Mark Zuckerberg's stated goal is to make advertising as simple as entering a credit card number and a business goal. That vision is real and it is moving fast. But agency leaders running Meta Advantage+ campaigns daily are describing something messier: worse placements, ballooning creative workloads, and systems that require constant human babysitting. Both things are true at once.
In April 2026, Marketing Brew spoke with agency leaders across the US about their experience with Meta's AI ad tools. The picture that emerged does not match the keynote version. It is more complicated, more labour-intensive, and in some cases more expensive than the fully automated future Meta is selling.
This post breaks down exactly what each tool does, what the performance claims actually measure, and what practitioners are experiencing on the ground. If you are running Meta campaigns right now, this is the context you need.
For the broader picture of where AI advertising is heading across platforms, read what the $47 billion AI marketing market means for advertisers. And for Google's equivalent shift, see how AI is changing Google Ads optimisation.
TL;DR: Meta's AI ad tools, including Andromeda and Advantage+, are real and improving. Meta claims a 14% ad quality improvement in Q3 2025 from Andromeda alone. But agency practitioners report mixed or worse results, a creative volume problem nobody warned them about, and systems that still require significant human oversight. (Marketing Brew, 2026)
What Do Meta's AI Ad Tools Actually Do?
Meta's AI advertising infrastructure now covers three distinct layers: Andromeda handles ad retrieval and matching, Advantage+ handles creative, targeting, and budget automation, and Manus is an AI agent some advertisers are beginning to see in Ads Manager. Together, they represent Meta's push toward fully automated campaign management. Understanding what each layer does is the starting point for evaluating whether any of it is working.
Andromeda: the matching layer
Andromeda is Meta's new ad retrieval system, rolling out since late 2024. The shift it makes is significant. Traditional Meta targeting worked by defining fixed audience segments, then serving ads to those segments. Andromeda flips that model.
Instead of targeting a fixed audience, you upload a wide range of creative assets and let the system match your messages, formats, and angles to different users. The AI decides who sees what. The implication: the system's ability to match well depends entirely on the diversity and volume of creative you provide.
Advantage+: the automation layer
Advantage+ is Meta's AI tool suite for creative, targeting, and budget optimisation. It progressively reduces the manual controls advertisers have traditionally relied on, replacing human decisions with algorithmic ones. Meta frames this as efficiency. Practitioners describe it as loss of control. How you structure your Advantage+ campaigns matters more than the tool itself.
Manus: the agent layer
Manus is an AI agent Meta acquired in December 2025. Some advertisers are now seeing it appear inside Ads Manager. Its exact capabilities are still limited in public documentation, but its presence signals Meta's intent: eventually, an AI agent handles the full campaign cycle from setup to optimisation. That future is not here yet. But the infrastructure is being assembled visibly.
What Does the 14% Improvement Claim Actually Measure?
Meta's most cited internal metric for Andromeda is a 14% improvement in Facebook ad quality in Q3 2025, reported to Marketing Brew. That number sounds significant. But ad quality is not the same as business outcome. Before you weight it heavily in your planning, it is worth understanding what the metric covers, and what it does not.
In practice, "ad quality" in Meta's measurement framework refers to relevance and engagement signals within the platform, not downstream conversion performance or cost per acquisition. A 14% improvement in ad quality scores can coincide with flat or declining ROAS depending on placement mix, audience match, and the creative inputs going into the system.
Meta has a structural incentive to report platform-side quality metrics positively. That does not make the number fabricated. It does mean it should be read alongside your own account data, not instead of it.
On the record: Meta's Andromeda system delivered a 14% improvement in Facebook ad quality in Q3 2025, according to Meta's own reporting to Marketing Brew. The metric measures platform-side ad relevance signals rather than advertiser-side business outcomes such as cost per acquisition or return on ad spend. (Marketing Brew, 2026)
What Are Agencies Actually Experiencing on the Ground?
The agency reactions reported by Marketing Brew in April 2026 range from cautious to openly critical. Several patterns emerge from what practitioners told the publication, and they are consistent enough to take seriously.
Aaron Edwards at The Charles Group described widespread hesitation around GenAI for ad creative specifically. "There's a lot of hesitation as to using GenAI specifically for ad creative right now," he told Marketing Brew. The concern is not that the technology does not work. It is that the output quality is not yet reliable enough to trust at scale without significant human review.
Hayley Owen at Deutsch described a pattern many performance marketers will recognise: "We're constantly having to go through and play Whac-A-Mole to figure out what's the new thing they didn't tell us." Meta releases features and updates frequently. Agencies often discover changes through their campaign data before any official communication arrives.
This communication gap is not a minor operational annoyance. When a platform updates its delivery algorithm or introduces a new automation layer mid-flight, it can shift campaign performance significantly before an account manager has any indication that something changed.
Daniel Johnson at We Scale Startups was more direct. His team experiments regularly with Advantage+ pre-existing functionality and "consistently sees worse results." Not mixed results. Consistently worse.
Jeremy Schulkin at Hawke Media flagged a specific mechanical issue: Advantage+ campaigns steer toward lower-quality placements, and the system still requires continued human oversight to manage that tendency.
On the record: Agency leaders running Meta Advantage+ campaigns in 2026 report mixed to negative performance outcomes. Daniel Johnson of We Scale Startups told Marketing Brew his team "consistently sees worse results" from Advantage+ features. Jeremy Schulkin of Hawke Media noted campaigns steer toward low-quality placements and require human oversight to correct. (Marketing Brew, 2026)
What Is the Creative Volume Problem Nobody Told You About?
The least-discussed implication of Andromeda is also the most operationally significant. The system's matching capability depends on creative diversity. To match different messages, formats, and angles to different users, it needs a wide library of creative inputs to draw from. That means Andromeda does not reduce your creative workload. It increases it, substantially.
One agency task initially requiring 300 unique assets ballooned to 1,000 when accounting for different personas and concepts, as reported to Marketing Brew. This is not an edge case. It is a structural feature of how Andromeda works. The more diverse and numerous your inputs, the better the matching quality. The system is designed to consume creative variety.
On the record: Meta's Andromeda system requires high creative volume to function effectively. One agency task requiring 300 unique assets grew to 1,000 assets when accounting for different personas and creative concepts, according to Marketing Brew (April 2026). Andromeda shifts the workload from targeting management to creative production.
This creates a real resourcing problem. Most marketing teams do not have the infrastructure to produce creative at that volume without significantly increasing production spend. Agencies that entered Andromeda expecting efficiency gains are finding they need to restructure creative workflows entirely to feed the system adequately.
The volume requirement also interacts poorly with quality control. When you are producing 1,000 assets instead of 300, the marginal review time per asset drops. Brand consistency, messaging accuracy, and compliance checks all become harder to maintain. Building a systematic creative testing framework before scaling into Andromeda is the prerequisite, not an optional extra.
What Should You Actually Do With Meta Campaigns Right Now?
The agency reactions in Marketing Brew's April 2026 reporting converge on one clear point: human involvement remains necessary "for quite some time." That is the practitioner consensus. Here is how to approach it practically.
Test Andromeda with a real creative budget, not a token allocation
Andromeda does not produce useful signal on a small or homogeneous creative set. If you want to test it properly, you need genuine creative diversity: at least 20 to 30 distinct creative concepts, not variations of the same concept. Watch placement quality and delivery data, not just platform-reported quality scores. Give it enough budget and time to optimise before drawing conclusions.
Retain manual placement controls until you have account-level data
The Hawke Media finding about low-quality placements is worth taking seriously. Advantage+ automated placements can include inventory that would not pass a manual review. Run parallel campaigns with placement controls active and compare CPA and quality metrics across both. Do not remove manual controls based on Meta's aggregate reporting alone.
For a clear view of what is actually driving performance across campaign types, a solid attribution setup is what separates informed decisions from guesses.
Build a modular creative production system before scaling
If Andromeda's creative volume requirement is going to double or triple your asset count, the production workflow needs to change before the campaign does. Modular creative systems, where you build components separately and combine them, are more efficient at scale than producing finished assets one at a time. Establish that system first, then scale into Andromeda.
Track Meta's feature releases as an operational risk item
The "Whac-A-Mole" dynamic Hayley Owen described at Deutsch is a real operational risk. Set up a process to monitor Meta's changelog, agency forums, and practitioner communities. You should not be discovering platform changes through your campaign data after performance has already shifted.
Frequently Asked Questions
What is Meta Andromeda and how does it affect advertisers?
Andromeda is Meta's ad retrieval system, rolling out since late 2024. It shifts from targeting fixed audience segments to matching diverse creative assets to users algorithmically. The practical impact is a significantly higher creative volume requirement. One agency saw a task grow from 300 to 1,000 assets when Andromeda's persona and concept requirements were applied, according to Marketing Brew (2026).
Is Meta Advantage+ actually delivering better results in 2026?
Results are mixed across agencies. Meta claims a 14% improvement in ad quality from Andromeda in Q3 2025. But practitioners report inconsistent outcomes. Daniel Johnson at We Scale Startups told Marketing Brew his team consistently sees worse results from Advantage+ features compared to more manually controlled campaigns. Your account-level data matters more than platform-level averages.
Should I use Advantage+ Shopping or Advantage+ Audience campaigns in 2026?
Worth testing with controls in place. The key issue flagged by Hawke Media is that Advantage+ campaigns tend to steer toward lower-quality placements without manual oversight. Run parallel campaigns with and without placement controls, compare CPA rather than platform quality scores, and make decisions based on your own account data rather than Meta's aggregate reporting.
What is the Manus AI agent in Meta Ads Manager?
Manus is an AI agent Meta acquired in December 2025. Some advertisers are beginning to see it appear in Ads Manager in early 2026. Detailed public documentation on its current capabilities is still limited. It represents Meta's broader direction toward AI-managed campaign cycles, but its practical impact at the account level is not yet well-documented by practitioners.
How much creative do I need to run Andromeda effectively?
Significantly more than a standard campaign. The system's matching capability scales with creative diversity. Based on agency reporting to Marketing Brew in 2026, planning for two to three times your current creative volume is a reasonable starting point. Build a modular production system first. Creative testing at scale requires a different workflow than traditional campaign creative production.
Zuckerberg's vision of advertising reduced to a credit card number and a goal is not impossible. The infrastructure Meta is building points in that direction. But the honest read of where things stand in April 2026 is that the tools are evolving faster than the results are catching up. Multiple agency leaders with real budgets and real client accountability are describing mixed or worse outcomes, higher creative workloads, and systems that require more human oversight than Meta's marketing implies. Run the experiments. Build the modular creative systems. Keep manual controls in place until your own data gives you a reason to remove them.
