Agentic AI is the new buzzword. The ad industry is embracing the technology, which goes a step further than generative AI, able to autonomously handle complex decision-making. But is it actually ready to run the show?

The year of agentic AI 

Over the past few years, AI has worked itself into the everyday. Already, we seem impossibly far from a world before the likes of ChatGPT burst into the mainstream, catapulted into the latest tech revolution led by Silicon Valley.  

While generative AI made the loudest splash, another type of AI has been working its way into the spotlight. Now, agentic AI has become the buzzword. Will 2025 be 'the year of agentic AI'? Many think so. Members of the advertising industry will no doubt have seen the buzz around agentic AI, as experts detail how the technology is already reshaping the ecosystem.  

To recap the basics, agentic AI goes a significant step further than generative AI, designed to autonomously handle complex decision-making and problem solving on behalf of a user. It has the capacity to understand high-level goals, make decisions under uncertain conditions, then plan actions and execute these to achieve specific goals, without the need for continuous human intervention. Agentic AI systems can also learn over time, based on feedback and new information, in order to self-correct. 

It’s designed to be able to do whatever it needs to in order to complete a task and reach a goal, including jumping between different platforms and interacting with other systems, unlike conversational generative AI, such as ChatGPT. They can work unprompted once set up, making the tech ideal to automate processes and tasks which have no end, such as website optimisation or tracking social media to uncover trends in real time, for example.  

Agentic agents, not to be confused with the broader term, are the specific, autonomous components within an agentic AI system. While AI agents may use a single agent to focus on a particular task, agentic AI makes use of multiple agents to manage complex workflows autonomously. 

How is agentic AI being harnessed by the ad industry? 

The ad industry has been embracing the technology, particularly for the media buying process. Platforms have been launched by ad tech players, designed specifically for the building and selling of agentic media products. 

Plenty of companies have launched agentic AI systems,  with aims to deliver improved targeting, streamlining the industry supply chain, and addressing inefficiencies in digital media buying. 

Many companies have also repositioned existing platforms as agentic AI, a term which Gartner coined 'agent washing', to refer to the rebranding of existing products which lack substantial agentic capabilities. Out of the very large number of vendors offering agentic AI solutions, Gartner estimates that only about 130 are real. 

The list of purported benefits from these platforms is long, it is touted as being safe and sustainable by design, dodging non-brand safe or risky inventory and made-for-advertising sites by default, while helping to eliminate ad fraud and invalid traffic. Ultimately, leveraging custom algorithms and AI agents to deliver more effective media buying. 

Scope3’s general manager EMEA Paul-Antoine Strullu, said that their agentic AI platform "fundamentally reimagines media buying through advanced AI". Reimagines is a hefty word, although it’s looking like harnessing agentic AI for media buying is definitely part of the future of advertising, as Ciaran O’Kane suggests in his latest MadTech sketch on the unbundling of ad tech. 

Considering the state of the climate crisis, we should always push for the adoption of more sustainable approaches. Plus, which advertisers wouldn’t want to cut down on ad fraud, decrease the chance of their ads ending up on a made-for-advertising site, and facilitate improved brand safety? 

Hold on… too good to be true? 

As with most technologies, agentic AI solutions also present some potential pitfalls. Aside from being fooled by agent washing, more significant risks exist. 

Like other types of AI, agentic AI has been known to hallucinate. Although agentic AI typically has a lower hallucination rate than say, a non-agentic LLM like ChatGPT-4 or Gemini, they will still make some errors, and these could actually be more complex and compounding. 

These hallucinations could have considerable consequences, causing advertisers to make decisions based on incorrect data leading to wasted budgets. In more extreme cases, misrepresented data could also lead to a company putting its reputation on the line. When it comes to brand safety: agentic AI systems may hallucinate that inventory is brand-safe, when in fact, it’s not. 

There is also a risk of media spend being directed to competitor buyers, given that despite its advancement, agentic AI still lacks the ability to reason, and fails to understand context or relationships. 

There have also been reports of agentic AI systems incorrectly removing attribution or measurement data. 

How ready is agentic AI to run the show? 

Is agentic AI ready to run the show? On some level, yes – given its fundamental purpose and design. These advances allow human teams to offshore complex tasks we would rather not do, particularly the data-heavy ones, which we are not built to handle. An agentic AI system, on the other hand, is specifically built for those.  

However, agentic AI’s advancement does not mean an absence of risks. We are yet to bear witness to any AI which does not make errors. Moreover, we are yet to create any AI which does not require some sort of human oversight. There is still so much human involvement needed. So, they can’t run the show completely on their own. At least certainly not for the moment.