Talking to the AI agent

The AI agent is where you talk to your campaigns. There is no separate query language or filter to build: you type the question, the agent looks at your data and answers with the numbers attached.

The agent only sees your workspace. It cannot read another workspace’s campaigns or another client’s spend — that is not an instruction but a boundary enforced in the data layer.

The kind of things you can ask

• What changed compared with last week?
• Which campaign is losing money?
• Lower the budget of this ad set.
• Review my catalog health and show the lowest scoring products.
• Produce three ad images for this product.

The steps under the answer

While producing an answer the agent lists the tools it called: “Read metrics”, “Campaign list”, “Catalog health” and so on. You can open those lines to see what it asked and what came back. That is where to look when you wonder where an answer came from.

Conversations and quota

Every conversation is saved and can be returned to: if you want to approve a suggestion later, it is still where you left it. Starting a new conversation for a new topic keeps the agent from mixing up context.

You have a daily message allowance and the header shows how many are left. When it runs out the chat is locked until the next day; reading screens are not affected.

The work the agent actually applies, and the approval gate, are covered in their own article.What the agent can do