James Bond, Librarian.

The 'Almost Agent Hub' doesn't have the same ring, does it?

But that's what WPP's CTO actually called it. Even he won't fully commit.

So why does the press release?

In launching their "Agent Hub" this week, WPP's CTO Stephan Pretorius, when describing their four "Super Agents" to Campaign, he called them a move from "things that do tasks to almost agentic products."

Almost.

So why does every other sentence in WPP's breathless announcement use "agentic" and "autonomous" or "the power of generative and agentic AI" when describing tools that answer questions about brand equity data when you ask them?

Here's what I don't understand: WPP built something genuinely valuable. They made 30 years of Brand Asset Valuator data accessible through natural language queries. They systematised Ogilvy's behavioural science frameworks so anyone can apply them consistently. And perhaps most importantly, they have democratised access across their organisation where 75,000 employees can access institutional knowledge that was previously trapped in PowerPoint decks and individual experts' heads.

This is meaningful work. And it addresses the greatest gap in AI adoption: the human one.

So why make the overclaim? Why not just get on with it? Well there is a pattern of this. Adding the extra 5% that moves the announcement beyond the truth. In doing so, bad clients may be interested but good clients will roll their eyes and great clients laugh.

In 2022, WPP launched the Metaverse Foundry through Hogarth, a "global network of over 700 creatives, technologists, and developers" with a Global Executive Creative Director of Metaverse. They positioned metaverse capabilities as the next essential infrastructure for clients. By 2023, it had quietly disappeared. No announcement. No explanation. Just gone.

Then, after a series of overclaims since 2024 about their proprietary AI tools and workflow management approaches, in October 2025 WPP came clean and expanded their Google partnership where they would build on Google's AI-optimized technology stack (let's be clear, they already were).

Now it's January 2026 and suddenly we have "Agent Hub" with "advanced agentic AI marketing expertise" that "scales WPP's collective intelligence" through "AI agents."

I'm not criticising pivots away from metaverse hype. That was smart. I'm not questioning the Google partnership. That makes strategic sense. I'm asking why we need the terminology inflation when the actual product is already good.

Let me be specific about what WPP launched, because the 95% truth is a good thing.

The Brand Analytics Agent provides conversational access to Brand Asset Valuator data spanning three decades. Instead of requesting custom research or digging through historical reports, anyone at WPP can now ask questions and get back brand equity insights immediately. That's valuable institutional knowledge, democratised.

The Behavioural Science Agent applies Ogilvy's proven frameworks systematically. It ensures teams consider psychological principles consistently rather than relying on whoever remembers to apply them. That's quality control through systematic knowledge application (and I'd quite like to have a play with this one.)

The Analogies Agent searches across industries for parallel situations. It expands the reference library beyond what individual strategists carry in their heads. That's organisational memory serving creative ideation.

Creative Brain structures access to 150 years of WPP creative intelligence covering campaigns, approaches, techniques that worked. It's institutional knowledge as sparring partner rather than buried archive.

These systems make WPP's accumulated expertise more accessible, more consistently applied, more useful to more people. The 75,000 employees using WPP Open aren't wrong to find value here. The clients who've adopted it including Coca-Cola, Nestlé and others aren't mistaken about its utility.

So why not stop there? Why add the 5% that transforms "excellent knowledge management" into "autonomous agents"?

After building Simbioniq last year, I've learned that explaining AI agents requires cutting through vendor terminology to get at the fundamental distinction.

Think of James Bond.

M gives Bond an objective: "Stop the villain from [specific threat]." That's the meta-instruction. Then Bond disappears. He doesn't return for approval on each decision. He doesn't go back over the research on similar missions conducted by other 'double-ohs'. He doesn't ask permission before choosing to infiltrate the casino rather than the yacht. He doesn't wait for M to prompt him about checking the villain's financial records or recruiting an informant.

Bond autonomously determines the strategy, adapts when plans fail, tries alternative approaches, makes independent decisions toward the goal. He owns the outcome. That's agency.

Now imagine Bond as a librarian.

M asks: "What do we know about the villain's financial patterns?" The librarian searches the files and returns what's available. Excellent service. Genuinely helpful. But when M doesn't ask the next question, the librarian waits. When the first search doesn't reveal enough, the librarian doesn't independently decide to cross-reference shipping manifests or property records. The librarian responds to queries. Doesn't pursue objectives.

Here's what technical capability actually requires to qualify as "agentic":

1. Autonomous Decision-Making. The system makes choices without constant human input. Bond decides whether to infiltrate through the kitchen or the guest entrance. He doesn't radio M for approval. An agent determines HOW to act, not just executes what you tell it to do.

2. Goal-Driven Behaviour. The system pursues objectives, not just responds to commands. "Stop the villain" vs. "Search the database for villain." One owns achieving the outcome. The other executes a specific task when asked.

3. Multi-Step Problem Solving. When the casino infiltration fails, Bond tries the yacht. When that's blocked, he recruits an insider. The system handles complex workflows autonomously—breaking down objectives, trying alternatives when initial approaches fail, coordinating across multiple steps without human orchestration at each stage.

4. Learning and Adaptation. The system improves based on outcomes. Bond learns the villain's patterns and adapts his approach. Real agents don't just retrieve best practices—they develop new practices based on what works.

5. Outcome Ownership. This is the critical distinction. When the first approach doesn't work, the system tries alternatives. When it lacks information, it seeks it out. When it encounters obstacles, it routes around them. The system is responsible for achieving the result, not just executing the task.

When a vendor pitches "agentic AI" or "autonomous agents," here's how to wade through the bullsh*t in 5 simple questions :

The Instruction Test: "After I give you the initial goal, do I need to provide step-by-step direction, or does the system independently determine how to achieve it?" If the answer involves "prompts" or "queries" or "requests," you're looking at a sophisticated assistant, not an agent.

The Failure Test: "When the first approach doesn't work, what happens?" Agents try alternatives autonomously. Assistants wait for your next prompt. Both are useful, but you need to staff and price accordingly.

The Coordination Test: "Can it orchestrate across multiple systems to achieve a goal, or does it operate within a single domain requiring human integration?" Real agents coordinate independently. Assistants provide better information for humans to coordinate.

The Adaptation Test: "Show me specific examples of how performance changes based on outcomes." Agents learn and improve their strategies. Assistants provide consistent responses to similar inputs.

The Ownership Test: "When this system encounters an obstacle preventing goal achievement, does it independently find alternatives, or does it report the obstacle and wait for my guidance?" This is where the distinction becomes clearest. Agents own outcomes. Assistants execute tasks.

Most systems marketed as agents will be honest when you ask these questions directly. The overclaim lives in the marketing language, not the technical conversations.

Here's what WPP could lead with that's both accurate and compelling:

"We've made 30 years of brand equity research instantly accessible to 75,000 people through conversational interfaces. We've systematised our best behavioural science thinking so it's consistently applied, not sporadically remembered. We've turned institutional knowledge from individual expertise into organisational infrastructure."

This is interesting! This addresses a real problem CMOs face: brilliant frameworks trapped in expert heads, historical research buried in archives, best practices applied inconsistently because nobody remembers them during deadline pressure. (It is also what all holdcos are doing with sophisticated prompting, they just don't announce it).

When WPP overclaims by that critical last 5%, every other vendor thinks they need to match. When every vendor overclaims, CMOs lose the ability to distinguish genuine capability from sophisticated marketing. The market for excellent knowledge management tools is enormous. The market for genuine autonomous agents is still emerging. Both are valuable. But conflating them helps nobody.

And the industry loses the language precision needed to evaluate meaningfully different capabilities.

Gartner predicts 40% of enterprise applications will embed AI agents by end of 2026, up from less than 5% in 2025. But what does "embed AI agents" actually mean if every sophisticated prompt system calls itself an agent?

PwC's survey shows 79% of organisations report adoption challenges due to coordination complexity, cost uncertainty, and unclear business value. That's not surprising when "agent" describes everything from autonomous multi-system coordinators to searchable databases.

When CMOs can't distinguish capability levels through vendor language, they resort to brand recognition and pricing as quality signals. Premium pricing becomes the marker of genuine capability rather than accurate description of what the system actually does.

Here's what I genuinely don't understand: WPP has an incredible institutional knowledge base in holding company marketing. The Brand Asset Valuator data is powerful. Ogilvy's behavioural science practice is legitimately excellent. The creative history is astonishing.

In short, their past is much better and stronger than their present, so it makes sense to access that past better, to help the present.

WPP built something valuable. They should call it what it is and let the actual capability drive adoption: very very good librarians of a very very good library. The 'Almost Agent Hub' might not have the same ring, but at least it would prepare clients for what they're actually getting, which is good enough to succeed on its own merits.

Brandflow is written by Justin Billingsley, who has spent his career on all three sides of the industry's table: senior client, global agency leader, technology founder. First published 8 January 2026 in the Brandflow newsletter on LinkedIn.