In this exclusive piece, Dmitry Sverdlik, CEO at AI and software development company Xenoss, reveals how the once-hyped promise of intelligent, multi-agent systems is becoming a reality in ad tech. Drawing from hands-on experience building custom AI solutions for marketers, Sverdlik makes the case: the multi-agent ad ops team of the future isn’t just coming – it’s already here, and shares a strategic and technical blueprint to build your own multi-agent system.
As part of a team delivering custom AI solutions for ad tech and Martech teams at Xenoss, I have seen a surge in demand for AI agents – and the operational gains they unlock are real. This shift isn’t theoretical. It’s already underway.
Anyone working in ad ops has seen the same workflow play out countless times. A campaign manager clarifies campaign objectives, target audience demographics, campaign budget, and KPIs and assigns tasks to the rest of the team based on what the agency and the client agreed upon.
An audience data analyst retrieves and analyses consumer insights from the company’s databases like CRM, CDP, and third-party sources.
Next, a creative strategist designs customised assets for the upcoming campaigns, tailoring the text and copy to audience properties and preferences discovered during segmentation. The creative team designs multiple ad variations to ensure dynamic in-flight optimisation.
A programmatic media buyer chooses ad placements, manages bidding strategies, and evaluates real-time bidding outcomes and market conditions.
A budget allocation manager then monitors expenditure based on the channel’s impact on performance and dynamically reallocates spending across channels, audience segments, and campaigns.
Finally, performance analysts collect and analyse data across all active campaigns, generating detailed campaign reports and actionable insights to help improve future campaign performance.
Today, AI agents, autonomous and ever-learning AI systems that reason, act, and collaborate, can fully automate and handle this familiar workflow.
Why AI Agents Are Your Next Great Hires
The AI agent market is growing rapidly. Capgemini estimates that its value will surpass $47 billion by 2030. Salesforce, Microsoft, Adobe, HubSpot, and Relevance AI have all released agentic tools spanning advertising, marketing, and sales.
Scope3 followed the trend by creating an agentic marketplace where ad tech professionals can connect multiple agents to build intelligent automation workflows for media buying.
Just recently, Google released a general-purpose multi-agent system that allows building intuitive and complex AI agent workflows.
Here’s how the AI agents function like a human team. They take on roles, coordinate tasks, share data, and adjust in real time.
In this workflow, each agent mirrors a human role – yet operates with speed, consistency, and adaptability that scales far beyond what traditional teams can achieve.
How Multi-Agent Systems Are Rewriting Ad Ops
If you want to understand how much full-scale adoption of multi-agent systems is a game-changer in ad ops, start by considering the limitations of traditional ad tech tools.
Deterministic automation platforms force ad ops teams to operate within these tools’ narrow feature sets. Customising these out-of-the-box solutions has been the bane of ad ops managers’ existence for years.
With AI agents, a lot of these will become history.
Once multi-agent systems are embedded in the ad tech stack, domain teams will no longer have to contact vendors for customisations, build in-house engineering teams to enable unified reporting, or set up custom attribution models.
Unlike traditional automation tools, AI agents reason independently and adapt over time – unlocking a new level of flexibility.
Adobe’s agentic platform is a case in point. It allows domain teams to configure workflows using granular, no-code components. With dozens of ready-to-deploy agents, ad ops managers can easily orchestrate complex, flexible campaign systems.
Traditional automation platforms also age quickly.
Over time, ad agencies usually find themselves dealing with a poorly connected patchwork of newer tools and legacy systems, a “Frankentech” stack. This fragmentation breeds silos, slows down workflows, and undermines the very productivity gains automation is meant to deliver.
Cognitively endowed agents are different because they do not necessarily need complex integrations to connect with other tools in your team’s stack.
Historically, agents interfaced via APIs or GUI automation – essentially mimicking how a human might use the software. Now we’re seeing a third integration pathway entering the agentic landscape: Just-in-Time (JiT) compilation. It allows intelligent agents to navigate software by writing code and communicating with the third-party platform’s internal codebase.
Once adopted at a large scale, JiT can empower the integration-free navigation of legacy ad tech stacks.
Independence from engineering teams and speed of reacting to market changes are other milestones multi-agent systems can help bring to reality.
Automating processes in a complex deterministic ERP platform usually goes beyond the skillset of ad ops teams, pushing agencies to set up internal process automation teams to design workflows. Relying on IT makes it harder for ad ops teams to test market trends and quickly capitalise on emerging opportunities.
With intelligent multi-agent systems, it is a different story.
Building complex workflows without reliance on IT empowers ad ops teams to develop and test pilot projects without committing to lengthy engineering projects.
This is powerful because agents don’t just follow static instructions. After deployment, they use cognitive reasoning to adapt to market conditions and course-correct without human involvement.
Even when ad tech leaders are clear on the benefits of multi-agents, a fuzzy understanding of technical considerations in such systems often stalls the pace of adoption.
What I Know Matters Most When Building Scalable Multi-Agent Systems
When helping multiple ad ops teams build multi-agent systems, I noticed that no matter how granular the teams’ needs are, MASs tend to share foundational capabilities.
First, you need a library of prebuilt agents – modules that handle core adops tasks like budgeting, segmentation, and performance analytics. But out-of-the-box tools aren’t enough. That’s why I recommend a no-code or low-code environment where domain teams can define agent roles, objectives, and behavioural parameters.
Orchestration is equally important. Teams need a coordination layer where agents collaborate dynamically to solve complex tasks.
I also recommend that team leaders integrate all top-of-the-mind LLMs into their MAS to switch between models dynamically depending on the tasks and avoid downtime (if the default LLM is down, the system automatically uses a different one).
The system also has to plug into the organisation’s data stack, which means it needs direct access to CRMs, databases, and email systems.
Finally, AI agents should interact with your team’s entire ad tech infrastructure, whether hybrid (a mix of on-premise and cloud solutions) or distributed across multiple cloud environments.
The tech needed to support these features is already on the market, and future-facing companies should use it to build intelligent MASs to streamline workflows.
Why the Future of Ad Ops is Human-AI Collaboration
Here’s what I believe: Within the next two to three years, ad ops will become a hybrid model of human-AI collaboration.
Multi-agent systems will handle the lion’s share of the executional load while humans provide the guardrails and strategic guidance. KPMG data suggests that fifty-one percent of advertising and marketing teams explore AI agents for content personalisation and real-time campaign optimisation.
The ad ops process will be almost entirely AI-driven in five to seven years, and multi-agent systems will operate with high autonomy. The virtual ad ops team will become a self-driving car for marketing campaigns, where humans set the destination and AI agents handle the journey, navigation, and real-time optimisation.
However, ad ops team leaders may wonder if humans still have a part to play.
I firmly believe they do. In fact, the leaders I’ve seen successfully leverage AI tend to focus on productivity augmentation over job role substitution. Even as agents handle most day-to-day operations, they keep humans in charge of ideation and ongoing supervision.
For example, though AI agents can already generate and test creatives, having a human review them for brand safety never hurts.
Eventually, well-oiled campaign management will no longer be a competitive differentiator. In the new landscape, authentic creative concepts, brand storytelling, and novel campaign ideas will make a real difference.
How AdOps Leaders Can Future-Proof Their Teams
The most critical shift ad ops managers can make now for success in the agentic era is learning to think like engineers so that they can manage intelligent systems as effectively as they guide their human peers.
Agents understand natural language well, but vague or ambiguous instructions can lead them astray. To make sure humans and AI understand each other without a hitch, ad ops managers need to formulate tasks in an agent-friendly way.
Designing Effective Agent Interactions
In a multi-agent system, each agent handles a complex task by breaking it into actionable steps – often requiring specialised tools or knowledge. When I work with clients on MAS architecture, we don’t just design agents individually – we model how they interact as a system.
Multi-agent systems have the unique ability to leverage collective intelligence. However, achieving such synergy requires domain teams to have a basic understanding of system design principles.
Suppose you are building a system where an audience segmentation agent communicates with a creative design agent, who coordinates with the performance management agent to optimise assets dynamically based on real-time campaign performance.
That workflow only functions if you’ve defined how the agents communicate – directly (decentralised protocol) or via an intermediary (often denoted as the orchestrator).
Beyond that, there are critical technical considerations: What protocols do agents use to exchange information? In what sequence? What happens if one agent fails to deliver a signal?
By mastering system design principles, ad ops leaders can prevent miscommunication, build robust AI workflows, and unlock the full value of multi-agent collaboration.
Final Thoughts
I’ve seen so many teams confused or at a loss by the AI agent hype wave. I think it’s important to understand that multi-agent systems have the potential to rewrite ad ops practices.
Still, at the current level of market maturity, the best way to gear up for the change is not to tear workflows apart in hopes of full AI-driven automation.
Instead, teams should build point-based multi-agent systems to automate simpler workflows and gradually expand the foundation environment by throwing new agents and workflows into the mix.
I am confident that the future of ad ops is agentic, and it is already here for those ready to build it.
