> Source: https://builder.ema.ai/v2/getting-started/create-your-first-ai-employee
> Title: Create Your First AI Employee

# Create Your First AI Employee

This walkthrough takes you from an empty AI Employees page to a published, working AI Employee (AIE). You'll create a Chat AI Employee that answers questions from documents — the most common starting point — by beginning from a template, reviewing its workflow in the AI Employee builder, testing it, and publishing.

**Time:** about 15–30 minutes. **You'll need:** a builder role (the "Build an AI Employee" button only appears for `builder`, `builder_admin`, `system_admin`, or `env_admin`), and, optionally, a document or two to use as a knowledge source.

> [TIP]
> **Prefer to describe it instead of building it?** Open [Autopilot](/builder/v2/autopilot) and tell it what you want — it can create the AI Employee, build the workflow, and set up a knowledge base for you. This page covers the manual path so you understand what Autopilot is doing.

## Step 1 — Start a new AI Employee

1.  From the sidebar, open **AI Employees**.
2.  Select **Build an AI Employee** at the top of the page. This opens the **Create AI Employee** picker.
3.  The picker shows two ways to begin:
    -   **Choose a template** — browse curated templates by category (Support, Sales, Marketing, HR, IT & Ops, Finance, Other) or search by keyword. Each card shows a badge for its interaction type and category.
    -   **Start from scratch** — the first card in the grid. Select it to build a custom AI Employee from a blank workflow.

### Option A — Use a template (recommended for your first build)

1.  Select a template card to open its detail view. Review the overview and, where available, the **Configuration** tab showing the workflow and recommended data sources.
2.  Select **Use template**. Ema creates the AI Employee and opens it.

Starting from a template gives you a working workflow you can run immediately, then customize.

### Option B — Start from scratch

1.  Select **Start from scratch**. The create dialog opens.
2.  Fill in the fields:
    -   **Name** (required) — for example, "Policy Assistant."
    -   **Description** (required) — a short summary of what it does.
    -   **Group** (optional) — a category to organize the AI Employee on the dashboard (for example, HR or Knowledge). You can pick a suggested group or type a new one.
    -   **Icon** — pick an icon for the card.
    -   **Interaction type** — choose **Chat**, **Dashboard**, or **Voice**. For this walkthrough, choose **Chat**. (See [AI Employee Types](/builder/v2/getting-started/ai-employee-types) for what each does.)
3.  Select **Create**.

## Step 2 — Land in the right place

What happens next depends on the interaction type:

-   **Chat** and **Dashboard** AI Employees open directly in the **AI Employee builder** on the workflow editor, ready to edit.
-   **Voice** AI Employees open on the AI Employee's detail page, where the **Voice Configuration** tab lets you configure the voice agent. (Voice is in beta.)

For a Chat AI Employee, you're now looking at the workflow — a directed acyclic graph (DAG) of typed nodes.

## Step 3 — Understand the workflow

An AI Employee's workflow is built from a small set of node types:

Node type

What it does

`start`

The workflow's entry point. Defines the input the AI Employee accepts (for a Chat AIE, the user's message).

`agent`

A reasoning step that calls an LLM. Each agent has a type — for example `search_respond` (retrieve and answer), `intent_classification` (categorize the request), `extraction` (pull structured fields), `rule_validation` (check against rules), or `custom`.

`transform`

Reshapes data between nodes by mapping fields.

`publish`

Marks a value as the workflow's output — what the AI Employee returns to the user.

`end`

The workflow's terminal node.

A basic retrieval-and-respond Chat workflow connects a `start` node to a `search_respond` agent, then to a `publish` node:

```text
start → search_respond (agent) → publish → end
```

The `search_respond` agent retrieves relevant passages from a knowledge base and writes an answer grounded in them, with citations. If you started from a template, this structure is already in place.

> [INFO]
> The deep mechanics of nodes, agents, conditions, and the editor live in [Core Concepts](/builder/v2/core-concepts) and [Builder Guides](/builder/v2/builder-guides). This page keeps to what you need for a first working build.

## Step 4 — Connect a knowledge base

To ground answers in your own content, give the AI Employee a knowledge base:

1.  Open the **Knowledge bases** tab on the AI Employee.
2.  Create a knowledge base and add sources — upload files or connect a data source through an integration.
3.  In the workflow, make sure your retrieval agent (the `search_respond` agent) is configured to use that knowledge base.

Knowledge bases and ingestion are covered in depth in [Integrations & Data](/builder/v2/integrations-data).

## Step 5 — Configure the agent's instructions

Open the agent node and review its **Instructions** — the natural-language guidance that shapes how it responds. For a knowledge assistant, you might add:

-   "Answer only from the provided sources; if the answer isn't there, say so."
-   "Keep responses concise and cite the source for each claim."

Instructions are how you tune behavior without changing the workflow's structure.

## Step 6 — Test it

Before publishing, confirm the workflow behaves as expected:

-   **Chat AI Employee** — open the **Chat** tab and send a message. Ema runs the workflow live and streams back the answer with citations. Try a question your knowledge base should be able to answer, and one it shouldn't, to check both paths.
-   **Dashboard AI Employee** — run the workflow against a sample input and inspect the resulting row, including its per-row execution history.

Iterate on the instructions and the knowledge base until the responses are right.

## Step 7 — Publish

When you're satisfied, publish so the AI Employee is live for the people you share it with:

1.  In the AI Employee builder, select **Publish**.
2.  Confirm in the publish dialog.

Publishing creates a new version. You can review and restore earlier versions later from the version history.

## Step 8 — Share access

Open the **Permissions** tab to control who can view, run, and edit the AI Employee. Grant the right people access so they can find it on their AI Employees page. Permissions and roles are detailed in [Administration](/builder/v2/administration).

## What to do next

You now have a published AI Employee. From here you can:

-   **Add branching** — route the workflow on conditions (for example, by intent) so it handles different request types differently. See [Core Concepts](/builder/v2/core-concepts).
-   **Connect Tools** — let the AI Employee read from and write to external systems through integrations. See [Integrations & Data](/builder/v2/integrations-data).
-   **Evaluate quality** — measure response accuracy and track it over time. See [Testing & Operations](/builder/v2/testing-operations).
-   **Hand the heavy lifting to Autopilot** — describe a change in plain language and let [Autopilot](/builder/v2/autopilot) make it.