The promise, pressure & performance of agentic AI: ‘We’re looking for unicorns’

That gap between the industry’s narrative and practitioners’ reality is a recurring theme. Emily Lesinski, group director of analytics and insights at Rightpoint, puts it simply: “There’s been a lot of hype, but not as much thought put into the data structure. If you don’t organize data in a way AI can digest it, it won’t deliver.”

Raising the stakes

Still, hype has its upsides. For Lucy Darbon, data strategy director at Armadillo CRM, AI’s star power has finally elevated long-overdue projects.

“Big data projects that have been going on for as long as I’ve worked with clients are suddenly the most important business projects,” she says. “AI is giving more precedence to these data projects, but if you get the structure wrong with the data, agentic AI is going to make every wrong choice for you and drive the worst profitable decisions.”

That’s the paradox: AI is exciting clients about personalization at scale – “I can literally give everyone on this call a bespoke experience,” says Darbon – but it’s also forcing them to confront their data quality and governance gaps.

Clients often jump straight to the AI solution before they’ve done the gap analysis, explains Tina Allan, global chief data and intelligence officer at FCB: “We need to map the structure first, see the gaps, see interesting ways where we can get different approaches or understandings of that data of your customer and then decide how AI can help.”

Bridging the literacy gap

This is exposing a critical disconnect inside organizations: the gap between those who engineer data and those who make decisions with it.

Ebrahim Bakhtar, senior analytics director at PMG, sees agentic AI as a possible bridge. “There’s a gap in data literacy where a lot of the juice doesn’t get squeezed,” he says. “That’s where I’m optimistic that agentic AI could step in,” giving marketers more self-serve analysis – querying databases in natural language and getting useful outputs – without always needing a business analyst in between.”

Lesinski agrees: “Historically, you were three steps removed from the data and the decision-making. Now, agentic AI can empower marketers themselves to ask questions and bring in macro factors – like weather patterns affecting energy consumption – directly into models.”

Unicorns wanted

The bigger concern is the talent needed to make AI work. “We’re looking for unicorns – people that can bridge all of that,” Allan explains. “We’re going from siloed expertise to T-shaped talent. I need people who can be client-facing, understand how to build agents for clients, but [at the same time] respect the data, understand the security and parameters.”

It’s a challenge echoed by many around the table: while AI tools may promise to automate, the demand for human judgment, contextual creativity and cross-disciplinary skills is only growing.

If there was one consensus, it’s that AI can’t run unchecked. “For safety and bias reasons, we have to have some high-level understanding – at least enough to poke holes in it,” says Lesinski. “Otherwise, we’ll just precipitate the same issues.”

Hauptman is even more emphatic: “If you’re going to rely on AI to do things, you have to be able to do them yourself first or else you won’t know what it’s doing and how it works.”

While brands’ expectations of agentic AI are high, data analysts – the experts at the data coalface – say they’re yet to see its full promise in their day-to-day operations. Today’s tools are “great machine learning for marketers packaged as AI,” says Darbon. They may be accelerating workflows, but they’re not yet the autonomous colleagues promised in the hype cycle.

The takeaways for 2025

Agentic AI is reshaping the conversation around data, but not in the way most headlines suggest. Rather than replacing marketers, it’s highlighting the urgent need for better structures, smarter governance and more adaptable talent.

For the roundtable, three lessons stood out:

• Don’t skip the foundations: AI won’t fix bad data structures; it will amplify them.

• Close the literacy gap: Self-serve AI tools can empower marketers, but only if they know enough to question the outputs.

• Invest in talent: From “unicorn” hybrids to creative strategists who bring the “magic and logic,” people remain the differentiator.

As the experts put it, we have to bring curiosity, creativity and responsibility into every AI conversation. AI may be the shiny new assistant, but humans are still the boss.

Read part one of our ‘Breaking Bad (Data Habits)’ series.