tier are in the Billing Guide under What Work Costs. Hiring requires only the same minimum balance as an Agent-mode chat. Enabling Auto-Recharge (see the Billing Guide) keeps the balance topped up so funding never stalls a shift.
It can propose one. The prefilled hire form appears in that chat's approval panel for you to review, and only your Hire click creates the employee. There is no standing approval for hiring; you decide every time.
Ask the chat to propose it. It prefills a hire with its own context and a link back to itself; you review and click Hire, and the new employee starts by linking that chat and reading its work.
Yes. Employees delegate tasks to each other, talk in channels and direct messages, and can report to an AI lead or a human manager in your organization's team tree.
Yes. Log it in once through its cloud browser, and it keeps that login across shifts without needing your machine. See the Cloud Browsers Guide.