One marketer is doing the work of five. Here's the actual stack.
by Luis Gomes, Founder & Growth Lead
The "vibe marketing" framing stopped being a meme this year. What started as a joke about prompt-driven marketing turned into something more specific: operators publishing actual stacks, with actual numbers attached, that show one marketer shipping what used to need five.
We hire one marketer at a time at the brands we operate. The number that keeps showing up across published playbooks — vibemarketer.com, adamigo, digitalfirst, and the AI-first fractional CMO writeups — is one operator delivering between 3× and 5× what one operator did three years ago. 70-80% cost savings on production. 20-30% lifts on ROI. Weeks of work compressed into days.
Those are the headline numbers from people pitching the model. They are real, but the part the headlines skip is what the stack actually looks like. Three layers, each load-bearing. If any one is missing, the rest does not work.
Layer 1 — Research
The bottom of the stack is the layer that feeds context into everything above it. Without it, the marketer is doing the same Google searches they were doing in 2022, then asking AI to write something based on whatever they remembered. That is not a 5x marketer. That is a copilot.
The research layer pulls live context from three places: the company itself (CRM, analytics, product usage, support tickets), the public market (competitors, category news, search queries that are moving), and the customer voice (reviews, social, sales calls, conversations the team is having today).
The shape of the layer is not a SaaS dashboard. It is a set of MCP-style connectors that an agent can call into when it needs context — the customer's last 90 days of behavior, the competitor's last week of pricing-page changes, the top 20 reviews on a competing product, the last five sales calls with people in this segment. The agent assembles the brief. The marketer reads the brief and decides what to write.
Without this layer, the marketer ends up doing the pre-work themselves and then asking AI to format it. That is where the time savings vanish.
Layer 2 — Methodology
The middle of the stack is the codified marketing skill — what the senior person on the team would do, written down in a way an agent can apply.
This is the layer most operators skip. They install tools and assume the tools know how to do marketing. They do not. A general-purpose model writes general-purpose copy. The methodology layer is what turns the model into something that writes the way a senior marketer writes for this specific brand.
Marcel Santilli's GrowthX argument is the cleanest version of this — the gap between AI that works and AI that produces slop is documentation discipline. The brand voice rules. The ICP. The banned words. The KPI definitions. The approval thresholds. Written down in plain language, in a place the agent reads before every action.
Martin Fowler is making the same argument from the engineering side: codify the tacit team-specific standards as instructions. The argument is the same in marketing. Without the methodology layer, the agent is freelancing in your brand's voice. With it, the agent is the second pair of eyes on every junior decision.
Layer 3 — Process
The top of the stack is the order operations run in. This is where most published "AI marketing stacks" get vague. They describe tools, not a sequence.
The sequence we run is closer to: research → foundation → structure → assets. Research is the brief assembled from Layer 1. Foundation is the strategic call the operator makes — what is this campaign trying to do, who is it for, what is the one number we will move. Structure is the outline of every asset the campaign needs, with the methodology rules from Layer 2 applied. Assets are the actual deliverables the agent generates and the operator finalizes.
The operator's time is concentrated in Foundation. That is the one step where AI assists but does not lead. Foundation is the moment of strategic judgment, and the moment that AI gets wrong most often. Junior operators try to skip Foundation. They go from research straight to assets. The output is fluent and pointless. Senior operators spend most of their time in Foundation and let the agent run the rest.
What the math actually looks like
Industry-quoted ranges from the published playbooks: 70-80% cost savings on production, 20-30% lifts on ROI, weeks of work compressed to days.
The way those numbers show up in real engagements is not "the cost savings is 75%." It is more like: the operator ships three campaigns in the time they used to ship one. Two of the three are lower-effort iterations on what worked last time. One is a new structural play that earns the experiments. The compounding is in the iterations — the second campaign is faster than the first, the third is faster than the second.
The lift on ROI is not the agent making better creative. It is the operator running more cycles in the same calendar quarter. More iterations is the leverage. Better tools per iteration is the second-order effect.
Where the model breaks
Three places. Worth naming because the published playbooks tend not to.
The first is when the operator's strategic judgment is weak. The stack does not save a junior operator from junior decisions. Every layer above the methodology assumes someone is making the brand-defining calls. AI without that operator produces fast slop. Five-times-slop is not an upgrade.
The second is when the methodology layer is generic. Templated CLAUDE.md files copied from a thread on X are worth less than nothing — they encode the average instead of the specific. The methodology layer earns its leverage when it captures what is true for this brand and only this brand.
The third is when there is no measurement. The stack works because the operator can run more cycles. Cycles compound only if you can tell which ones worked. A team that ships three campaigns and cannot tell which one moved the number is not running a 5x marketer. It is running a faster guess.
What we install
We build this stack inside client engagements. A research layer connected to the brand's actual systems. A methodology layer codified per client — written down, version-controlled, owned by the operator. A process the team can see and audit. A measurement loop that closes back into the next iteration.
The operator stays the operator. The stack is what makes the operator's time count for five.