Meta is testing an AI agent in Ads Manager: the guardrails you need now

Meta wants you to run campaigns through a chat window. At Advertising Week New York on October 6, 2026, the company announced that its AI assistant in Ads Manager will no longer just make recommendations but act on them: creating campaigns, changing audiences and adjusting budgets. For now it is a test. For performance teams, it is still the right moment to write down their own rules for AI agents.

Meta is testing agentic capabilities in its AI business assistant inside Ads Manager: it is meant to create campaigns, change targeting and adjust budgets, alongside scheduled tasks, alerts and the import of brand guidelines. According to coverage of the announcement, this is a test without a fixed launch date. Prepare guardrails now: budget limits, approval levels and a measurement setup that makes agent changes visible.

What did Meta announce at Advertising Week?

Meta summarised its updates in a Meta for Business post on Advertising Week New York 2026, and several trade outlets covered the announcement on October 6 and 7. An analysis by Relevant Audience sorts the features by status, and that distinction matters, because not everything is live.

According to that overview, the agentic actions of the Meta AI Business Assistant are in testing, meaning creating campaigns and changing targeting and budgets, along with scheduled tasks and alerts and the import of strategy, tone of voice and brand guidelines. Generally available are the Customer Lifecycle Strategy, which lets you steer audiences by new versus existing customers, and video from static images in Advantage+ Creative. Ads Creative Studio is rolling out to more advertisers, and AI-assisted audience discovery in Detailed Targeting is expected by the end of the year. Meta has also expanded its ads MCP server, which lets external AI tools create campaign drafts in Ads Manager. According to Adgully, permission controls and human approval before publishing remain in place.

What does this mean for budget and control?

Agents shift work from execution to oversight. Until now the bottleneck was time spent on setup and optimisation. Next, the question becomes which decisions you delegate to an AI and how you notice when it gets them wrong. An agent that moves budget "in response to performance signals" optimises for whatever signal it can see. If the account only has pixel conversions without lead quality, it will fund the cheapest forms, not the best customers.

A quick calculation shows the risk. Suppose you run two ad sets: A at 40 euros per lead with a 25 percent close rate, and B at 25 euros per lead with an 8 percent close rate. On cost per lead, B looks better. On cost per customer, A costs 160 euros and B roughly 313 euros. An agent without closed-deal data would push budget in the wrong direction.

Guardrails to set now

1. Fix your signals: Send lead quality or revenue back via the Conversions API. Without that data, any automation optimises for the wrong goal.
2. Define budget limits: Set daily and campaign caps that an agent may not exceed, and document them.
3. Agree on approval levels: New campaigns and audiences only with human sign-off, budget shifts within fixed corridors automatically.
4. Log changes: Review the Ads Manager change history weekly and flag what the assistant changed. That is the only way to attribute results.
5. Prepare brand guidelines: When guideline and tone import arrives, have a short, unambiguous document ready.

What not to do

Do not switch everything on at once when the test reaches your account. Start with a clearly ring-fenced part of the budget, compare it over several weeks with a manually managed control group, and judge it on cost per qualified lead or contribution margin. And do not rely on vendor figures without a transparent method: Meta says users of its recommendations typically see 5 percent more conversions, but according to Relevant Audience it does not disclose how this was measured.

How to run and measure paid social properly is covered under performance marketing and tracking and analytics. How AI agents take over processes beyond ads is covered under AI automation.

FAQ

Can Meta's assistant already create campaigns today?

According to current coverage, the agentic actions are being tested and no general launch date has been given. Availability may vary by account and market.

What is the ads MCP server?

An interface that lets external AI tools work with Ads Manager, for example to create drafts, generate previews and pull reports. According to Adgully, human approval before publishing is still required.

Which feature is worth trying right away?

According to the coverage, the Customer Lifecycle Strategy is available to all advertisers. If you want to prioritise new customers, you can test it now, provided your customer data is clean.

Sources

• Meta for Business: Advertising Week New York 2026
• Relevant Audience: Meta Advertising Week 2026 Ad Tools, Live vs Testing
• Adgully: Meta unveils agentic AI capabilities in Ads Manager (October 7, 2026)

Want more of the attention you get to turn into customers? Let’s spend 15 minutes on your setup, free and without a pitch deck. Book your free call