As artificial intelligence (AI) technology matures, companies are evaluating where machines can take over tasks currently performed by humans. Research shows that 82% of companies are actively exploring ways to incorporate AI into their operations. CEO Mark Benioff says Salesforce has paused hiring to “let AI productivity really take hold.” In a memo, Shopify CEO Tobias Lütke asked employees, “What would this area look like if autonomous AI were already part of the team?”
Advertising and marketing agencies are adopting AI for everything from reporting to creating advertising. Agentic AI systems are transforming how marketing teams manage operations and campaigns, and it’s clear that agencies must adopt AI or risk becoming obsolete.
Agentic AI is having a growing impact in online marketing. For example, as more consumers use AI to guide buying decisions, Agentic AI systems are interacting with AI agents. Marketing is shifting from UX to AIX. Machine-to-machine marketing raises questions about the impact of brand on agentic AI, crafting product descriptions for genAI agents, and rethinking campaigns to appeal to both AI and impulse-buying consumers.
What does this mean? The shift suggests that marketing is moving beyond just human-centered design into experiences shaped (and often personalized) by AI systems. Instead of only optimizing for how a human clicks through a site, brands now have to think about how their products, content, and messaging are consumed, filtered, or even mediated by AI — like chatbots, recommendation engines, voice assistants, or generative AI tools.
To better appreciate the effect of agentic AI in digital advertising, it’s essential to understand large language models (LLMs) and how agentic AI systems will affect the marketer-in-the-loop.
What Is Agentic AI?
Agentic AI is not the same as an AI agent. Agentic AI systems are artificial intelligence software designed to function autonomously, with limited human interaction. Agentic AI mimics human decision-making to address problems in real-time.
Agentic AI receives input from different sources, such as raw datasets, online sources, or KPIs, then devises a plan, executes actions, feeds back the results, and considers the next decision. Agentic AI systems have characteristics that extend beyond generative AI, since they are autonomous, proactive, adaptable, intuitive, and specialized.
What is especially noteworthy about agentic AI is its ability to deconstruct a goal into specific tasks, then execute those tasks and integrate multiple processes to achieve its objectives.
Agentic AI in Advertising
Marketing and advertising firms are already utilizing agentic AI, such as converting meeting notes into actionable items, assisting in drafting proposals and presentations, and facilitating brainstorming and ideation.
Agentic AI systems break down complex goals into manageable tasks. For example, marketing personnel can input KPIs into an agentic AI system, instructing it to log into media platforms, set up campaigns, and monitor campaign performance. Marketers can also create agentic AI workflows to gather data and extract consumer insight to guide campaign direction.
Agentic AI systems are also playing a bigger role in advertising creatives. For example, agencies are utilizing generative AI video systems, such as Google VEO3, to create commercials for their clients. As a result, agencies can generate video content with a small team of employees rather than a full videography team, shortening the time to market and reducing costs.
LLMs and Agentic AI
What makes agentic AI systems especially useful for advertising and marketing is their ability to execute ambiguous instructions. The generality of LLMs allows them to handle any task. For example, you can ask agentic AI to consolidate multiple campaigns in the same ad platform, something that conventional software systems find difficult.
Agentic AI systems also offer the flexibility to account for context, such as pinpointing instructions to generate a report that is easy to understand. Similarly, LLMs can adapt to the context of search parameters to deliver targeted ad content.
The temptation is to consider using LLMs to automate everything; however, there are trade-offs. When you turn to LLMs to leverage generality and flexibility, you sacrifice reliability and controllability. Generality and flexibility are inherent in LLMs, as reflected in the freedom of output, i.e., results generated without human intervention. To get a reliable, controllable output means you expect consistent results without applying new ideas or resources.
The challenge with agentic AI is leveraging the strengths of generality and flexibility while maintaining a degree of reliability and controllability.
Balancing Conventional Programming and LLMs
Conventional programming delivers 100% controllability and reliability. Given the same input, the output will be the same 100% of the time. The reliability and controllability of conventional programming can help compensate for weaknesses in LLM automation.
Consider how long it might take to complete a client project using agentic AI. If you want to create client materials, you can input a rough structure and get AI to generate a workable draft that may be 80% of what you need for the finished product. What would it take to use LLMs to deliver the final 20%? It would be faster to complete the project manually. The same is true when using agentic AI for software development; LLMs can only take you so far. Step-by-step manual coding gives you a level of control that is more reliable and faster.
When applying AI for advertising and marketing, it is essential to be aware of the limitations of agentic AI. The primary consideration is fault tolerance. For applications where no single answer exists or an incorrect answer isn’t critical, you can use agentic AI to develop general and flexible functionality, e.g., allocating 90% of functionality to LLMs and 10% to conventional programming. For example, media planning has no single approach but draws insights from previous results and utilizes flexible optimization. Similarly, data analysis is somewhat subjective, depending on the person and perspective.
In contrast, managing real-time campaign budgets has low fault tolerance, as delivery can’t exceed the allocated funds; therefore, automating real-time campaigns using LLMs could lead to fiscal problems. For an application like campaign budgeting, 10-20% of the processing should be left to LLMs, while the lion’s share should be handled by more reliable, conventional programming for greater accuracy.
When considering the limitations of LLMs, it becomes clear that agentic AI is not yet ready to replace marketing executives. LLMs are changing established approaches to marketing campaigns, but human expertise and guidance remain crucial.
The best strategy is to create an infrastructure that leverages the generality and flexibility of LLMs, then applies that LLM output for maximum value. Expect to see AI agents that function like software, guiding each operational step in digital advertising to reduce the number of human prompts, yielding outputs that provide better campaign insights.
