Agentic Advertising Is About Giving Marketers More Leverage, Not Less Control

Google's testing of an AI agent called Remy has raised questions about how much autonomy advertisers are willing to hand over to autonomous systems, but the real promise isn't about replacing human decision-making - it's about freeing up marketers to focus on what actually requires judgment.

The advertising industry is staring down a fork in the road. On one side, AI agents promise to handle the tedious, multi-step workflows that eat up hours of a marketer's day. On the other, there's a creeping anxiety about what happens when those agents start making decisions without asking permission first.

Google's internal testing of Remy, an AI personal agent that runs inside the Gemini app, is the latest signal that agentic advertising isn't a distant concept anymore. According to internal documents seen by Business Insider, Remy can keep track of events, handle complex tasks, and learn user preferences over time. The agent integrates with Google's services and operates in a staff-only version of Gemini, though a public launch timeline remains unclear. Google Marketing Live is happening later this month, and there's speculation the company might demonstrate the agent there.

But Remy isn't the only player in this space. Adobe announced several agentic ad tools at its Summit conference in April, including CX Enterprise, which can execute entire campaigns based on a single goal provided by the advertiser. There's also CX Enterprise Coworker, which coordinates multiple other agents to carry out complex marketing workflows. Adobe Brand Intelligence is another tool in the mix, designed to ensure all AI-generated advertising content remains on-brand by continuously learning from real-world feedback like campaign approvals or rejections.

Even with all this momentum, there's a problem: advertisers aren't ready to hand over the keys.

Greg Collison, Adobe's head of product and design, told MediaPost that the complete vision of AI agents creating and serving ads in real time requires significant changes and a lot of trust that advertisers simply aren't prepared to give yet. "Before AI agents can create and serve ads in real time it will require advertisers to trust technology to approve and serve ads as it creates them," Collison said. "There must be quality control, and the industry is not yet ready for something like that."

Jon Roberts, Chief Innovation Officer at People Inc., echoed this sentiment in a recent interview with Digiday, noting that "there is a trust gap to be solved before this really gets deployed." That's not a small hurdle. Advertising is a high-stakes field where brand reputation can be damaged by a single misstep, whether that's an ad appearing in the wrong context or creative that doesn't align with brand guidelines.

Collison pointed out that automated reviews that score ads based on a brand's requirements will need to mature before real-time creation and serving becomes viable. Programmatic ad assets are already built in milliseconds to fit into the timeline of getting the page loaded and the ad being served, but that's a far cry from trusting an AI to generate entirely new creative on the fly without human oversight.

One piece of the puzzle that's getting more attention is the concept of digital twins. Collison mentioned that Adobe is working on this technology, and it's not hard to see why it matters.

Digital twins in advertising are AI-powered replicas built from real behavioral data. They're not just hypothetical personas - they're simulated versions of actual consumers, trained on demographics, psychographics, interests, affinities, and media consumption habits. StatSocial launched a product called Digital Twins in June that allows brands to simulate audience research using AI-generated profiles built from behavioral data.

Shepherd, an audience strategy consultancy, was an early tester of StatSocial's Digital Twins. Dean McBeth, managing partner and co-founder at Shepherd, explained that the tool allowed the agency to accelerate its research capabilities. "We work with really highly niche audiences," McBeth said. "No one has these panels, but what Stat provided was hundreds of millions of people anonymized across tens of thousands of data points. We could get those behaviors and start to understand what people were doing online without asking them."

For Shepherd, Digital Twins meant they could test creative concepts, ask questions, and evaluate products without recruiting traditional focus groups. One early use case involved a news and entertainment company exploring whether audiences would pay for a new editorial offering. Instead of spending weeks recruiting a cohort that doesn't really exist in standard research panels, Shepherd matched the publisher's first-party subscriber data with StatSocial's behavioral graph, then used Digital Twins to segment core subscribers, casual users, and prospects.

The results challenged some assumptions. Core audience members liked the concept but weren't the most eager to pay. Prospective users showed a greater willingness to pay, particularly when the product was framed more closely to the creator-driven content they were already supporting. Those findings helped Shepherd rethink positioning and pricing before investing in broader testing.

McBeth said the agency has spent much of its testing period validating Digital Twins against historical research, first-party customer data, and live surveys. "We're actually getting a lot of the same feedback," he said. "There isn't a big delta between how they're responding as a twin and how we're seeing them show up in real life."

The phrase "agentic advertising" has been floating around more frequently, and it's worth clarifying what that actually means. According to Digiday, agentic AI in the context of advertising specifically is the use of autonomous and semi-autonomous AI agents to plan, transact, and optimize media.

Scope3, a company focused on AI-driven marketing, described agentic advertising as one of the clearest early examples of what AI-native marketing looks like in practice. "In agent-to-agent media buys, a buying agent structures and relays a single campaign brief to an ecosystem of sell-side agents representing surface owners and service providers," the company wrote in a blog post. "What used to require coordination across multiple platforms and parties now happens through AI agents, within governance parameters set by the brand."

That's a significant shift. Most advertising technology companies adding agentic capabilities are doing the same thing: layering AI agents and LLM connectors on top of existing platforms. The promise is that these agents can handle operational complexity while advertisers focus on brand strategy.

TensorOps AI published a field guide on agentic AI in advertising in May, noting that "agentic AI describes autonomous systems that can perceive a situation, reason through a multi-step plan, and take actions toward a goal with minimal human intervention." But the guide also included a crucial caveat: "An agent is only as good as the data, tracking, and structure beneath it."

Google's broader strategy involves something called the Agent2Agent (A2A) Protocol, an open standard designed to ensure that agents from different platforms can communicate. The goal is to create a multiagent ecosystem where agents based on security, coding, or data analysis can collaborate to solve broader challenges.

Demis Hassabis, CEO of Google DeepMind, has spoken about how agents are core to DeepMind's heritage. Projects like Atari game research and AlphaGo are examples of agent systems. DeepMind created AlphaGo and trained the computer to play Go on human knowledge, but the ultimate successor in the lineage became MuZero, which can master games without being told the rules beforehand.

That's the vision: agents that discover the answers without humans telling them what's required or needed. Hassabis has referenced this as a step toward AGI, though that moment still feels distant for most practical advertising applications.

Parallel, a digital-twin activation platform that launched in 2026, is betting on a different angle. The company uses human-calibrated digital twins to simulate and score the resonance between a customer, an ad, and the content it runs against. Parallel's Managing Director Ben Dimond explained that the platform introduces the concept of Customer Suitability to advertising, aiming to bring the voice of the customer into every advertising impression.

Parallel built a nationally representative cohort of 1,000 digital twins in the UK and 1,000 in Australia, each trained on real data and enriched with personalities, interests, and beliefs. The company says its digital twins are 80.4% accurate against real human responses.

The idea is that context isn't always king - the underlying customer need state is actually the key thing. Dimond gave an example: imagine a running shoe ad creative promoting a gritty 6am workout. In Scenario A, it airs during Premier League highlights. It's sports content, but the viewer is lying on the couch in relax mode. The creative doesn't resonate. In Scenario B, it airs during a video about improving football drills. The viewer is in a self-improvement frame of mind. The message resonates perfectly.

Both are football content, but only one captures the right need state. According to IAS, receptiveness to an ad increases by 1.5x when the ad is sentiment-aligned. That's not a trivial difference.

The conversation around agentic advertising often gets framed as a binary choice: either marketers embrace full automation or they resist it entirely. That's the wrong framing.

The goal should be to identify which parts of the advertising workflow genuinely benefit from automation and which parts still require human judgment. AI agents can handle repetitive coordination, testing, and optimization work across campaigns. They can run simulations, score resonance, segment audiences, and adjust bids in real time. Those are tasks that don't require deep brand understanding or creative intuition - they just require speed and consistency.

What AI agents can't do yet is make the kind of judgment calls that require context, cultural awareness, or long-term brand strategy. They can't decide whether a campaign concept aligns with a company's values or whether a creative direction will resonate with audiences in six months when the cultural moment has shifted. Those decisions still need humans.

Collison's point about quality control is critical here. Brands don't just need AI that can generate ads quickly - they need AI that can generate ads that won't damage the brand. That requires layers of oversight, feedback loops, and continuous learning from real-world campaign performance. Adobe Brand Intelligence is designed to address this by learning from campaign approvals and rejections, but that's still a work in progress.

The advertising industry has a tendency to swing between extremes. When a new technology emerges, there's either unbridled enthusiasm or complete skepticism. The reality with agentic advertising is somewhere in the middle.

AI agents are going to become a bigger part of how campaigns are planned, executed, and optimized. That's not really up for debate anymore. The question is how much autonomy those agents will have and how much oversight marketers will maintain.

For now, the smart approach is to treat agentic advertising as a tool that gives marketers more leverage, not as a replacement for human decision-making. Let the agents handle the grunt work. Let them run the simulations, test the combinations, optimize the bids, and coordinate the multi-step workflows. But keep humans in the loop for the decisions that still require judgment, context, and brand understanding.

That's not a compromise. That's just good strategy.

Sources