A person who doesn't know how to write code, single handedly withstood all of Anthropic's growth marketing for ten months To what extent can AI improve a person's work efficiency? Recently, a post about Anthropic has sparked a lot of reposts on social media. The poster Ole Lehmann stated that the entire growth marketing team of Anthropic, a company valued at $380 billion, surprisingly had only one person - a non-technical marketer who was solely responsible for paid search, paid social networking, app store optimization, email marketing, and SEO, for nearly ten months. Shortly after the post was posted, it was questioned in the comment section, but soon the parties involved came out to confirm it themselves. The growth marketer named Austin Lau replied that he was indeed the only person doing growth marketing when writing that report, and he lasted almost ten months alone. Image | Related tweet (source: X) Anthropic released an official case study at the end of January this year, detailing Austin Lau's working methods. During the same period, Anthropic also released an internal white paper titled 'How the Anthropic Team Uses Claude Code', covering use cases from ten teams, ranging from data infrastructure to legal departments, with growth marketing being one of them. The white paper states that the growth marketing team focuses on channels such as paid search, paid social networking, mobile app stores, email marketing, and SEO. It is a "non-technical one person team" that relies on Claude Code to automate repetitive marketing tasks and build an automated workflow that traditionally requires a large amount of engineering resources to achieve. (Source: Anthropic) Austin Lau is not an engineer. He said in the official Anthropic case video that he had "never written a single line of code" and when he first started working with Claude Code, he even had to search on Google for "how to open a terminal on a Mac". When Claude Code was first released, his first reaction was that he had no idea who this product was intended for. As a marketer, he felt that the purpose was not obvious. The turning point occurred when a colleague in the company's Slack group shared a Claude Code installation guide for non-technical employees. Austin installed it out of curiosity, and a week later, he built two automated processes that completely changed his way of working. The first set is a Figma plugin. Doing paid social advertising and app store marketing requires processing a large amount of visual materials in Figma. The past process was to manually copy the framework in Figma, switch between Google Docs and Figma, and copy and paste titles one by one when creating multiple variations of the same design proposal. If there are 10 variations of the copy that need to be adapted to 5 different aspect ratios, this mechanical labor can easily consume half an hour. Austin Lau (source: Anthropic) described this pain point in natural language to Claude Code and asked it to help write a Figma plugin. During the process, he asked Claude Code to refer to Figma's API documentation while researching and prototyping. The first version of the prototype generated was not perfect, but it was sufficient as a starting point. He continuously debugged it on this basis and eventually made a functional plugin. The working method of the Anthropic plugin is to select a static image frame, and the plugin automatically recognizes its components such as titles, call to action buttons, code blocks, etc. Then, it generates independent Figma frames in batches from a prepared copy list, with each variant corresponding to a new set of copy. Generate up to 100 ad variants in a single batch, with each batch taking approximately half a second. The manual operation of the past 30 minutes has now been shortened to 30 seconds. The second set is the advertising copy generation workflow for Google Ads. Google Ads' responsive search ads have strict character limits for titles and descriptions, with a maximum of 30 characters for titles and 90 characters for descriptions. In the past, he needed to write a draft in Google Sheets, manually check the character count, and then paste the content into the Google Ads backend one by one. Austin created a custom slash command "/rsa" in Claude Code, which, when triggered, will require input of advertising data, existing ad copy, and keywords, and then cross reference his pre-set "Agent Skills", including Anthropic's brand tone, product accuracy specifications, and Google Ads RSA best practices. The system uses two sub agents with clear division of labor, one dedicated to writing titles and the other dedicated to writing descriptions, each working within their own character constraints. The output quality is much higher than stuffing two tasks into a single prompt word. Finally, Claude Code packaged 15 titles and 4 descriptions into a CSV file that can be directly uploaded to Google Ads. Austin emphasized that the generated copy is only the starting point, and he will evaluate each item one by one: whether the value proposition is in place? Is the tone correct? Is there any differentiation between competitors? But at least the tedious process of generating and formatting initial drafts has been completely automated. The efficiency improvement of these two workflows is already quite astonishing, but Austin's system goes beyond that. He also built an MCP server (Model Context Protocol) that connects to the Meta Ads API. Through this integration, he can directly query advertising performance, spending data, and the effectiveness of each advertisement in Claude's desktop application, without the need to open the Meta Ads backend dashboard. Which ads have the highest conversion rates this week? "" Where did I waste my budget? "These questions can be directly asked to Claude for real-time data answers. More importantly, the closed loop. Austin built a memory system to record the hypotheses and experimental results in each round of advertising iterations. When he starts a new round of variant generation, Claude will automatically retrieve all the data from previous tests, which ones perform well and which ones do not, and build the next round of generation on the basis of historical experiments. This system becomes smarter after each cycle. This systematic experimental tracking across hundreds of advertisements typically requires a dedicated data analyst in traditional teams. According to Anthropic's white paper, the result of this work approach is that the creation of advertising copy is compressed from 2 hours to 15 minutes, and the creative output is 10 times higher than before. The advertising variants he tested alone cover more channels and quantities than most full scale marketing teams. In that white paper, growth marketing is just one of ten cases. The data infrastructure team used Claude Code to debug Kubernetes cluster failures, resolving issues that originally required contacting network experts within a few minutes; Members of the reasoning team without a background in machine learning use it to understand model functions and settings, reducing the time spent reviewing documents from one hour to 10 to 20 minutes; The product design team directly used Claude Code to modify the front-end code, and the engineer found that the designer was making "large state management changes that you usually don't see designers making"; Someone in the legal team created a predictive text assistance application for family members with language barriers in just one hour, and they had no programming experience before. The usage of technical and non-technical positions is different, but the conclusion is consistent: Claude Code is blurring the boundary between "can do" and "can't do", which used to be almost entirely determined by technical abilities. Austin Lau himself has a summary in the case that goes: "The distance between 'I hope this thing exists' and' I can make it myself 'is much shorter than most people think. Of course, it should be noted that growth marketing does not equal the entire GTM (go to market). Anthropic has a complete brand, product marketing, and communication team, while Austin Lau is responsible for the performance marketing line, which includes quantifiable channels such as paid advertising, app store optimization, and SEO. In February of this year, Anthropic placed a TV advertisement for the Super Bowl, which was clearly not something that one person could handle alone. The copy and brand assets that his workflow relies on were initially produced through collaboration between product marketing and copy teams, with Claude conducting variant generation and scale testing based on this foundation. Austin Lau recently added some background information on LinkedIn. He pointed out that the widely circulated article described his experience as the only growth marketer in the second quarter of 2025, which has been almost 8 months since then. The team did indeed expand its manpower later, although the scale was still much smaller than the outside world imagined. In his words, 'our combat power far exceeded our numbers'. Nevertheless, the signal is strong enough. A company with a post investment valuation of $380 billion and an annualized revenue of $14 billion managed its core growth channels for ten months on its own with an inexperienced marketer during the fastest growth phase, and the results were quite good. This should already prove that AI's ability to amplify knowledge workers may be much greater than what our current organizational structure and recruitment inertia assume. It is currently unclear to what extent this model can be replicated. Growth marketing is highly data-driven, process oriented, API friendly, and naturally suitable for automation. The situation may also be very different in fields that require more interpersonal judgment or creative intuition. Anthropic's white paper provides three suggestions at the end of the growth marketing chapter: find repetitive workflows with API interfaces for automation; Break down complex processes into multiple specialized sub agents, rather than attempting to cover everything with a single prompt word; Before starting to write code, thoroughly consider the overall process design on Claude. These three suggestions essentially illustrate that the bottleneck of efficiency often lies not in technical ability, but in whether you are willing to take the time to break down your workflow and hand over the parts that can be taken over by machines.