By Michael Fuchs, Head of Strategy & Planning, Data & Technology  

Marketing organizations are asking the wrong AI questions. It’s not which AI agents to deploy, but what sits beneath them, and who owns it.  

In a recent POV, I argued that marketing and advertising are entering a new operating era shaped by persistent AI, compressed execution cycles, converging decision systems, and increasing demands for strategic control. That transition has a structural implication most organizations are moving through without even recognizing.  

Over the past year, enterprise AI adoption has moved rapidly from experimentation into operational proliferation. Teams are now deploying agents across multiple vendors, orchestration environments, workflow systems, cloud platforms, and governance models simultaneously. In some large enterprises, internal estimates already suggest agent counts reaching into the thousands across disconnected business units, platforms, and operational environments.  

What’s emerging is not simply agent growth, it’s the early formation of fragmented intelligence estates — disconnected layers of memory, orchestration, governance, workflow coordination, and accumulated operational learning.  

Most current industry discussion remains focused on the agents themselves: their capabilities, reasoning quality, autonomy, or model performance. But the more consequential strategic layer sits behind the agents: the harness infrastructure responsible for orchestrating workflows, coordinating permissions, managing memory, enforcing governance, maintaining observability, and connecting systems, tools, and models together into persistent operational environments.  

That distinction matters because modern marketing systems no longer operate as isolated execution tools. Increasingly, they influence everything from audience strategy, media allocation, campaign optimization, and creative iteration to budget development, analytics, and operational decisioning.  

In other words, these systems are beginning to shape the operational intelligence layer through which growth decisions are made. And that is where accumulation begins.  

From Workflow Automation to Intelligence Accumulation  

Much of the first wave of enterprise AI adoption has been framed around productivity and workflow automation. That framing is directionally correct, but incomplete.  

As agentic systems move deeper into operational environments, they do not simply execute tasks. They begin accumulating context: prior decisions, optimization histories, workflow patterns, organizational heuristics, approval structures, performance outcomes, and institutional memory  

Over time, these systems increasingly function less like temporary tools and more like persistent operational intelligence environments. That changes the nature of the strategic question.  

The challenge is no longer simply: “Which models should we use?”  

Increasingly, the question becomes: “Where does organizational intelligence accumulate, who governs it, and how portable does it remain over time?”  

This is where harness architecture becomes strategically important.  

Two Architectural Decisions Enterprises Are Already Making  

Many organizations are implicitly making two foundational decisions, often without fully recognizing the long-term implications.  

1. Should the Harness Layer Be Owned or Rented?  

As systems accumulate workflow intelligence, operational context, and institutional memory, the distinction between renting AI capability and owning persistent intelligence infrastructure becomes strategically significant.  

Beyond simply a software procurement question, it increasingly becomes a question about operational control, portability, governance, dependency, long-term differentiation, and accumulated organizational learning  

If enterprise intelligence compounds inside commercial orchestration layers controlled externally, organizations may eventually discover that switching costs extend well beyond models or workflows themselves.  

The accumulated intelligence layer may become the more valuable asset.  

2. Should Persistent Intelligence Be Centralized or Federated?  

The second question may ultimately become even more important: Should organizations operate through a single coordinated persistent intelligence layer, or through multiple federated harnesses accumulating disconnected memory, governance, workflows, approvals, and operational histories over time?  

At small scale, federated experimentation appears manageable — and in many cases desirable. Organizations should continue experimenting across models, vendors, and agentic approaches. Flexibility matters.  

But experimentation across systems is different from fragmenting your intelligence infrastructure. This distinction is increasingly important.  

APIs, interoperability standards, and MCP services may help systems communicate or access shared tools and data, but they don’t inherently solve for unified governance, memory coordination, workflow continuity, audit inheritance, approval harmonization, accumulated organizational learning, and operational accountability  

Those are harness-level concerns.  

In marketing and advertising environments, where agencies, clients, media systems, creative operations, analytics environments, commerce platforms, and measurement frameworks are deeply interconnected, coordinating persistent intelligence across multiple disconnected orchestration layers will only get harder.  

Technically interoperable systems are not necessarily operationally coherent systems.  

Why Advertisers Should Care  

For advertisers, this is not simply a technical architecture discussion.  

Over time, these systems may increasingly influence:  

• How audience intelligence accumulates  
• How media decisions are optimized  
• How campaign learnings persist  
• How operational workflows evolve  
• How measurement logic develops  
• Who controls the intelligence shaping growth decisions across the ecosystem  

This creates an important strategic tension for the industry.  

Agencies need persistent intelligence systems to compete effectively in more compressed, AI-mediated operational environments.  

As those systems deepen, advertisers will face harder questions about what is actually owned. Not just data, but the operational intelligence that has accumulated around how that data is used, interpreted, and acted upon.  The future debate may not simply center on media transparency or data ownership.  

It may increasingly center on intelligence ownership, whether that sits with the advertiser, agency, or platform coordinating them both.  

The Emerging Strategic Control Point  

This isn’t an argument against agents, models, or innovation velocity. Nor is it necessarily an argument for rigid centralization.  

It is an argument that the industry is underestimating where long-term strategic control points are actually forming.  

The agents themselves will likely become more replaceable than the persistent intelligence layers coordinating them.  

Over time, that is where organizational memory accumulates, governance frameworks harden, workflows converge, operational heuristics develop, and institutional intelligence compounds.  

The organizations that navigate this transition best will be the ones that recognize this architectural distinction earliest.  

As agentic systems move deeper into the operational core of marketing and advertising organizations, these decisions around harness architecture, portability, governance, and intelligence coordination may become increasingly consequential.  

The next major competitive battleground will not simply be model access or agent sophistication.  

It may will be the architecture through which enterprise intelligence accumulates, compounds, governs, and transfers over time.