Forwarded this email? Subscribe here for more.
👋 Hey there, I'm Nick. Each week, I share field notes from the people building the future of marketing. For more: LinkedIn | The Marketing Engineer | Profound University.
Most marketing AI conversations are still stuck on adoption: how do we get people to use the tools?
Holly Chen is already on the next question.
Holly is VP of Growth Marketing at Samsara, where the marketing org is roughly 230 people. They have about 90 agents and thousands of internal skills. That is the number that made this episode click for me.
Once people actually start building, discovery, ownership, maintenance, token usage, data access, and governance become the real system.
The thing I was most surprised by was the role of marketing ops. At Samsara, marketing ops owns the marketplace, the systems, and the deployment motion while acting as consultants, reviewers, and guardrails for the rest of marketing.
I see a lot of marketing leaders know it's important to adopt AI, but they're overwhelmed. They think they can hire their way out of it. They say, "I'm looking for a marketing engineer who can AI-fy my organization." I think that's the wrong approach. We cannot hire a marketing engineer to AI-fy or transform the organization. Every single marketer needs to be a marketing engineer.
I agree with the spirit. I disagree with the strongest version of the label.
Every marketer should become AI-native. They should be able to build for themselves, because they have the deepest context on their role, workflow, edge cases, and judgment calls. A lifecycle marketer knows where the handoffs break. A growth marketer knows which campaign workflows are a mess. A PMM knows which positioning nuance a generic agent will miss.
That kind of personal building is now table stakes.
But I still think "marketing engineer" should mean something more specific. A true marketing engineer builds and maintains shared systems. They own the plumbing: sources of truth, permissions, deployment, maintenance, and the handoffs around workflows other teams depend on.
So the model I took from this episode is layered. Everyone should be capable of personal agent building. Then team-level or org-level people orchestrate the systems that make that work useful, reliable, and reusable.
There is no one-size-fits-all marketing engineering org design
Holly's model stood out because marketing ops is the enabling function.
She described the rollout as a mix of top-down mandate and bottoms-up building:
One approach is top-down: leadership mandates AI usage, ops people or engineers build agents, and the rest of the organization mostly consumes. Then there is bottoms-up, where everyone experiments, tries things, and builds. Ops and engineers help, consult, review, and sometimes maintain. Long term, the bottoms-up approach is what's most sustainable because the people closest to the workflows and pain points know how to iterate on agent building.
That feels right. Leadership has to make AI adoption real. They give people permission, budget, urgency, and air cover. But the durable workflows have to come from the people closest to the pain.
What I like about Samsara's model is that marketing ops owns the marketplace, guardrails, deployment process, and governance layer. The builders still sit close to the work.
That is also why I do not think there will be one canonical marketing engineering org chart.
At Figma, my read is that PMMs and growth marketers may drive more of the creation and adoption themselves. At Samsara, marketing ops seems to be the systems and governance layer. At a smaller company, this might sit with one technical growth marketer for a while. Different companies will pick different operating models based on size, culture, risk tolerance, and workflow context.
The old marketing org defaults still matter. Comms, growth, PMM, lifecycle, and ops are not going away. But the work is being re-sorted around systems, shared context, and agent ownership.
The bottleneck moves from creation to discovery
This was the most important systems point in the episode:
That's top of mind for me right now: what's the guardrails, what's the orchestration layer of our AI system? For a 230-person team, we have 90 agents now. How do these agents talk to each other? How do they leverage the output from each other? How do we make sure they're pulling the right information, they're efficient use of AI tokens? What's the discovery process look like? Not everyone knows what the other 89 agents are.
That is the next bottleneck.
Holly also mentioned a smaller moment that explains the same problem. She tried to build a skill herself, it did not work that well, and then she found a better one in Samsara's internal skills marketplace.
Once you have thousands of skills and dozens of agents, the question becomes: how does the right person find the right system at the right moment?
A static directory is useful for version one. A large marketing org needs a role-aware discovery layer. If someone works in growth, show them the agents other growth marketers actually use. If someone owns paid social, surface the creative testing, landing page, ad library, and reporting skills tied to that job.
The first wave of AI adoption is getting people to build. The next wave is helping the organization remember what it has built.
What I am stealing
I would steal three things from Samsara's model.
First, the internal marketplace: a real place where agents and skills can be discovered, reused, improved, and governed.
Second, AI Power Hours and demos that focus on process. Holly's point on judgment reminded me of Reema's episode. The scarce human skill is knowing when to do something, when to stop, and how much effort to spend. That judgment spreads when people show what failed, what surprised them, and how they improved the system.
Third, the longer hackathon format. Samsara gives people a month to build something real and then demo it. That is much more useful than a one-hour training. Training creates awareness. A month-long build window creates artifacts, confidence, and internal examples other people can copy.
The more of these conversations I have, the more convinced I am that marketing is in the middle of an org design reset.
The live question: who owns the systems after everyone starts using AI?
Holly's episode is the clearest large-company version of that question so far.
Watch or listen
Listen to Holly's episode of The Marketing Engineer.
If you are building your own version of this agent marketplace, reply and tell me what is working. I read everything.
See you next time,
Nick