No, 007 hasn’t been wheeled out of retirement. James Bond isn’t about to unleash a phalanx of spies on the marketing world. “Agentic AI” is a bit less dramatic than it might sound, yet also has the potential to take organizational effectiveness to another level.
The “year of the agent” (as Nvidia’s CEO put it at this year’s CES) is all about agentic AI and the impact it will have on transforming workflows, as well as employee and customer experience.
Let’s get up to speed.
The term agentic AI appears to have sprung from out of nowhere into ubiquity over the last six months, but it’s not an entirely novel concept. IBM suggests that AI agents have been around for some time, acting to enable autonomous cars, virtual assistants, and software ‘co-pilots’.
However, agentic AI still manages to confuse, so it helps initially to establish where it sits in the AI hierarchy, or evolution.
AI agents are then coordinated by the agentic AI—essentially another AI agent tasked with ‘managing’ the team to achieve a certain goal.
The intelligent aspect comes to the fore when agentic AI takes the outputs from the various AI agents and makes decisions based on the goals it has been set.
Let’s unpack a real-world analogy. Your home has thermostats, radiator controls and water heaters—these are your agents.
Meanwhile, your smart home controller fills the role of the agentic AI, making decisions based on the information these ‘agents’ are feeding back to it and then getting them to act on it: turn the heating up, shut down the air conditioning unit, set a new on/off timer, and so on.
Your business, of course, is not an HVAC system. In terms of how this technology will affect brands and marketers, “the biggest impact of agentic AI is going to be in efficiency, data unification, and autonomous decision-making, as well as resource management,” explains SmartAssets’ CTO and co-Founder, Eric Walzthöny.
“Instead of creating fresh assets for a new marketing campaign, AI agents can pick the most relevant content from the library."
—Eric Walzthöny, CTO and co-founder of SmartAssets
Other use cases for agentic AI include automated campaign optimization, particularly in areas such as bidding. AI agents could adjust bids, targeting parameters and creative variations in real time.
Yes, we know: Bidding adjustments do already happen, powered by AI—but not in the coordinated fashion provided by agentic AI, or with the ability to optimize creative in tandem to drive even better ROI.
“Instead of having a data aggregator, for example, you get a connector that links to all your advertising data—including platforms like Meta. Then an agent that knows how to interact with each of the data points through smaller, specialized agents, is wrapped around it. The marketer only needs a single point of interaction,” Walzthöny adds.
If the marketer’s job is reduced to interacting with a single, automated, AI agent, the elephant in the room becomes: Will this finally be the AI that takes my job?
“Usually what happens is that, as we automate, it creates jobs for people that didn’t exist before,” clarifies Stagwell Marketing Cloud’s Head of AI Solution Development, Louis Criso. “It’s tough to predict what the actual outcome will be, but it should create better opportunities and more fulfilling work.”
The future of marketing skills and roles is just one example of why, while the concept of agentic AI is relatively simple, its implications can be complex and far-reaching. It requires a certain leap of faith, something that could be difficult as executives in late 2024/2025 don’t seem to believe they have a burning reason to hasten their adoption of agentic AI solutions.
According to KPMG’s AI Quarterly Pulse from December 2024, over half (51%) of organizations are exploring the use of AI agents and another 37% are piloting AI agents. But currently, only 12% of respondents have deployed AI agents for use.
If you're currently a non-adopter, it's important to remember that agentic AI is neither difficult nor expensive to experiment with. It doesn’t require sophisticated tech stacks or even highly trained staff. Most solutions in operation today or whose release is imminent will focus on user-friendly dashboards that have a similar look and feel to existing tools.
Ease of use does have its potential pitfalls, however. The concept of ‘shadow AI’, where individuals within an organization forge their own path with AI tools outside any formal usage policy, is a worry.
Indeed, per that KPMG report, enterprise technology practitioners are worried about security (62%) and data governance (49%) when it comes to the development and adoption of AI agents.
When data—accessing it, interpreting it and creating more of it—is one of the core use cases for agentic AI, a bit of anxiety is understandable.
Consider MeldAI, a Stagwell Marketing Cloud application that combines private and public data to create dashboards that allow the user to ‘chat’ with that data. The user logs in, sees different widgets pertinent to the things that are most important to them, and finds a chat interface where they can ‘interrogate’ the agent. They can essentially ask 'what information do I need to answer my question?'"
The idea that agents are sent off under their own steam to hunt down data from a potentially infinite number of sources can feel risky, but take a deep breath.
"Agents only have access to the information your user or company already has access to. You know they’re picking from the right data by creating systems that go in and check the agent’s responses. It can then improve itself over time, based on that feedback,” Criso reassures—though he does stress that, while this is true of Stagwell Marketing Cloud’s agentic AI, it’s not universally true.
This sort of agentic checks and balances system is referred to as ‘scaffolding’. Harvard Business Review reveals that this element of learning science which gives “learners exposure to real-world practice with safeguards” is also applied to agentic AI.
Scaffolding can be created to keep AI agents’ activities within acceptable boundaries, based on the importance of the decision it is likely to make, the confidence in the data used to train models and the degree of human in the loop.
Similarly, organizations will not be able to run before they can walk if their data is not in the shape needed to help AI agents perform their function. No matter how sophisticated AI becomes, the adage remains true: Garbage in, garbage out.
Walzthöny explains: “The most important part about using your own data is what its current structure looks like. Is your data lake organized, or messy? If the latter, you’re going to have to find a way to organize it into something that’s usable. If it’s not organized, the agent will struggle to find the information it’s looking for.”
He notes that Google’s Knowledge Graph is a prime example of how to make life easy for AI agents. “Everything is already connected,” he says. “They can start from one edge of the data and go all the way across, finding the relevant information.”
AI adoption might sound daunting. The need to reframe entire data lakes’ worth of information...Restructuring organizations so that the human resource works effectively and knowledgeably in partnership with machines...Maintaining data security and governance when technology can operate autonomously and at speed...
The reality is much more mundane. Agentic AI does not currently require full data or tech stack overhauls. Certainly, tech companies are working to innovate their systems and create entirely new ones to address new challenges, but OpenAI models are very accessible, and can be cost effective to build an agent using open-source models.
A lot of layers are being built on top of existing solutions. You may find you’re already using agentic AI by default, as agents are wrapped around existing tools—a CRM platform, for example.
As the user, you pay for your software as a service plus agent. Eventually, these levels become more and more abstracted, so that the user is no longer interacting directly with the software—the agent is.
“Adoption is inevitable,” Walzthöny insists. “If one company wants to maintain the status quo, there will be a newer, younger one that doesn’t want to and takes advantage of these agentic AI techniques.
In the average eight hours of work marketers have a day, which ones will be able to output more, have better insights, and allow employees more time to make crucial business decisions?”
