The AI Advertising Podcast: S2
Episode 5
AI Agents & the Future of Advertising
About This Episode
AI in advertising is entering a new phase of autonomy.
This episode unpacks what agentic AI really means, how Ivy™ is evolving into a true marketing agent, and why the Model Context Protocol (MCP) is quietly becoming one of the most important technologies in ad tech.
Yang Han | CTO at StackAdapt
Lee Odden | CEO at TopRank Marketing
Transcript
Diego Pineda (00:00:00)
Marketing is entering the agentic AI era—and the numbers are staggering. According to the Martech for 2026 report, 90% of marketers say they’re already using AI agents somewhere in their workflow, but only 23% have those agents in full production use. Most teams, a full 67%, are still stuck in pilots, experiments, or very narrow use cases.
At the same time, customer behaviour is shifting even faster. Half of consumers already use AI-powered search, putting 20 to 50% of traditional search traffic at risk.
So while AI agents are rapidly becoming the operating system of marketing… Most teams are still figuring out what these systems actually are, how they work, and how they’ll reshape campaign strategy, creative workflows, and media execution.
That’s why today’s episode matters. We’re going to break down what agentic AI really means, and what marketers should prepare for next. Joining me is Yang Han, CTO of StackAdapt, who’s been leading StackAdapt’s investment in agentic systems, including the development of Ivy, the platform’s marketing AI agent. We’ll also hear from Lee Odden, CEO of TopRank Marketing, and his take on AI agents from the agency side. So let’s start at the very beginning.
Podcast Intro (00:01:24)
Welcome to the AI Advertising Podcast, brought to you by StackAdapt. I’m your host, Diego Pineda. Get ready to dive into AI, Ads, and Aha moments.
Diego Pineda (00:01:40)
There’s a lot of noise in the industry about “agents.” Some vendors call every chatbot an agent. Others say only fully autonomous systems count. Here’s Yang Han.
Yang Han (00:01:50)
There’s certain capabilities that have to leverage predictive AI that works behind the scenes, that’s more of a black box. These types of AI are complementary to what AI agents can do. So both of these types of AI need to be used properly in the right situations. And if so, they won’t conflict. And in fact, they can complement off one another. So AI algorithms that occur behind the scenes that are more black box are suited for solving like a very specific problem that’s more deterministic. And it typically involves, especially in our industry, high volumes of data in real time that has to be processed very quickly in a matter of milliseconds. And it has to happen at low cost, essentially. So real-time bidding environments are ideal for this because we need to make decisions on millions of data points per second in a split second. And also historical data has to be available to train these algorithms so they can solve a specific task. So this might include like optimizing towards a KPI, calculating specific but complex analytical metrics as more of a part of an overall result.
Agentic AI systems, on the other hand, they can handle more dynamic interaction and inputs in flexible environments. So these are more suited for broader and general purpose applications, especially problems that can allow different types of outputs depending on what the inputs are. So anything that can mimic, let’s say, a human interaction or provide guidance to people, execute workflows based on dynamic situations. Oftentimes you can use agentic systems to also operate a series of black box AI algorithms to execute a broader task and a higher level problem you’re trying to solve.
The key difference of agentic AI versus other forms of AI is you can directly interact with it and it can continuously adapt to complex situations and environments with varying external inputs. So whether it be inputs from humans or even other AI agents.
Diego Pineda (00:03:57)
That part is important: adaptation. Traditional AI models give you answers. Agentic systems take actions. And Yang explained that Ivy, StackAdapt’s agent, is already evolving along that spectrum.
Yang Han (00:04:10)
Ivy started out as an LLM, but we have developed capabilities now where it’s beyond just a standard chatbot. It can do reporting, visualizations, measurements. It can take in inputs to generate forecasting, planning, and various other capabilities. And we are also developing IVY to be able to interact with other agents as well. So it is evolving from just an LLM into something that can be plugged into a workflow that can execute its own workflows and so forth.
Diego Pineda (00:04:44)
So think about the difference like this:. Old AI = “Give me an answer.”
Agentic AI = “Understand the goal, gather inputs, make decisions, and take action.”
And that evolution is reshaping creative production, optimization, and cross-channel orchestration.
But even with all this promise, there’s a real question: Are marketers actually ready to use agents?
Diego Pineda (00:05:12)
If you’ve been listening to this show, you know that adoption is often slower than innovation.
In Martech for 2026, one of the most striking findings is that only 8% of organizations feel fully confident in their AI governance. That tells us a lot—this is as much an organizational shift as it is a technological one.
Yang Han (00:05:32)
I think education is key. I think on our platform, we are starting to see more and more adoption. The results are really promising when it comes to our own AI agent internally. But there is still a lot of capabilities to build out in the AI agent to be able to execute autonomously end-to-end the entire platform. Right now, there’s different capabilities the AI agent can do. So that’s always, in the near term, provide certain limitations for, let’s say, complete adoption. But that’s going to change every single year, we’re going to see more and more capabilities supported on the AI agentic framework on the agentic platforms and we hope to see agents being able to integrate with one another. That will make it very powerful once integrations get supported where we can get the intelligence from one agent into IVY that can execute something very holistic and very strategic end to end.
Diego Pineda (00:06:26)
To add another perspective, I turned to Lee Odden, who is implementing AI at his agency. I asked him whether AI assistants will eventually evolve into more independent agents.
Lee Odden (00:06:37)
I do believe there are absolutely use cases for that in an agency environment. And maybe even someday there will be a BAM agent, a Best Answer Marketing agent, right? That, I mean, initially that would be something that our own team members would use, but inevitably I could see something like that that our clients could use, you know, so sort of as an on-demand resource for them to get insightful information according to a pair ah framework that will then be continuously fed with best, you know, performance data. I know what’s working, what’s not and that sort of thing. And that would be limited to our clients, obviously. But yeah, I absolutely see that there’s an opportunity for, agents to perform certain tasks. And I think the obstacle is, or not obstacle, it’s just the consideration is that where can we have confidence of accuracy? Where can we have kind confidence in consistency of output from these agents? When we solve for consistency and quality and you know that continuity that we need in our work with an agent, then absolutely, why not rely on that resource?
Diego Pineda (00:07:39)
So the moment we’re in is the shift from curiosity to experimentation to trust. And trust is built not just through better AI, but through better infrastructure.
Diego Pineda (00:07:54)
Here’s the big unlock: For agents to do more than answer questions—for them to act—they need a way to interact with other systems, tools, and even other agents. This is where MCP comes in. What is it?
Yang Han (00:08:07)
MCP stands for Model Context Protocol. It’s basically a standard for AI to communicate with another service or another AI, essentially. So if we think about how AIs can integrate with one another to create a greater workflow to solve a greater problem, MCP is the protocol and integration method to enable that.
Stackdap uses MCP internally and also as an offering to our clients. So some background Stackdap has IVY, which is our marketing AI agent. Historically, that lived within Stackdap’s platform. It’s an agent that you can communicate with to help you solve problems, analyze data and whatnot. But we really wanted our customers to leverage IVY even in their normal workflows and day to day in their own systems. And so we developed an MCP integration in which customers can utilize IVY and call IVY from their, you know, ChatGPT, or Claude, their favorite LLM, and then incorporate StackAdapt’s, data, planning capabilities, strategies, and whatnot into their greater workflows in a seamless flywheel, essentially.
Diego Pineda (00:09:22)
Let me add context here: Before MCP, integrating systems required engineering work — APIs, custom scripts, weeks of development time. Now two AIs can simply talk to each other. So what does this look like in real campaign execution?
Yang Han (00:09:39)
What MCP and AI offers is a lot of, by nature, a lot of the fundamental pieces here and there. The complex part when it comes to the end marketer is really stitching it together to come up with a comprehensive flywheel from the planning phase that feeds into the execution phase, that feeds into the data measurement and uses that to learn. Because the power of AI isn’t AI in isolation. You know, talking with one AI, you can get some value, but it’s limiting.
The power of AI and MCP is really when it all comes together so that you have multiple AIs and you figure out how they can collaborate with one another. They all have their own goals and missions. One informs the other and so forth, and you can then close that loop so that the system is ever learning. The end goal is to get to a learning system. That’s the purpose of AI. Otherwise, it’s just like any other system. You want to set things up in which the system can seamlessly learn on its own without, with how or little human intervention. And the only way you can do that is to have multiple AIs that can flow as inputs and outputs to one another. But to configure this, that is where a lot of the… it’s open ended right now. And that’s where companies are figuring out how do we piece it all together so you can create this very powerful flywheel. And I think that’s an ongoing evolution every company is trying to figure out.
Diego Pineda (00:11:14)
This is the beginning of true autonomous orchestration. Not real-time bidding — that still requires millisecond-level systems — but strategic decision-making at scale. And now that agents can talk to other agents, the flywheel becomes possible: Planning → execution → measurement → learning → back to planning. That’s the future we’re steering toward.
Diego Pineda (00:11:42)
What about the rising consumer-side agents like shopping agents, recommendation agents, and conversational commerce? These systems will reshape how people discover products and how brands show up in the AI ecosystem.
Yang Han (00:11:57)
There’s a lot of synergies between shopping agents and advertising as well because as people ask these agents to buy the right products for them or recommend the right products, you know, marketing is always going to play a role in terms of being able to show something that could be relevant to them as well. And to make this work, it’s really important for brands to ensure that there’s relevant information about the brand and the products available on the web for LLMs to suggest. So you want to ensure it makes it to the output of the LLM.
And LLM optimization is very much a concept now, similar to traditional SEO. And going forward, as the agent-to-agent ecosystem continues to develop, even though this is a relatively new concept, brands should prepare to have information accessible to other agents, such as using the MCP, so that suddenly there’s information shared across the board in terms of the whole digital ecosystem, so that agents can be very intelligent to recommend something really relevant when it comes to shopping purposes for a customer.
Yang Han (00:13:11)
If we look in the future, there’s no doubt that agents are going to be able to do more and more. But the hope is that as agents develop to be able to handle more tasks, it will allow them like marketers and just you know humans in general to focus on more human tasks, more things that involve human to human. Because I think in general, regardless of your role, I don’t think any person really enjoys the robotic and mundane part of their jobs. But the reality is that in many roles today, there’s still a lot of like repetitive, pretty mind-numbing tasks that tend to take up people’s times. And those are going to be ripe for disruption anyhow.
And it’s true that AI can do basic critical thinking on behalf of people too. But when that does become the norm, there’s going to be a greater need, I think, for like true human interaction, right? And deeper interpersonal development and communication among people itself. And so the hope is that by elevating humans to get closer to one another, instead of us just interacting with machines all day, we can let machines talk to each other. You know, it will be better for like, I’d say human society overall, because we can produce more harmony, mutual understanding between people beyond like the artificial boundaries and hopefully it will bring the, you know, people in general societies closer together.
Diego Pineda (00:14:36)
That’s the heart of it. So here are the three big takeaways from today’s conversation:
1. Agentic AI is already here. But adoption is uneven. Teams that invest early will gain a real competitive advantage.
2. MCP is the hidden breakthrough. It’s what allows any agent to access tools, coordinate actions, and learn from multiple systems.
3. The human marketer becomes more valuable, not less.Execution gets automated. Insight, strategy, creativity, and empathy become the differentiators.
This is the era where marketers shift from doing everything to directing intelligent systems that work alongside them.
Podcast Outro (00:15:19)
Thanks for listening to this episode of The AI Advertising Podcast. This podcast is produced by StackAdapt. Visit us at stackadpat.com for more information about using AI in your advertising campaigns. If you liked what you heard, remember to subscribe, and we’ll see you next time.
You May Also Like
Infographic
GLOBAL
Nov 17, 2025
2026 Programmatic Advertising Trends Sneak Peek
Webinar
NORAM
Jan 28, 2026
2026 Programmatic Advertising: Trends to Watch, Moves to Make
