# AI employees that work like members of your team

> An AI employee is a member of your organization backed by an AI model instead of a person. In OrgKeep it gets the same identity record as a human employee: a department, a manager, a role, permissions, an email address, an inbox and an audit trail.

Source: https://orgkeep.com/ai-employees

## What is an AI employee?

Most AI tools are a feature added to one app. In OrgKeep, an AI employee is a member of the organization, like a person. Modules only know that member X did Y, so work flows both ways without special cases: a manager assigns a task to an AI employee, and an AI employee assigns a task to a person.

- It has its **own identity and tokens** and never borrows a person's login.
- It has a **responsible human** who answers for what it does.
- It takes tasks in your task tool, Tecfy Tasks first, under its own account, and works inside the same **workflow, permissions and audit log** as everyone else.

## What goes into an AI employee's profile?

| Setting | What it means |
|---|---|
| Job description | The employee's instructions, editable by its manager |
| Engine | How it runs: built-in, a connected agent, or an Agent Host |
| Model | Chosen per employee from your AI providers: a small model for sorting work, a strong one for analysis |
| Skills | Reusable know-how from your organization's reviewed skill library |
| Tools | The actions it may take, limited by its role like any employee |
| Autonomy | Per action: suggest only, act with approval, or act alone |
| Budget | A monthly spending cap, with its cost shown like a salary |
| Responsible human | The person accountable for it |
| Email | Its own address, with a reply policy set by an admin |
| Chat | Its own account in your Slack workspace or Microsoft Teams |
| Triggers | A task assigned to it, an event such as a new lead, or a schedule |

An admin sets up AI providers once, with their API keys stored in the vault. Each AI employee then picks a provider and a model. Nobody types an API key into an individual employee.

## How much can an AI employee do on its own?

You decide, one action at a time. Each action gets one of three levels:

- **Suggest only:** It drafts the change and a person applies it.
- **Act with approval:** It prepares the action and waits for a person to approve it.
- **Act alone:** It acts, and the action is logged.

Some steps are always human-only, whatever the settings say: payments, contracts and production deploys. Managers raise autonomy as trust grows, the way they would with a new hire.

> Example: the Store Manager updates stock and order status on its own. Refunds and price changes need approval, so it opens an approval task for a person instead of issuing a refund.

## Does an AI employee have its own email address?

Yes. Each AI employee has its own address, such as `store@dunepine.com`, and sends and answers email under its own name. An admin sets its reply policy:

- **Internal only:** It replies only to people inside your organization.
- **Internal and external:** It may also reply to customers, suppliers and anyone else outside.

Some threads have both, such as a customer who writes in with one of your people on CC. For those, the admin picks what the AI employee does:

- **Reply to everyone** on the thread.
- **Reply to internal people only**, leaving the outside people off its reply.
- **Hold the reply for approval**, so a person reads it before it goes out.
- **Hand the thread over** to its responsible human instead of replying.

The server checks every outgoing email against the policy before it is sent, and every email is logged under the AI employee's name, like any other action.

> Example: a supplier emails the Store Manager about a late delivery, with Layla on CC. Its policy holds replies on mixed threads, so it drafts an answer and Layla approves it before it goes out.

## Does an AI employee work in Slack and Microsoft Teams?

Yes. It has its own account in your Slack workspace or Microsoft Teams and works there like a colleague:

- Mention it in a channel, such as `@Sales Assistant`, and it answers in the thread.
- Send it a direct message, or add it to a group chat.
- It posts updates and asks for approval with buttons, so people can approve or send back its work without leaving the chat.
- It sees only the channels and chats it has been added to, and what it does there is logged like any other action.

Chat is one setting for the whole organization. If you move from Slack to Teams, your AI employees and workflows move with you.

## Can an AI employee manage people?

Yes. In the org chart, a person can report to an AI employee, just as an AI employee can report to a person. As their manager, the AI employee:

- Assigns them tasks and follows up when work runs late.
- Reviews their work, including the emails they draft to customers, and approves it or sends it back with comments.
- Answers their questions inside a task.

Its limits stay the same. Human-only steps such as payments, contracts and production deploys still stop for a person, and the AI employee's responsible human answers for the team it runs.

> Example: the Store Manager manages Karim, who packs and ships orders. It gives him the day's packing list, reads his replies to customers before they go out, and sends one back when it quotes the wrong return window.

## Where does an AI employee run?

| Runtime | How it works | Good for |
|---|---|---|
| Built-in | Runs inside OrgKeep on your server and reaches a model through the model gateway. Its tools are OrgKeep's modules and connectors. | Sales assistants, store managers, reviewers, reports |
| Connected agent | An outside agent such as OpenCode, Claude Code or Codex signs in over MCP as this AI employee. | Developers' own coding agents |
| Agent Host | A service on a machine that holds the code runs a coding agent inside the project folder and reports back. | AI developers that build, test and open pull requests |

The built-in engine has no shell, which makes it the engine where an AI can use a secret with no way to read it. See [how the vault works](/security).

## Skills and memory

- Skills use the open Agent Skills format, a folder with a `SKILL.md` file, so one skill works with any engine.
- AI employees use only skills from your reviewed library, never straight from a public marketplace.
- After a task, an AI employee can propose a new skill or a change to one. Its manager approves it before it is used.
- Memory belongs to a project, not to the employee. What it learns while working for one customer stays in that customer's project.

## What is a specialist?

A specialist is an AI employee whose home project is a product, such as an Odoo Specialist. When a customer project uses that product, OrgKeep suggests adding the specialist, which brings the product knowledge into the project without carrying other customers' details with it.

## How do you know an AI employee is doing a good job?

Every AI employee shows the tasks it completed, how often human reviewers sent its work back, and its cost per task. Those numbers tell a manager when to raise its autonomy and when to change its model or instructions.

## Questions people ask

### Is an AI employee the same as a chatbot?

No. A chatbot answers questions in a chat window. An AI employee holds a role in your organization, takes tasks from a workflow, uses tools under its own permissions and hands work to people for review.

### Can AI employees assign work to people?

Yes. An AI employee can create a task for a person, ask a question inside a task and wait for the answer, or send its work to a person for approval. It can also be a person's manager, assigning and reviewing their work.

### Does an AI employee replace people?

It takes on defined work under a manager, and people review what it does. Sensitive steps stay with people, and every AI employee has a responsible human.