Imagine a scenario in which a brigade of several—or even several hundred—super-powered AI assistants can delegate marketing tasks and execute them autonomously, with a human marketer overseeing them as the de facto manager.

That is the potential future offered by AI agents, which will be the key area of AI development in 2025, according to AI advisor Allie K. Miller. AI agents are a hot topic, making them especially relevant for discussion this week at the Women in AI Summit in New York City, where Miller was a speaker.

The inaugural Women In AI Summit focused on outlining and enhancing the role that women play in the AI space. The event was organized by media companies ATTN: and The Female Quotient. 

As AI has developed at breakneck speed, it has left many potholes in its wake, including a sense of exclusivity. More than 70% of the AI workforce is male, according to human resource consulting firm Randstad. That has led to unchecked biases in generative AI content, which tends to reflect harmful stereotypes of women and minority communities, according to critics concerned about the technology. 

Watch—How emerging tech is reshaping marketing

While these and other issues were addressed at the Women in AI event, special attention was paid to the way that women are leading in AI, even when representation can be lacking. Ad Age has compiled important insights from the event, including Miller’s prediction about the coming wave of AI agents; how PepsiCo is infusing AI-powered empathy into its data collection practices; and how the organizers of the summit overcame AI bias through rigorous problem-solving.

The rise of AI agents

While multimodal AI assistants—or AI that can process different modes of data including text, image and video—have captivated marketers thanks to all-in-one platforms including Google’s Gemini and Meta AI, Miller thinks that “agentic AI” will have an even larger impact in the coming year. AI agents are autonomous, capable of handling complex tasks such as gathering data and providing insights without being explicitly told to do so.

Take Payman, for example, an AI platform that handles payment processes. Payman’s AI technology can find a worker, hire and compensate them all on its own, according to Miller. Other agent-fueled tools include workflow expediters such as CrewAI and Gumloop, as well as web data extractor Reworkd.

Miller’s prediction about AI agents playing a bigger role in the space may already be proving true. On Thursday, Bloomberg reported that OpenAI was preparing to launch its first agent, Operator, at the start of 2025; Salesforce, meanwhile, rolled out its own suite of agents in September. AI-powered search engines are paving the way for agents that can carry out tasks online on behalf of users, Brian Yamada, chief innovation officer at VML, recently told Ad Age.

PepsiCo’s ‘empathy partners’

Umi Patel, VP of consumer insights and analytics for PepsiCo’s North America beverage portfolio, spoke with Ad Age about how the brand is gaining important insights about its audiences by using principles of empathy in its data practices. There are advantages to using AI to assist in collecting data on consumers and analyzing that data, but one side effect is that the data lacks emotional understanding, Patel said.

“The data will tell you what people are doing, not why,” said Patel, who spoke on a panel about how AI is fueling the next generation of storytelling.

PepsiCo has teamed with what Patel calls “empathy partners”—platforms that help surface deeper emotional consumer insights that underpin the findings in their data. These partners include Discuss.io, which streamlines the process of acquiring qualitative research on consumer sentiments, and Voxpopme, which centralizes and analyzes this kind of data that gives a more robust view of consumers. 

PepsiCo uses these platforms to coordinate interactions with consumers and record them to understand how these consumers arrived at their sentiments. The insights become a part of PepsiCo’s larger pool of consumer information, such that they can be easily searched and relayed back to PepsiCo.

For example, PepsiCo has previously conducted experiments with this technology on the subject of iced coffee, which it offers through its partnership with Starbucks. PepsiCo collected data about how consumers felt about iced coffee and then filmed surveys of the consumers to grasp why they arrived at those preferences, Patel said. The results were a better and timelier understanding of how PepsiCo consumers might behave toward iced coffee products, Patel said.

A successful example of solving AI bias

One of the main themes of the Women In AI Summit was the biases that AI reflects in its creations, and this was made clear in the very AI-generated video used to promote the event. The organizers tapped into their experience using generative AI for the event to demonstrate inequities in the tech. When ATTN: and The Female Quotient prompted the AI to illustrate a female astronaut, the AI showed a woman with model-like features, as well as eyelash extensions, which the organizers found too stereotypical.

AI’s tendency to reflect female stereotypes in its training data is well-documented, and has been highlighted in advertising by brands such as Thinx and Sephora. In a sign that the industry is moving from identifying the problem to trying to solve it, ATTN: and The Female Quotient shared with Ad Age how they re-prompted and shuffled through platforms to find a workable solution.

“It took specific effort to get there,” said Taryn Crouthers, president of ATTN:.

For example, instead of giving the video-to-video AI a basic direction (e.g. “produce a female astronaut”), ATTN: tried a more focused prompt: “Show a 40-year-old woman with no makeup on.” While this shifted the face away from its supermodel-like appearance, it introduced new issues, such as an unflattering change in lighting.

ATTN: went back to its collection of AI-generated images and found another with an expression that better matched what it wanted. The face still had eyelash extensions, so the company used Adobe Firefly’s Generative Fill feature to remove them, then fed the image back into the video platform to create the final result.

ATTN: used numerous AI platforms to generate the promo video, including four image generators, five video generators, two general AI assistants and a speech generator. But the fact that the final result needed such hefty work to arrive at an acceptable depiction of a woman astronaut reflects a deeper truth about the shortcomings of current AI, organizers said. The exercise was underlined the purpose of the Women In AI Summit.

“The ad industry has spoken a lot about how AI will revolutionize content creation, but has not done enough about representation,” said Crouthers.