The Great Bifurcation: Where PMM Goes From Here

Last Tuesday at Cannes, Forrester released a report in partnership with the 4As—the agency trade group—that landed with less fanfare than it deserved, perhaps because the headline finding was so uncomfortable that people didn't quite know what to do with it. Nine in ten US marketing agencies now use generative AI, and half have moved on to agentic AI for campaign execution, which represents an adoption curve steeper than anything I've seen in twenty years of watching marketing technology cycles. But the punchline—the finding that matters for anyone who runs a PMM team or sits on one—is that all this adoption is actively undermining marketing effectiveness, creativity, and long-term brand growth.

Not because the technology doesn't work. Because everyone is using it for exactly the same thing.

The agencies that have embraced AI most aggressively are optimizing for cost reduction and efficiency—which makes sense from a margin perspective, and which their CFOs no doubt appreciate—but the consequence is that they're all producing more content, faster, at lower cost, and it all looks increasingly indistinguishable from what everyone else is producing. The 80% that's competent and interchangeable, the content that could have been written by any of a hundred people or by no person at all, is exactly the content that AI generates most easily. And when the marginal cost of producing "good enough" drops toward zero, the supply of "good enough" explodes while the demand—the finite attention of buyers who can only read so many whitepapers and watch so many product videos—stays exactly where it was.

The Fork in the Road

The same week Forrester dropped that report, Adobe announced partnerships with WPP, Accenture Song, Omnicom, and Stagwell to deploy what they're calling "agentic content supply chains"—integrated workflows that connect brand intelligence, content generation, activation, delivery, and measurement into a coherent loop that runs with minimal human intervention. Amazon released data showing that Rufus, their AI shopping assistant, appeared in 38% of shopping sessions during the 2025 holiday season, and those sessions converted at 3.5x the rate of non-assisted sessions on Black Friday. Klaviyo published research showing that 32% of consumers trust brands less when they detect AI-generated content—which means that while marketers are racing to produce more synthetic content, consumers are simultaneously developing antibodies to it.

Put these signals together and you start to see something that looks less like "AI is changing PMM" and more like "AI is splitting PMM into two fundamentally different jobs that happen to share the same title."

On one side, you have what I'd call the AI System Architects—people who orchestrate AI systems rather than compete with them, who train models on brand voice and positioning frameworks, who design the prompting workflows that connect content generation to activation to measurement, who build the infrastructure that makes agentic marketing actually work at enterprise scale. These people need to understand both the marketing strategy and the technical architecture, and they're commanding premium compensation to do it. The freelance consultants are billing $50-200 an hour; the full-time roles run from $90K to $250K depending on seniority and what they can actually ship. Adobe's Cannes announcements make clear that this is where the major holding companies are placing their bets—not just using AI tools, but architecting AI systems.

On the other side, you have what might be called Human Premium Specialists—people whose value comes from judgment, empathy, and narrative craft, the things AI cannot replicate even as it becomes more capable at execution. Strategic decisions under uncertainty, when the data is ambiguous and someone has to make a call. Customer insight that comes from having been in the room when the deal fell apart, from understanding not just what the CRM says but what actually happened. Stories that stick because they're grounded in lived experience rather than synthesis—the Porsche on the tarmac at the Atlanta airport, the specific question the CFO asked that changed the direction of the conversation, the energy in the room when the pilot results came back.

And in between those two poles—in what used to be the comfortable middle of the PMM career ladder—there's a tier of work that's getting steadily compressed. Drafting decks from templates, formatting one-pagers, running basic competitive analysis, doing the routine campaign setup and standard content production that used to fill calendars and justify headcount. This is the work that's "good enough," the competent execution of standard tasks, and it's precisely where AI is most disruptive because it's precisely what AI does most easily.

This isn't speculation. The evidence is accumulating faster than most people realize.

The Efficiency-Effectiveness Trap

The Forrester report deserves a closer look, because the finding isn't just that agencies are adopting AI—everyone knew that was happening—but that the way they're adopting it is systematically undermining the outcomes they're supposed to be producing. The agencies most focused on AI as a cost-reduction play are producing more content faster and cheaper, which looks great on a productivity dashboard, but they're producing content that looks exactly like what every other agency is producing with the same tools and the same prompts optimizing for the same efficiency metrics.

The efficiency gains are real, and I don't want to dismiss them—there's genuine value in being able to produce a first draft in minutes rather than hours, in automating the mechanical parts of content production so that humans can focus on the parts that require judgment. But efficiency without distinctiveness is a race to the bottom. When everyone has access to the same tools producing the same outputs, the competitive advantage shifts to whoever can do something the tools can't—and right now, what the tools can't do is be genuinely distinctive.

We cover this dynamic extensively in Unit 6 of the Academy curriculum, where we talk about the Content Collapse and why 80% of marketing content is becoming invisible. The 80% that's competent and interchangeable is exactly the content AI produces most easily, and when the marginal cost of producing it drops toward zero, the only content that breaks through is the 20% that's genuinely differentiated—the content that reflects specific experience, contrarian takes backed by evidence, stories nobody else can tell, technical depth that can't be faked.

The Rise of the AI Orchestrator

Adobe's Cannes announcements tell the other half of the story. When WPP, Accenture Song, Omnicom, and Stagwell all announce partnerships to deploy "agentic content supply chains," they're not just adding another AI tool to the stack—they're building something qualitatively different, an orchestration layer that connects brand intelligence and content generation and activation and measurement into an integrated system that can run with progressively less human intervention.

The skill this requires isn't prompting ChatGPT well, though that's table stakes at this point. It's understanding how to design the workflows, how to train the models on positioning frameworks and brand voice, how to build the integration points between systems, how to set the guardrails that keep the automation from going off the rails. It's systems architecture applied to marketing, and the people who can do it credibly—who understand both the marketing strategy and the technical implementation—are becoming increasingly valuable.

I should note that I'm skeptical of the "prompt engineer" hype as a standalone career path—the models are getting good enough at inferring intent that the alpha in clever prompting is eroding quickly—but the broader capability of designing AI-native workflows, of understanding how to integrate these systems into marketing operations, of being the person who makes agentic marketing actually work rather than just work adjacent to it, that capability is real and increasingly premium.

The Trust Penalty

The Klaviyo research points to a constraint that most discussions of AI-generated content tend to skip past, which is that consumers aren't passive recipients of whatever marketers produce—they're developing their own pattern-matching capabilities, their own sense of what synthetic content looks and feels like, and their own preferences about how much of it they're willing to tolerate.

Nearly one in five consumers now report encountering low-quality or generic AI content from brands on a weekly basis—which suggests the volume problem is real and getting worse—and while 61% claim neutrality toward AI-generated content in principle, 32% say that detecting AI content makes them trust brands less. That's not a trivial number. It means that every time your AI-generated content trips the "this feels synthetic" detector in a buyer's brain, you're paying a trust tax that compounds over time.

This is where the Human Premium earns its name. Not by avoiding AI entirely—that's neither practical nor desirable—but by knowing when human voice, human judgment, human craft are worth the investment. The stories that stick are the ones grounded in experience rather than synthesis, with the specificity that comes from having been somewhere and paid attention. An AI can generate a competent analogy, can produce a hypothetical scenario that illustrates a point, but it can't tell you what the energy was like in the room when the pilot results came back, can't give you the exact question the skeptical CFO asked that changed the direction of the conversation, can't draw on a reservoir of tacit knowledge that only exists in the heads of people who were actually there.

The AI Discovery Shift

The Amazon Rufus data points to a structural change in how buyers discover and evaluate products that I think most marketers haven't fully internalized yet. When an AI shopping assistant appears in 38% of sessions and those sessions convert at 3.5x the rate of non-assisted sessions, something fundamental is shifting in the buyer journey—the discovery and consideration phases are increasingly happening through AI intermediaries that synthesize and recommend rather than list and link.

This matters for PMM because it means your positioning doesn't just need to work for human readers anymore—it needs to work for AI systems that will parse it, summarize it, and potentially recommend or not recommend you based on how clearly you've articulated what you do and for whom. If your positioning is vague enough that an AI can't accurately represent it, if your content is structured in ways that AI systems can't parse effectively, you become invisible at the exact moment when buyers are making decisions about which vendors to consider.

We call this dynamic GEO—Generative Engine Optimization—and it doesn't replace SEO so much as layer on top of it. You're writing for two audiences now: the human who might click through and read, and the AI that might cite you or summarize you or recommend you in response to a buyer's query. The good news is that what works for AI systems—clarity, specificity, structured claims, verifiable evidence—also tends to work for human readers who are short on time and patience. The skill is learning to satisfy both.

What This Means

If you run a PMM function—or you're a PMM thinking about your own trajectory—the question isn't whether to use AI. That question was settled a year ago, and anyone still debating it is already behind. The question is which side of the fork you're building toward, which capabilities you're developing, and whether you're consciously making that choice or letting it happen to you by default.

Some PMMs will become AI System Architects, the people who design the workflows and train the models and build the orchestration layer that makes agentic marketing work at scale. They'll need to develop real technical fluency—not necessarily writing code, but understanding how these systems work well enough to design effective integrations—while maintaining the marketing judgment to know what the systems should be optimizing for. The path into this role runs through experimentation, through building things that work, through developing the portfolio of evidence that you can actually ship AI-native marketing at enterprise scale.

Others will become Human Premium Specialists, the people whose judgment and empathy and narrative craft justify their role in an AI-abundant world. They'll need to double down on the things AI can't replicate—the customer intimacy that comes from years of real conversations, the stakeholder alignment that requires reading rooms and building coalitions, the strategic judgment that emerges from having been wrong before and learned from it, the storytelling capability grounded in experiences that actually happened to them. The path into this role runs through depth, through developing real expertise in domains that matter, through building the reservoir of tacit knowledge that makes judgment possible.

Both paths are viable. Both command premium compensation. Both have room to grow as the technology evolves.

The path that's closing—not all at once, but steadily—is the middle: competent execution of standard tasks without distinctive value on either dimension. The PMM who drafts good-enough positioning from templates, who runs adequate competitive analysis without real insight, who produces serviceable content that could have been written by anyone, who executes campaigns without building systems or stories. That tier of work isn't going away overnight, but it's getting compressed, and the compression will continue.

Building the Skills That Matter

This bifurcation is why we built the Future of Product Marketing curriculum the way we did—not as a survey of AI tools, though those matter, and not as a defense of traditional PMM, though the fundamentals remain, but as a map through the transformation. The twelve units trace both paths, developing both the system-building capabilities and the human-premium capabilities, with the understanding that the most effective PMMs in the bifurcated world will be fluent in both languages and will know when to deploy which.

The middle ground is collapsing. The question is which side of the fork you're building toward—and whether you're making that choice consciously.