The real AI challenge for WPP isn’t scale, it’s control 

This isn’t theoretical. WPP is already deploying agents to handle media planning, content generation, analytics and optimization. For now, their capabilities are limited to helping human employees without agency to full autonomy to access systems and data sources for safety reasons. But the promise of agentic AI involves coordinating numerous AI systems, orchestrating multiple intelligent systems to connect agents across teams, clients and platforms. Without it, the risk of conflicting behavior, redundancy, or outright failure goes up fast. 

In other words, part of WPP’s bet is that the real competitive edge won’t come from using AI so much as it will come from responsibly managing, integrating and scaling it before rivals catch up. 

Forget chatbots. These are adaptive systems

Hulme wants to nix the idea that AI is merely software that mimics human tasks. His focus is on something more dynamic. He draws from a 1980s concept called “goal-directed adaptive systems” — a framework for AI that learns, adapts and evolves through experience instead of static code.

“Most of the systems in production up to now don’t adapt themselves,” he said. “They’re not adapting to a changing world. They’re static. What I think generative AI and agents in particular allow us to do is actually to build systems that are adapting and learning. But for me, the true power of AI are systems that are always constantly getting better and building adaptive systems is extremely hard.”

That’s the urgency. AI agents are starting to operate independently, in live business environments, and many companies are shipping them without proper testing or oversight. WPP is trying to get ahead of that curve to create its own safeguards before the industry walks into a credibility crisis.

One way Hulme is planning for the future is with neuromorphic computing — AI systems modeled on the structure and function of the human brain. Instead of training AI in the traditional way with static datasets, a research organization called Conscium — partially funded by WPP and co-founded by Hulme — is exploring simulated environments to shape their behavior. The goal is to teach agents using dynamic and experience-based learning to simulate environments to teach agents behaviors learned through experience and outcomes. Rather than hard-coding ethics, it shapes them by designing systems where cooperation leads to success — and survival. (Unlike a scenario where AI agents learn to survive like characters in William Golding’s classic novel, Lord of the Flies.)

“I can create an environment or a game where agents have to lie, cheat, steal, and kill each other to survive,” Hulme said. “Or I can create a game or environment where you have to cooperate, be altruistic, be sacrificial and share to survive.”

One of Conscium’s projects is an app called “Moral Me,” which crowdsources moral dilemmas from users to help train AI agents on a wider range of human moral values. The logic is simple: as people hand more decision-making to personal AI, marketers and technologists will need to learn how to influence the agents themselves. 

Influence the agent, not the audience

It’s not just about personalization, it’s also about building systems that can anticipate intent. That’s the next frontier: agents that know what someone is about to want and marketers who can respond in real-time. Hulme’s team is integrating data from various knowledge graphs and data sources through WPP’s Open, the holding company’s privacy-first campaign planning tool powered by federated learning.

Holding companies aren’t the only ones betting on AI agents. A January research note by analysts at William Blair noted Meta is well-positioned thanks to its scale and fast-moving business AI tools. Ralph Schackart, one of the analysts who authored the William Blair report, said Meta and other companies could benefit from training chatbots on CRM data for customer service, sales and marketing.

“I think the CRM data is going to be honest, I don’t want to say magic, but the differentiation here,” said William Blaire analyst Ralph Schackark. “I’ve heard some people speculate it’s like a 10x step up in data for these bots when you’re able to train with the CRM data.”

Schackart thinks advertisers could end up relying on holding companies like WPP to help manage AI agents on their behalf. According to William Blair’s Q1 2025 survey, 65% of ad execs rated generative AI as “very” or “extremely effective,” citing clear gains in performance.

Governance and risk

More complex systems also require more guardrails. Hulme sees a shift to dynamic, autonomous agents capable of learning and evolving in real time—which makes them riskier. An untested agent misallocating ad spend is one thing; one that adapts and spreads flawed logic or unethical use of data across a platform is another.

That requires a broader and more nuanced perspective of what does and doesn’t apply in terms of various types of machine learning and optimization algorithms. For audience perspectives, performance predictions, and media planning. Hulme believes under-tested agents could trigger failures that prompt future regulation across the AI industry.

“What’s happening is lots of people are going to be creating agents without actually testing them and then putting them out there,” Hulme said. “They’re going to be going wrong. And not only that, they can adapt themselves in the real world. So they need constant verification. They need constant testing.”