Should we let AI agents or Meta's agentic tools (Ads MCP, API agents) manage campaigns?
Our answer
Use AI agents to analyze your account and build and upload ads, with a person approving anything that goes live. Don't give an agent control of live budgets, bids or targeting yet.
- Confidence
- Low
- Strongest proof
- 1 controlled test
- 13 Meta rules, 2 datasets, 13 practitioners, 8 opinions or claims
- Sources
- 21 sources
- 37 statements
- M
AFCP
New here? Take the 1-minute tour
Why this is the answer
3 reasons and our judgmentAgencies already run agents on a written playbook, with people checking the work. Common Thread Collective runs Claude over an MCP link guided by its playbook, and its media buyers check the output and add context.
Meta's own tool keeps a person between the agent and live spend. Its ads MCP server creates agent-built campaigns and ads paused, and requires a separate step to make them live.
Meta's planned agentic buying keeps the same check. Nik Ketkar of Meta says advertisers will be able to accept each change before it is made, or ask Meta to decide (Marketing Operators podcast, October 2026).
Our judgment
Keeping budget, bid and targeting changes with a person is our reading of guidance, not of a test. It follows Meta's advice to start an account read-only or approval-required and widen permissions as confidence grows, and its October 2026 note that agent budget and targeting changes are only starting tests with approvals. Revisit monthly, since the tools change fast and this question is on a monthly review. A holdout or A/B test of an agent against your team, judged on contribution margin rather than platform ROAS, would change the answer. That draws on Meta's note that its ads MCP can run A/B tests and lift studies, and on The business metric that governs spend's case for judging spend on margin.
Our judgment, built from Meta: our reading of these sources, not a finding from any one of them.
What the sources say
37 statements from 21 sourcesIn a 17 day Haus geo test, Thrive Causemetics gave 20% of the US the budget moves of Haus's own AI agent. Those regions got 2.4% more new customer Shopify revenue, after shifting spend from AppLovin and TikTok.
“The test ran during a low-spend evergreen period on DTC alone.”
Meta says it put proposed changes to its commercial terms on hold after publishers raised concerns about how they apply to its web crawlers. It plans to consult publishers and regulators in early 2027 before any rollout.
“the proposed updates to our commercial terms of service have been put on hold”
Segwise, quoting Meta's help pages, says Meta's ads AI connectors are in open beta with a phased rollout, new campaigns are created as drafts by default and actions need the user's authorization.
AdExchanger reports Shopify's Campaign Autopilot buys Meta, Microsoft and Snapchat ads for merchants who pay only per sale. Merchants can pay more for new customers, target region or time, and cap or exclude channels.
“takes that media risk”
Meta paused its revised ad and commercial terms that were due on November 20, 2026, with no new date, and says its existing terms stay in effect. It has not tied the pause to recent court or regulator moves.
“In the meantime, our existing terms remain in effect”
Supermetrics says its Claude connector can now create and edit campaigns on six ad platforms with human approval, and more than 6,000 customers have used it since February.
“The goal isn't to remove marketers from the process.”
Meta told Reuters on October 28 that it will not change its terms without a full consultation with industry groups, which could run past November 20; until then its existing terms stay in place.
“Until then our existing terms remain in place.”
Cody Schneider says a coding agent running one startup's Google Ads account cut cost per lead from $1,100 to $250. It is one unnamed account with no time frame or control, and his post links to his own product.
“a coding agent is running the google ads account for a startup we're working with and cost per lead went from $1,100 to $250”
Meta told Reuters it was moving the revised terms from October 30 to November 20 to explain what is changing; PPC Land notes Meta did not promise any edits, only that robots.txt preferences still apply.
“Meta did not say it would amend anything”
Phil Kiel describes an agency using an MCP connection to prepare Ads Manager work under a written SOP: it turns off every Advantage+ creative enhancement, and a buyer approves every change.
“It suggests and prepares. A buyer reviews everything and makes the call. Nothing goes live without their yes, and every change is logged and can be undone.”
Meta's customer story about a one-person shoe brand using Muse for Small Business, which builds ad sets and flags weak spend. It reports no spend or results and does not say if Muse acts on its own.
“Muse builds out ad sets, reads the numbers and tells her when a strategy needs to change or where spend isn't earning its keep.”
PPC Land relays Meta engineering saying a single AI agent hallucinated in production, so Meta replaced it with an orchestrator routing work to 14 subagents, cutting column prep from days by hand to under an hour.
“In production, it started hallucinating”
Meta's Commercial Terms from October 30, 2026 make businesses solely responsible for what its AI features produce, bar removing AI watermarks, and let Meta crawl your product pages for ads and business products.
“You are solely responsible for and assume any and all risks associated with such AI features and any resulting Content”
Meta's Nik Ketkar describes planned, not live, agentic buying. He says Meta's automation rests on transparency and previews, plus control so advertisers can approve each change before it is made. No data.
Ben Feys says AI can analyze retention data to find the top messaging talk tracks that are underrepresented in current ad spend for a brand's top lifetime value persona.
“what are the top five messaging? Talk tracks right now that are underrepresented in ad spend”
A YouTube channel describes Meta's Ads MCP Server, which lets outside AI tools pull reports and manage campaigns, with user confirmation for budget edits.
“It lets outside AI tools pull campaign reports and manage campaigns directly.”
Cody Plofker says Winks gets 60 videos a month from five retained creators, and an AI agent writes copy, picks landing pages and uploads ads to Meta paused for his review. Its first drafts usually need tweaks.
“Tbh, V1 of ads usually need tweaking.”
Ads Uploader counted 36,289 Meta ad-creation batches in September 2026: 17.5% came from AI agents, mostly adding ads to existing campaigns. It shows adoption only, with no performance comparison.
“the monotonous part, building the ad sets and writing the ad variations”
Meta says agentic actions in its business assistant are only in early testing, with previews and approvals for what to delegate. For agencies, work via the Meta Ads MCP server arrives as a draft in Ads Manager.
Meta announced in October 2026 a wider rollout of Ads Creative Studio, conversational campaign tools in testing, and an ads MCP server where agent work lands as a draft needing approval before going live.
“Every action is gated behind advertiser-set permissions and explicit approval before anything goes live”
Jess Bachman says she does not trust AI to analyze ad performance data.
“I also do not trust AI with analysis”
Meta says Muse, its AI agent in the US and Canada, can connect Instagram analytics, Pages and Meta ad accounts and draft campaigns, with nothing spent or published without approval. No results.
“You're in control: nothing publishes, sends, or spends without your approval.”
Meta says its ads MCP rules can give an AI agent read access to performance data while blocking budget, campaign or creative changes, and advises starting read-only or approval-required and widening access over time.
“Start an account in a read-only or approval-required posture and widen permissions programmatically as confidence grows”
Meta says its ads MCP server needs a token with ads_mcp_management plus ads_read or ads_management, more permissions open more tools, and only Employee-role system user tokens work, not Admin-role ones.
“Admin-role system user access tokens can't access the ads MCP server.”
Rooster Sanchez, on the AI DTC WTF podcast, says running his media buying plans past several AI models acting as a judge panel has helped him reach good plans faster.
“AI judges like a judge panel running my stuff through that has been a game changer because it helps me come up with good plans Quicker”
Cody Plofker says his new brand Winks is AI-first, with most ads AI-made, statics and videos, and a rule to prove a task cannot be done with AI before hiring for it.
“Most of our ads are AI too, statics and videos”
A podcast guest, the founder of Home Plunge, says he replaced all his Google, Meta and Klaviyo agencies with AI systems. He also plugs his own web app that packages those systems.
“I've relinquished myself from every agency from Google, to meta, to Clavio, all of it and I've replaced it with AI systems.”
Meta's ads MCP server rules let a business block agent actions, such as creating campaigns, changing targeting or creative, or raising budgets past a cap like 20%. In limited availability.
“deny any budget increase above 20%”
Nik Sharma says brands with multi hundred million dollar revenue are already using agentic AI media buying in production, naming Triple Whale as a provider.
“Triple will has a gentic media buying and they have multi hundred million dollar brands doing it fully.”
CTC says over 60% of a typical brand's media budget goes to Meta, and more than half of a Meta buyer's time went to building campaigns and uploading ads, which it automated with its own tool.
“more than half of their time is spent on the process of creating an advertising account”
CTC's Taylor Holiday says humans are poorly suited to analyzing daily ad account optimization, and that AI oversight and automation should do that job.
“humans are not well designed for that job”
One agency anecdote: True Classic went from four people managing paid ads to one managing their system after automating all its Meta spend, with the others moved to new channels. No test.
“four people managing their paid ads to one person managing their system”
Meta's ads MCP server creates campaigns, ad sets and ads paused, and a separate tool activates them; the AI client asks for confirmation before anything starts spending.
“Write tools create entities in a paused state; your AI client asks for confirmation before activation.”
Matt Bertulli built AI tools and dashboards that pull live ad account data, but says he lost days debugging and shipped tools that broke the next day.
“the prototype is not the product.”
CTC runs Claude with written playbook skills over an MCP connection to analyze client Meta accounts, and its media buyers act as editors who check the output and add client context.
“they're an editor, they're a filter, they get to gut check.”
Ecommerce Alley's hosts say many clients export Meta data for AI tools such as Claude to analyze, and that the resulting recommendations are often inaccurate and have wrecked campaigns. No figures.
“we found a lot of it is actually inaccurate and it has absolutely wrecked campaigns.”
True Classic's CEO says it runs 100% of its Meta spend through Triple Whale's autonomous media buyer, Moby 2, on an episode with Triple Whale's COO. No results or price are disclosed.
“True Classic now runs 100 percent of its Meta spend through an autonomous media buyer, Moby 2”
In the full answer
For readers who sign in- What to do this week3 steps
- The numbers behind it2 figures from cited sources
- Where each source standsA chart placing each source on the question
- The strongest disagreementThe best case against the answer, from a cited source
- How sure we areWhat its confidence rests on, and what would change the answer