It all started with a blog post from Anthropic.
https://claude.com/blog/how-anthropic-uses-claude-marketing
It's a story about a "Growth Marketer who had never even touched a terminal but is now running entire marketing departments with Claude Code."
As I looked into it, I found more similar examples than I expected. Based on public information as of April 2026, I've summarized these case studies of marketers leveraging Claude Code.
Overview of the Case Studies
I have summarized three cases here, featuring marketing professionals from international companies and individual practitioners.
Case Study 1 | Anthropic Themselves — "One Person Running the Growth Marketing Department"
This was the first case I read, and it surprised me the most.
Austin Lau, who works on the Growth Marketing team at Anthropic, went from never having touched a terminal to building and operating all of the following using only Claude Code:
- Automated Ad Variation Generation: Inputting existing ad CSVs into Claude → sub-agents dedicated to headlines and descriptions generate hundreds of variations while strictly following character limits.
- Custom Figma Plugin: Built a system to insert dozens of ad variations into Figma with one click in about 45–60 minutes, despite having zero programming experience.
The blog also details exactly how he built the Figma plugin.
For social ads and app store assets, it's common to have combinations like 10 copy variations × 5 aspect ratios. Previously, this meant manually duplicating Figma frames and copy-pasting back and forth from Google Docs.
This task alone used to take over 30 minutes per update.
All Austin said to Claude Code was this:
"Claude, I’m working in Figma. I want to solve this repetitive copy-and-paste challenge. Can you help me build a Figma plugin to solve this for me?"
Claude reportedly started building a prototype on the spot by researching the Figma API documentation itself.
The first version wasn't perfect, but it was "enough for a proof of concept," and from there, he iteratively improved it to completion.
Now, the workflow involves copying catchphrases from a Google Sheet, selecting the target frames in Figma, and clicking one button in the plugin.
All variations, including different aspect ratios, are generated automatically. With a maximum of 100 items per batch, the system now saves about 30 minutes every time an update is made.
What’s even more amazing is that building this system took only 45 to 60 minutes.
By spending the equivalent of just two manual sessions on building the automation, he significantly reduced all future workload.
📎 Anthropic Official Internal Document PDF "How the Anthropic Team Uses Claude Code"
Case Study 2 | Adam Sandler (Former American Express Marketer) — Building a "1-Person CMO" System
This is a case study of Adam Sandler, a veteran marketer who has handled digital marketing for Fortune 500 companies like American Express and Nestlé.
He currently operates his entire marketing business solo by building an orchestrator agent called the "SLC CMO Agent."
How the SLC CMO Agent Works
This agent is designed with an emphasis on a "guided experience." Centered around campaign development (brand, competitor, SNS, launch, annual campaigns, etc.), it is programmed to ask all the necessary questions.
Adam constantly iterates to reflect his own marketing experience and intuition in the agent's behavior. He takes an approach of "correcting the AI on the spot if the output feels off, thereby adjusting the agent's thought patterns."
Specifically, he is doing the following:
- Automated Competitor Analysis CSV Generation — He instructs, "Research this competitor and save it as a structured CSV." Once a template is made, it can be used for the next competitor analysis, and as data accumulates, it can be iteratively expanded into trend analysis and email campaigns.
- Parallel Content Creation (Using Sub-agents) — If he wants to write 10 blog posts, 10 sub-agents run simultaneously in parallel. 10 posts are completed in the time it takes to write one. These sub-agents are "disposable," each having an independent context window and being discarded once the job is done.
- Strict Brand Voice Application — He accumulates call and video transcripts from clients to build individual "persona skills." He trains the AI to reproduce that specific person's way of speaking, phrases, and style.
- MCP Integration with Google Analytics & Search Console — He built custom MCP tools to retrieve GA/GSC data directly from Claude Code. By simply giving a voice command like "Look at recent performance and suggest 10 blog posts based on trends," he gets suggestions based on real-time data.
- Onboarding Flow — When setting up a new brand, he just drops brand assets into a folder, and the agent organizes them. It even automatically guides brand analysis using marketing frameworks like the Golden Circle or Blue Ocean.
He handles all of this alone.
Another impressive point was his philosophy on context management.
He describes it as "surgically managing what to show and what not to show in a session." He emphasized that if irrelevant context gets mixed in, it causes hallucinations, so intentionally narrowing the scope is crucial.
He also strongly recommends voice input (he loves Super Whisper), noting that "it's not just about typing efficiency; the mindset shift of explaining things out loud improves the quality of the output." This feels like a great match for a marketer's verbalization skills.
📎 Source: Personal Site "The Viable Edge" (Sells and publishes 14 types of AI marketing specialist systems using Claude Code)
Case Study 3 | Zapier — Product Marketing Team Fully Utilizing Claude Cowork and Code for Daily Tasks
Moving from individual cases to team and corporate examples.
At Zapier, Joe Stych (Head of Product Marketing), Matt Brown (Senior Manager of Influencer Marketing), and Larisa Cavallaro (AI Automation Engineer) use Claude Cowork to delegate actual tasks.
I found Joe Stych's story particularly interesting.
He reportedly researches the company's product database, creates messaging materials, and completes a landing page all within the same work session.
Previously, this flow required coordination with three different teams.
Joe has pre-connected three sources of information to Cowork:
① The company's existing homepage
② Team-specific "Skills" (A reusable set of instructions summarizing the product marketing team's writing rules, messaging policies, and operational assumptions)
③ Internal tools via Zapier MCP (Slack threads, the internal search tool Glean, Jira, etc.—anything needed can be pulled in)
By pre-integrating the team's philosophy and judgment criteria into the tool, he avoids having to re-explain everything from scratch every time.
Here is the actual workflow:
He only gives Cowork two things: "The current company homepage" and "The messaging policy for this project."
Claude opens the actual website in a browser, reads the page elements like the header, hero section, and feature introductions, and outputs a revised homepage proposal as a prototype HTML page aligned with the new messaging.
It produces something that can be shown to others immediately without needing a designer for the initial handoff.
Moreover, while Claude is working, Joe can focus on other tasks. As a result, the time required to share the draft with the CMO or CEO for feedback is only about 15 minutes!
Previously, they couldn't even evaluate the direction without first handing it off to a designer, which took several days. The sense of speed is entirely different.
Another striking point was Joe's philosophy of "not treating sessions as disposable."
At the end of each session, he asks Claude, "What should we remember from this session?" to accumulate learnings.
Additionally, influencer marketer Matt Brown self-built an ROI dashboard using Claude Cowork (constructing a real-time visualization dashboard on GitHub Pages that updates daily).
The idea of a marketer "building a dashboard themselves" is something that likely wouldn't have happened before.
Once you can do this much, it seems impossible to go back to working without AI.
📎 Zapier Blog "What is Claude Cowork?" (Feature explanation article)
Insights Across the Case Studies
① "I can't code" is no longer an excuse
Both Austin Lau and Adam Sandler started with no experience in terminals or programming. The approach of "talking to the AI as if explaining a problem to a colleague" directly translates into building workflows.
② The "Structure" is the same for individuals and large corporations
Whether it's an individual case (Adam Sandler) or a team case (Zapier), the essence of what they are doing is the same. They assign roles to sub-agents for parallel processing, while humans focus on review and decision-making. Knowing that the structure is reproducible regardless of scale makes it easier to "start small."
③ How far to allow "Human review only for final confirmation"
I don't have a clear answer yet, but the criteria for how much to delegate to AI will vary greatly depending on the industry, brand, and phase. We need to consider our own boundaries while looking at these examples.
Starting a Practice Community for "AI x Marketing"
For those interested in the practice of "changing marketing work using AI" as introduced in this article:
"aimark" is a community where we share practical knowledge of AI x Marketing and learn by doing together.
- "I'm using AI tools, but I don't know how to apply them to actual work."
- "I don't have anyone around me to talk to about AI x Marketing."
- "I don't just want to see examples; I want to reproduce them myself."
This place is recommended for people like that. Through challenges and discussions, we prioritize not just input, but getting to a level where it can be used in actual business.
We are currently recruiting the first batch of members until April 14th. If you're interested, please take a look.
