# Your first conversation

> **What you will learn**
> This is a chapter for **doing**, not reading. Follow the three below exactly.

## 1. Just talk to it

Entering a workspace drops you into a chat box. Type this:

```
Hello. I'm a (write your role here, e.g. marketing manager).
Tell me 5 things I could do in this workspace.
```

### What to check

- Whether the answer is **fitted to your role**
- Whether it appears character by character in real time

---

## 2. Make it use a tool

Now a request that needs a tool.

```
Calculate how many days are left in this quarter as of today.
```

Or, if a web tool is enabled:

```
Find how the phrase "AI Transformation" is being used recently
and summarise in 3 lines. Include sources.
```

### What to check

Look for **the tool call shown above the answer.**

```
┌─ Tool call ──────────┐
│ web_search("...")     │
│ → result              │
└──────────────────────┘
```

That is **the answer's basis** from [the threads chapter](/guide/cn-thread). It
is the point of difference from a consumer AI service.

> If you see no tool, no tools are enabled yet. Covered in
> [turning tools on and off](/guide/cn-tool-settings).

---

## 3. One real piece of work

Pick whichever is closest to your situation.

### If you have documents to organise

```
Turn the meeting notes below into a table.
Columns: Topic / Decision / Owner / Deadline.
Do not guess at anything not present in the source.

(paste the notes)
```

**The last line matters.** Because
[filling in a blank is the natural continuation](/guide/ai-hallucination), you
have to block it explicitly.

### If you have an email to write

```
Draft an email for the situation below, in five sentences or fewer.

To: (e.g. a supplier contact)
Situation: (e.g. I sent a quote last week and haven't had a reply)
```

### If you have research to do

```
Research (topic).
- Always attach a source (URL and date checked)
- Leave anything unconfirmable as "unconfirmed"
```

---

## If you do not like the result

**Keep talking.**

| Complaint | Say this |
|---|---|
| Too long | "Cut it in half" |
| Wrong format | "Make it a table" |
| Doesn't fit us | "We're (situation), so that won't work. Again" |
| No basis given | "Give me the source for each item" |

---

## Once you have done this

You have the basics. The next step is **reducing repetition.**

If you keep making the same kind of request, rather than rewriting the
instruction every time, [turn it into an agent](/guide/cn-agents-intro).

## Care when using it with a team

Threads live **inside the workspace, so teammates can see them.**

```
□ do not put customer personal data in conversations
□ do not put passwords or API keys in
□ split workspaces for sensitive subjects
```

The principles from [things to watch out for](/guide/ai-cautions) matter more in
a team environment.

---

## Check yourself

**1. Why should you check that tool calls are displayed?**

<details>
<summary>Answer</summary>

**Because it shows the answer's basis.** Knowing which tool a value came from
makes verification possible, and when something is wrong you can see which step
was wrong.
</details>

**2. Why add "do not guess" when asking for a table?**

<details>
<summary>Answer</summary>

**Because filling in a blank is the natural continuation**, so it invents
something. You have to block it explicitly.
</details>

**3. What if you keep making the same kind of request?**

<details>
<summary>Answer</summary>

**That is a signal to build an agent.** Put the instruction inside it and you stop
rewriting it every time.
</details>

---

Now for agents → [What agents are](/guide/cn-agents-intro)