Anthropic Marketer Runs Full Google Ads Operation Solo Using Claude AI Plugin

The mechanics of running a full Google Ads operation without a team came into sharp focus when Ole Lehmann, a growth marketing lead at Anthropic, published a detailed breakdown of how one non-technical colleague had been quietly sustaining the company's entire paid marketing output alone. Austin Lau, a growth marketer at Anthropic, single-handedly managed paid search, paid social, app store optimization, email marketing, and SEO for nearly ten months at a company valued at $380 billion. Lehmann's thread drew 1.2K likes and 63 reposts from PPC professionals trying to understand exactly how the workflow holds together under pressure.

The operating system Lau built centers on Claude Code, Anthropic's terminal-based agentic tool, and it starts with a CSV. He exports all existing ads from his ad platforms along with their performance metrics, feeds the whole file into Claude Code, and instructs it to find what's underperforming. Claude analyzes the data, flags the weak ads, and generates new copy variations on the spot. The step that previously consumed the most clock time, writing and validating Google Ads responsive search ad copy, now runs through a custom slash command Lau built himself. Typing /rsa into Claude Code triggers a prompt for campaign data, existing copy, and keywords, which Claude then cross-references against Agent Skills he created encoding Anthropic's brand tone, product accuracy requirements, and RSA best practices. The output is an upload-ready CSV, not a draft requiring manual character-counting inside the Google Ads interface.

The workflow cut ad-variation production from around 30 minutes to roughly 30 seconds. That 98 percent reduction compounds across every campaign cycle: with mechanical tasks absorbed by Claude Code, Lau runs more copy experiments per sprint and iterates faster on what actually performs. He had never written a line of code before building these workflows and had to Google how to open a terminal when Claude Code first launched.

What makes the setup replicable for agency operators is the guardrail layer. A CLAUDE.md file placed in the project folder gives Claude Code persistent memory of the brand's voice, SEO keywords, and persona documents without requiring copy-paste at the start of every session. That file is the change-control anchor: update the brand guidelines once and every subsequent /rsa run reflects them. The Agent Skills layer adds a second enforcement pass, rejecting copy that violates character limits or drifts from approved product messaging before any CSV is generated.

Human judgment still governs the loop, however. "Claude is a great brainstorming partner, but sometimes it doesn't get it right on the first try," Lau explained. "A lot of the work I do is riffing and going back and forth to help refine the copy over time." That refinement cycle, evaluating whether a headline resonates against live audience knowledge no model can access, is where the one-person model incurs its real cost. Query mining, negative detection, budget reallocation signals, and first-draft RSA testing all collapse neatly into the automated layer. Strategic read on competitive positioning, creative intuition about messaging fatigue, and final sign-off before a bulk upload hits a live account do not.

Lau has described the shift as growth marketing moving toward product management: "We're not just executing campaigns; we're building the products that achieve our targets." For solo operators and lean PPC teams evaluating whether agentic tooling can absorb department-level workload, Anthropic's own marketing stack now stands as the most concrete field test available.