The One Man Marketing Team: How AI Transformed Growth Marketing at Anthropic

There is a detail in the Austin Lau story that most of the viral threads skip over, and it might be the most important one.

When Anthropic first released Claude Code, their agentic coding tool, Lau had to Google how to open Terminal on his Mac. He is a marketer. He had never written a single line of code in his life. And his first reaction to the product his own company had just shipped was, by his own admission, complete confusion.

"I have zero idea what this product is for," he said internally. "As a marketer, it just didn't really click."

That was early 2025. What happened next has become one of the most talked about stories in tech this year. For ten months, Austin Lau was the entirety of Anthropic's growth marketing team. One person. Paid search, paid social, app store optimization, email marketing, SEO. All of it. For a company that would reach a $380 billion valuation and grow from $1 billion to $7 billion in annualized revenue in under nine months.

The Curiosity That Changed Everything

Lau came to Anthropic with a strong but entirely non-technical resume. He had spent time at Dropbox, then moved to Webflow, then to Notion, where he spent nearly four years leading startup growth and building programs like First Block, Founder Fridays, and The Startup Stack.

He joined Anthropic as their very first growth hire. There was no team. No existing paid search infrastructure. No lifecycle marketing engine. No one doing app store optimization. Nothing. Just a product that was gaining traction and a company that needed someone to figure out how to pour fuel on the fire.

Most growth marketers in that position would have immediately started spinning up campaigns. Lau did the opposite. He started with lifecycle marketing, not acquisition. His reasoning was simple: if you don't understand what makes a user stay, scaling acquisition just accelerates the leak.

A Figma Plugin in 45 Minutes

A colleague posted a guide in Anthropic's internal Slack channel explaining how non-technical employees could install and use Claude Code. Lau, curious but skeptical, followed the instructions. Within a week, he had built two tools that fundamentally changed how he worked.

The first was a Figma plugin. Running performance marketing at scale requires a constant churn of fresh ad creative. Different headlines, different descriptions, different aspect ratios, all needing to be generated across dozens of templates. Before the plugin, Lau was doing this by hand.

He opened Claude Code and described his problem in plain language: "Claude, I'm working in Figma. I really want to be able to solve this challenge of this repetitive copy and pasting. Can you help me build a Figma plugin to help me solve my challenge?"

Claude Code went and read the Figma API documentation on its own, prototyped a solution, and gave Lau something to work with. After roughly 45 minutes, Lau had a working plugin that could take a batch of headlines from a Google Sheet and generate up to 100 ad variations with a single click.

The Google Ads Workflow

The second tool was a Google Ads workflow. Lau created a custom slash command in Claude Code, /rsa, that would accept campaign data, existing copy, and keywords, then cross-reference everything against a set of "Agent Skills" he'd built for Anthropic's brand voice, product accuracy, and Google Ads best practices.

Then he built a memory system. Each round of ad testing generated data on what worked and what didn't. Lau set up a structure where Claude could automatically retrieve all prior experimental results before generating new variants. With every cycle, the system got smarter.

What the Numbers Actually Mean

Anthropic published these results in an internal white paper:

Those numbers are real, and they're impressive. But they need context.

Everything Lau automated shared a set of characteristics. The inputs were structured. The outputs were measurable. The feedback loops were fast. The cost of failure was low. A bad ad variant just underperforms. Nobody gets hurt. This is the ideal surface area for AI augmentation.

The copy and creative examples he fed into Claude were not generated from thin air. They were written in partnership with Anthropic's product marketing and copywriting teams. Human judgment was baked into the foundation before Claude ever started brainstorming. The AI wasn't replacing creative strategy. It was scaling creative execution built on top of human strategy.

The Super Bowl Moment

It's worth noting what was happening around Lau while he was quietly running Anthropic's entire growth marketing operation.

The company was in the middle of explosive growth. Claude was gaining millions of users. Revenue was climbing at a pace that made even Silicon Valley veterans do a double take. The company's first real brand campaign, "Keep Thinking," launched in September 2025, positioning Claude not as a productivity tool but as something more philosophical: an AI built to expand human thinking.

When Anthropic arrived at the Super Bowl in February 2026, the "A Time and a Place" spots imagined a world where AI conversations get hijacked by irrelevant ads. The campaign won the Super Clio for Creative Excellence. The ads reached an estimated 120 million viewers.

Coverage: Adweek | Campaign US | The Drum

What This Actually Tells Us

In September 2025, the team grew. Lau himself described where the role is heading: "Growth marketing is going the way of almost like a product manager. We're not only able to execute on campaigns, we're able to actually build products in order to help us achieve our targets."

That sentence is worth sitting with. The marketer as builder. Not someone who files tickets with engineering to get a tracking pixel installed. Someone who identifies a problem, describes it in natural language, and builds the solution. In 45 minutes. Without writing code.

The uncomfortable implication is not that marketing teams will disappear. It's that the composition of marketing teams will change. The ratio shifts. Instead of one strategist supported by eight producers, you might have two strategists supported by one person managing AI-powered production.

The Part Nobody Talks About

Before Lau built any of his tools, before he automated a single ad variation, before he connected Claude Code to any API, he did something that sounds almost anticlimactically simple.

He sat down and deconstructed his own workflow.

He mapped out every step. Identified which parts were repetitive and which required judgment. Figured out which tasks had structured inputs and measurable outputs, and which ones needed a human to look at something and just know whether it felt right. Then he automated the first group and freed himself up to do more of the second.

That process—the willingness to honestly audit your own work and separate the mechanical from the meaningful—is the actual bottleneck. It's not technical ability. It's not access to tools. It's the clarity to see which parts of your job are truly yours, and which parts have been keeping you too busy to do the work that matters.

As Anthropic's white paper put it: the bottleneck of efficiency often lies not in technical capability, but in whether you're willing to spend time deconstructing your workflow clearly, and then handing off the parts that can be automated.

The Bottom Line

One person. Ten months. A $380 billion company. It started with Googling how to open Terminal.

The tools exist today. The question is whether you're willing to look at your own work that honestly.