AI for marketing leaders: Seven lessons for CMOs from Agentic Marketing Forum West

AI in marketing isn’t an ops project anymore. It’s a leadership job. That was the clearest message from the Agentic Marketing Forum, and it came from the people who should know: the marketing leaders already doing it.

One of them opened with an admission about AI. “I was a bit of a laggard, call it 12 months, 18 months ago,” Zuora CMO Kyle Christensen told a room of enterprise marketing leaders. He’d read the articles. He understood AI “philosophically.” He just wasn’t using it, and neither was most of his team.

That changed. And he wasn’t alone. Getting everyone into one room on Snowflake’s Menlo Park campus on October 29 helped us realize how many leaders had a version of the same story. Leaders from OpenAI, NVIDIA, CrowdStrike, Meta, Snowflake, Zuora, Adobe and Devicie compared notes on what’s working, what isn’t and where they’re placing their bets.

Nick Donaldson, our resident MOps specialist, already covered how their marketing ops teams are building it: the intake, the guardrails, the quality checks. If you run MOps, start there.

This one’s for CMOs and VPs of Marketing, who do two jobs at once. You run the team that has to change how it works. And you answer to the board and the rest of the leadership team for what that change delivers. For most of the CMOs I talk to, that means an AI mandate on a flat budget and a board asking what it bought. The leaders getting results don’t treat AI as a tool rollout. They treat it as their own job. Here are seven lessons from that day.

1. Lead by example, in public

Most AI rollouts start with a memo. Licenses go out. Someone books a lunch-and-learn. Leadership waits for adoption. It doesn’t come, because a mandate tells people what to use, not that it’s safe to be bad at it.

Kyle’s team changed when he stopped waiting and built something himself. “It wasn’t until I’m like, all right, I got to roll up my sleeves as the lead of the organization and … actually build stuff and put it out there that the rest of the org … kicked into gear.”

The second half is the part that matters. He didn’t build quietly and unveil a finished product. He posted the rough versions in the team’s Slack channel as he went. A CMO sharing unfinished work sends two signals: experimenting is safe, and the bar is trying, not perfect.

Give it one direction

Then you hit the opposite problem. Dozens of people vibe coding their own tools is activity, not change. Good luck explaining it to the board.

Hillary Carpio, VP Growth Marketing at Snowflake, isn’t against the experimenting: “Vibe coding is not going anywhere and it’s not a bad thing.” Her concern is direction. “To drive change we need to have a common goal together.”

Setting that goal is the leader’s job. It’s what turns a pile of experiments into a win you can name in a board deck. Nick covers how Snowflake turned its goal into a short list of pillars in his recap.

Let peers do the teaching

Top-down training only gets you so far. Snowflake’s adoption moved fastest through peer cohorts. People “learn better from their peers than they do top down,” and peers know the real problems. Go first. Set the direction. Then let the team carry it.

2. AI can’t live in marketing alone

Keep AI inside marketing and it becomes a marketing tool. Faster emails. More content. A nicer dashboard. Useful, and easy for the rest of the exec team to ignore. It also leaves marketing exactly where it was when budgets get set.

Devicie CMO Corey McCarthy took the other route. AI there runs through the whole leadership team, because the inputs are shared. In her words, it “can’t live in one silo.” Marketing can’t run on one set of assumptions while the CRO runs on another.

Show up with the same context

Alignment changes where marketing sits. Hillary described how shared context puts marketing on “an even playing field” with sales and product. Same customer data. Same read on the market. No more taking questions away and coming back next week.

Build what other teams use

At Zuora, Kyle said marketing now consumes more AI tokens than any function outside engineering. Some of that goes into tools for the field, like tailored outbound the AEs can switch on for their accounts. The result surprised him: “suddenly they’re coming to us for insights in ways they [hadn’t] in the past.”

That’s the seat at the table most CMOs want. You don’t get it by asking. You get it by building something the CRO’s team relies on.

3. Spend the time you get back on the market

For years, marketing rewarded the people who knew the tools. Building programs in Marketo. Fixing broken merge tags. Keeping UTMs straight so the reporting works. That was a career, and in a lot of teams it still eats most of the week. Senior marketers end up in production work instead of strategy, positioning, and the insightful conversations that move the number.

Jon Miller, who co-founded Marketo, put the shift in one line: “If AI can handle the ‘ing’ in marketing, then we all can focus more on the market.” Agents take more of the doing. People keep the part that drives pipeline: the audience, the offer, the message and the call on what ships.

The teams furthest along are redesigning the work around that. Max Martinez said Meta wants to “let the marketers focus on what they do best,” and set the pace by the business, not by how much the build team can handle.

Judgment is the scarce skill now

Here’s the catch. When agents can draft without limit, volume stops being an advantage. Your competitors can produce more too. What they can’t copy is your judgment about what’s worth sending.

Accountability doesn’t move to the agent, either. Speakers kept coming back to it: people own what their agents produce. That’s exactly why leaders need the time to review it.

So protect the time AI hands back. Fill it with more output and you’ve automated your way to a louder version of the same marketing. The board will still be asking where the savings went. Spend it on the decisions only your team can make.

4. Fund the mix, not more creative

Personalization pays. Everyone in the room knew it. What held it back was never ideas. It was production.

Mark Organ, the founder of Eloqua, calls the campaigns teams actually want “demand choreography”: multi-touch, multichannel, tested, and adapting as buyers respond. Most companies went the other way. The most advanced programs took big teams and outside agencies to build, so companies have been “scaling back on the complexity of their marketing campaigns when really we should be making them ultra complex.”

Don’t write a new song for everyone

The obvious AI answer is to let the machine make a unique asset for every person. It doesn’t hold up. Corbin Thoe said that at Meta’s scale, token cost and compute alone rule it out, at least for now. Meta personalizes “the experience and the channel and the workflow, less so on the content,” and drops the right data and proof points into approved templates.

Jon Miller has a better model, and it’s one every marketer already uses: a playlist. Spotify doesn’t write a new song for each listener. It has a catalog of millions of tracks, and when two people each ask for 80 songs, they get two different playlists. Same catalog. Different mix.

Map that onto marketing and the roles get clear. The songs are your approved building blocks: modules, offers, proof points and messages, each signed off once. The listener is the person, the account and the buying group. AI is the curator, picking what goes into each journey based on what you know about them. Your team owns the catalog. The machine handles the mix. The building blocks stay governed, and the mix does the personalizing. Snowflake is already running a version of this, and Nick has the details in his recap.

Spend where the response comes from

Jon started his career in direct marketing and catalogs, and he brought an old rule of thumb from that world: 40% of response comes from the offer, 40% from the list, 20% from the creative.

Most AI personalization is still chasing the 20%. Think AI-written openers about a prospect’s college team. The bigger opportunity is the right offer reaching the right people at the right time. Mark’s version for enterprise buyers: messaging built around the initiatives their CEO named on the last earnings call. Not their first name.

For a CMO, that’s a budget call as much as a creative one. Fund the system that assembles the mix from approved modules you already own. Measure it in pipeline, not assets produced.

5. Your go-to-market model is the new moat

Ask who knows which segments respond to which offers, what the brand will and won’t say, and which campaigns worked last year and why. You’ll get a handful of names. That knowledge lives in their heads, a few decks and a lot of Slack threads. It walks out the door when they do.

Mark Organ called that working model the real asset, and he meant it in board terms. As product advantages get harder to hold, a company’s ability to go to market “is going to drive most of the market capitalization.”

It came up all day under different names: harness, context, marketing OS. Call it what you like. Mark’s point stands: it’s “definitely not a commodity.”

Agents only know what you give them

So, build it on purpose. When James Finlay joined NVIDIA, the first thing he worked on wasn’t an agent. It was the context underneath them. Data was spread everywhere and inconsistent, and he had a choice: put everyone on his team through AI 101, 201 and 301 training, or make every agent smarter so anyone asking a question gets a better answer. He went after the agents.

He called pulling it into one place “a huge standardization effort,” and he didn’t pretend it was easy. The payoff is a team-level outcome instead of a handful of power users. His advice: build with AI “as if AI was part of your organization.”

That’s the leadership part. In James’s words, “your job is to manage what information you are making those decisions off of.” Agents do the work. You decide what they know. Every gap in what you’ve written down shows up in what your agents produce.

The payoff compounds

Mark calls the system that holds the model the harness: segments, journeys, messages, offers, guardrails, and a memory of every campaign you’ve run. Because it remembers, it gets better, and it earns agents more room to work. He expects 2027 to be the year companies get serious about it.

For how ops teams govern the agents inside it (risk tiers, checks, human sign-off), see Nick’s recap.

Where Knak fits

This is the part we’ve been building for at Knak. The piece of the model your customers actually see is the hardest to capture and the easiest to get wrong: brand rules, approved modules, messaging, templates and who signs off. In most enterprises it’s scattered across agencies, regional teams and a design system few people can find.

Your MAP is great at sending. It was never built to hold any of that. Knak is. Brand rules, modules and approvals live in one place, so a regional marketer building by hand and an agent building on their behalf start from the same approved pieces. The brand holds across every region and channel. Nobody has to play brand police. And the Marketo or Eloqua instance you already pay for finally gets campaigns worth sending.

Kai, our campaign agent, works the same way. It drafts from your brand guidelines, messaging, personas and playbooks, and it stops for a human before anything goes out. Nick walks through it live in his recap.

Whoever owns this model owns the advantage. Start writing yours down now.

6. Hire for willingness

Ask who on your team adopted AI fastest and you’d probably guess the most technical people. Kyle did. He was wrong.

She dove in and became one of Zuora’s top AI users. Kyle’s conclusion: “So it’s about willingness.” Title and tenure didn’t predict who pulled ahead. Jumping in did.

Generalists are back

For years, marketing teams grew by adding specialists, each owning one step in a long chain of handoffs. One person configures the journey. Another builds the audience. Someone else checks the claims with legal. Every handoff adds a queue, and the campaign moves at the speed of the slowest one.

AI makes it cheaper to learn the steps outside your lane, and that changes who’s most valuable. Corbin Thoe said Meta is now looking for generalists, and he was explicit about why.

Someone who sees the whole workflow can design how agents and people work across it, then scale that to the rest of the org. Mark Organ made the same call earlier in the day: “I think that generalists are going to be rewarded.”

For a CMO, that’s a hiring brief and a development plan. Look for the people who already ask what happens before and after their step. Then give them room to learn the rest.

Junior talent is an advantage

That favors people early in their careers. Hillary said Snowflake is “hiring builders instead of doers,” and the people who fit that best are often the newest on the team.

Part of the advantage is what they don’t carry: no approach baked in from years of doing it one way, which Hillary said “is not holding them back in any way.” Pair that with senior colleagues who bring judgment, relationships and the scar tissue of past campaigns, and you get reverse mentoring. The juniors show how to build. The seniors show what’s worth building. She called it “a really nice [complementary] relationship where we can all build together.”

AI also removes the barrier that used to hold junior marketers back. Corbin pointed out that they rarely had the relationships with data science or engineering needed to get anything built. Now they “can actually go build out a campaign or an objective themselves.”

For leaders, that changes the shape of the team you hire for. Fewer layers of specialists waiting on each other. More curious people who can build, backed by senior people who can judge.

One hire, many hands

For most CMOs, headcount is frozen and scope isn’t. Asked to choose between a new person, an agent or a tool, Hillary (and the whole panel) still picked the person.

Mark Organ expects the org chart itself to move: “There’s going to be new functions that are created. There’s going to be old functions that die.” Hire the people willing to build the new ones.

7. Demand more from partners

You’ll sit through a lot of AI pitches this year. Most will promise to make today’s workflow a little faster. The leaders on stage had a higher bar, and Hillary’s red flag is two words.

Corbin added a second test. A partner who shows up with no idea of the scale you operate at or the regulations you answer to is, in his words, “a waste. It sounds too good to be true.”

So hold every pitch to two questions. Does this change how my team works, or just how fast it does the same thing? And does it understand the scale and rules we live with? If you’re weighing building it yourself instead, Nick’s recap has the clearest answer I heard all day: own what makes you different, and buy the rest.

What this means for your next quarter

The leaders furthest along didn’t have bigger budgets. They had the same habits. If you need one win to show your board this quarter, start with these seven questions.

1. Have you built something yourself? Then share it, rough edges included.
2. Does every experiment point at one goal? If not, set one.
3. Are your CRO and CPO working from your assumptions? Bring them in early.
4. Where does the time AI saves actually go? Spend it on judgment, not more output.
5. Is your go-to-market model written down anywhere? It’s your moat. Treat it like one.
6. Who are you hiring for? Willingness counts for more than tenure now.
7. Would your partners pass Hillary’s test? If they only speed up today’s workflow, keep looking.

None of these is an ops checklist. Every one is a leadership call: what you build in public, what you point the team at, who you bring into the room, what you protect time for, what you write down, who you hire and who you buy from.

That’s what stuck with me from Menlo Park. Kyle Christensen went from self-described laggard to the person his team followed, inside 18 months, and he did it by going first. Your team is waiting to see whether you will.

AI in marketing used to be an ops project. On October 29, it sounded like a leadership job.