Meta's Ads Manager assistant moves from advice to agent, with approvals attached
At Advertising Week New York, Meta began testing agentic actions — creative, targeting and budget changes on a prompt — inside the Ads Manager business assistant.
On October 6, Meta used its Advertising Week New York slot to begin testing agentic actions inside the Meta AI business assistant in Ads Manager, converting what had been an advisory layer into something that can draft creative, adjust targeting, and move budgets from a conversational prompt. The pitch, per Meta’s own framing to Relevant Audience, is that marketers will “soon” be able to plan and create a campaign, generate imagery and text, update targeting or change a budget conversationally, all gated by “previews, approvals and controls.”
The scope of “soon” is doing a lot of work. Relevant Audience notes that agentic actions, scheduled tasks and alerts are each labeled as beginning to test, with no country list and no rollout date. Audience discovery inside Detailed Targeting is slated to arrive by year-end.
What’s actually new is memory. Social Media Today reports the assistant now carries “the context of what you tried last quarter, your seasonality, and the goals you’ve described,” and Meta is testing imports of brand voice and strategy guidelines. The 5% more conversions at the same cost per conversion that Meta cites applies to the older advisory product; Relevant Audience flags that no method, period, or vertical was published for the comparison.
The infrastructure story runs underneath. Meta’s Ads MCP server now has ten partner integrations, including Perplexity, Snowflake, and Triple Whale, letting external agents stage work as drafts inside Ads Manager under advertiser permissions. It’s a parallel small-business bet to Meta’s push through the Muse creator agent.
The governance gap is the quiet part. AIxH recommends written thresholds, budget ceilings, delegated scope, and a weekly change log before any account enables the test. Social Lady is blunter: Meta has published no spending cap, no mandatory approval step, and no rollback for assistant-initiated changes. A manually set account spending limit is the only hard cap. The approval layer is a design choice until it’s a documented policy.