One person. Five channels. No engineering support. Ten months.
That's the structure Anthropic's Growth Marketing team ran on — paid search, paid social, app stores, email, and SEO — managed by a single non-technical hire using Claude Code for growth marketing workflows that would otherwise require a full team.
The results aren't abstract. Ad creation: from two hours to fifteen minutes. Creative output: 10x more variations tested across channels. Tasks that traditionally needed dedicated engineering resources: handled without them.
Here's how the system is built.
Automated ad creative generation with Claude Code
The process starts with a CSV export of existing ads and their performance data. Claude Code analyzes the file, identifies underperforming ads, and generates new copy variations — all within Google's strict formatting constraints.
The design choice that matters: two specialized sub-agents rather than one. A headline agent (30-character limit) and a description agent (90-character limit) each handle their specific task. Splitting the work this way keeps each agent focused, which improves output quality and makes debugging straightforward. Hundreds of new ad variations generated in minutes.
Figma plugin for creative production at scale
Text is only part of the ad. The visual creative — banners, social images, display formats — still needs to be produced.
The team built a Figma plugin that identifies existing ad templates and programmatically swaps new headlines and descriptions across frames. Up to 100 variations per batch, generated in half a second. The hours previously spent duplicating frames and copy-pasting text manually now run automatically. Creative output scaled by 10x without adding a designer.
Meta Ads analytics, surfaced directly in Claude
Campaign data used to mean dashboard-switching: open Meta Ads Manager, pull reports, interpret numbers, return to work. The team eliminated that friction by building an MCP server integrated with the Meta Ads API.
Now campaign performance, spend data, and ad effectiveness are queryable directly inside Claude Desktop. Ask which ads converted best this week. Ask where spend is being wasted. Get real answers from live data without leaving the working environment.
The memory system that closes the loop
Each of the above systems generates results. The memory layer captures them.
The team implemented a logging system that records hypotheses and experiment results across ad iterations. When generating the next batch of variations, Claude pulls previous test results into context — what worked, what didn't, what to try next. The system refines its own output over time.
Systematic experimentation across hundreds of ads, tracked automatically. That's the work of an analytics function built into the workflow itself.
The design principle behind Claude Code for growth marketing
Every piece of this system follows the same logic: find a repetitive task, locate the API, build a structured workflow around it.
The tools — Claude Code, Figma's API, Meta's API — aren't the story. The approach is. Break complex workflows into specialized agents. Brainstorm the full process before writing a line of code. Work step by step rather than asking for one-shot solutions.
A non-technical marketer built an infrastructure that runs like a larger team. The constraint wasn't technical skill. It was identifying where to start.
Learn the workflow yourself
CC4.Marketing is a free interactive course that teaches non-technical marketers how to use Claude Code for exactly these kinds of workflows — campaign briefs, ad creative generation, analytics, and memory-driven iteration. The course runs inside Claude Code itself. No coding experience required.
Start the course at cc4.marketing.