Opinion. This week, two platforms moved buying, execution and business administration into agents that read the tools marketers already run. I think the approval step is the one thing a marketing team should refuse to hand over.
What shipped
Meta’s Muse for Small Business connects its agent to Klaviyo, Shopify, Stripe, Slack and Facebook and Instagram business accounts, among others. Google’s Vertical Video Unification puts Gemini behind “buying, execution, and full-funnel measurement” in Display & Video 360. Meta also announced an Enterprise Platform to take the Muse agent, Meta Business Agent and Muse API to businesses and developers.
I read each of these as a reasonable product decision. I also read them as one direction of travel, and marketing teams should decide how they meet it.
My position
A marketing stack used to be something a team assembled. The email platform sat here, the storefront there, the ad accounts somewhere else, and a person moved between them. A connector-based agent reverses that. The agent sits on top, and the tools become sources it reads and actions it takes.
Once that happens, three things a team owned by default turn into choices. Which tools can the agent see. Which actions can it take. Who signs off. I would treat the third as non-negotiable. The approval log is the only record that explains why spend moved, why a message went out or why a page changed. A team without it cannot audit its own marketing.
Meta’s release says “nothing publishes, sends, or spends without your approval.” That is the right default. My argument is that teams should adopt it as their own written policy and not treat it as a product setting a vendor can change later.
The strongest counter-argument
The best case against me comes from the people these tools are built for. Tom Mulholland, owner of Mulholland Grocery in Malvern, Iowa, said in Meta’s announcement: “I work about 65 hours a week, and there are so many things where I’m the only one who can do them.” For an owner in that position, approving every draft can look like the same workload the agent was supposed to lift. Meta also says Muse is free for most of what people need, so the cost of trying it is low and the pull is real.
I accept the point about hours. It does not change my answer, because reviewing a draft is a different job from producing one. An owner who reads a campaign draft and taps approve has kept the decision and dropped the labor. The policy I am arguing for costs a look at each action. It does not cost the owner the work the agent takes over.
The enterprise version
Enterprise buyers face the same question at larger scale. CJ Desai, Meta’s Chief Enterprise Platform Officer, said: “Our goal is to make Meta the place enterprises come to scale their businesses.” That is a plain statement of strategy, and a sensible one for a company that already serves millions of advertisers. It also means a marketing team may soon see one vendor on the media plan and on the software contract.
The Enterprise Platform announcement says security and privacy are built in from the outset. It does not say whether the approval default carries over to enterprise accounts. That is a contract question, and marketing should be the one asking it. Google’s announcement raises the same point from another side, since it credits Gemini with buying, execution and measurement in a single platform. When one system plans, runs and grades a buy, the person approving it is the only independent check left.
What to do
Write a connector policy before anyone connects anything. List the tools an agent may read and the tools it may act in, and name who can add to either list.
Put spend, publishing and customer messaging behind approval, and require the approval to be logged somewhere the team can export. Ask each vendor in writing how that log is kept and whether the default can be changed without your sign-off.
Keep customer data in a system you control and let agents read from it, so that removing a connector does not remove your history. Our earlier argument on this point is in AI Agents Need Trust Layers, Not More Access, and the deployment side is in GTM Teams Deploy AI Agents Faster Than Governance.
The agents will keep getting better at the work. The approval step is where a marketing team still decides, so write it down now.
Source: Meta Newsroom