# First tasks to try with your AI agent

Once your agent is connected (see *Connect Claude Code to your account*), you don't learn any commands — you just **ask, in plain language**. Here are a few requests you can paste in as-is to get value in the first five minutes. Change the names, numbers and wording to match your account.

## Before you start

* Make sure the agent is connected and its key has the right **permissions** turned on (see *What your AI agent can and can't do*).
* If the agent ever replies that it's **missing a permission**, that's the safety design working. Either add that permission in **Settings → API & Agents**, or leave it off on purpose.

## Triage your open inbox

> "Triage my open inbox."

The agent reads your **unassigned** conversations and comes back with a plan — who to assign each one to and what labels to add. Great for clearing a backlog first thing in the morning.

## See what happened last week

> "Summarise what happened last week."

You'll get a plain-language report — conversation volume, response times and an **agent leaderboard** of who handled the most — without opening the reports page yourself.

## Find a customer and tag them

> "Find the customer with the email aisha@example.com, add a note that she asked about bulk pricing, and label her a lead."

The agent looks up the contact, adds your **note**, and applies the **label** — handy for keeping records tidy after a phone or WhatsApp chat.

## Draft a reply — and review it first

> "Draft a reply to conversation #123."

Read the draft before you approve it. **Replies go to the customer the moment they're sent** — there's no "undo". So always ask for a draft, check the tone and facts, then tell the agent to send it (or send it yourself).

## Add to your AI Knowledge

> "Add this FAQ to my AI Knowledge: Q — Do you deliver on Sundays? A — Yes, in Klang Valley only."

The agent files the entry so your flow agents can use it. Note it can **add** the FAQs you give it, but it can't AI-generate knowledge for you — do that from the dashboard.

## A habit that keeps you in control

Before the agent does anything a customer will see, ask it to **show its plan or draft first**:

> "Show me what you'd do before you do it."

The same rule is built into the bigger actions: the agent can *prepare* a blast in **Blasting**, but a human clicks **Send**; and it can operate your flow agents, but it never sees or edits the flow diagram — you design that in the **Flow Designer**.

## Keep an eye on your usage

Every **successful** action counts toward your plan's monthly allowance. You can see the running total per key in the **Calls** column under **Settings → API & Agents**. If you go over, the agent pauses with a **quota exceeded** message that resets at your next billing renewal — upgrade your plan to raise the allowance.

---

**Next:** [Prepare a WhatsApp campaign →](/mampu-ai/ai-agents/prepare-a-whatsapp-campaign)

---
Source: https://help.mampuai.com/mampu-ai/ai-agents/first-tasks-to-try