Zero to Agent in 30 Minutes: How Max Johnson Built an Automated Content Engine

When Max Johnson, the founder of the AI boutique briix, first tried to tame his daily content grind, he quickly realized that the old "research‑write‑edit" loop was sucking up more time than he’d like to admit. Instead of hiring extra hands, he turned to Claude Code and, over a coffee‑break, cobbled together a tiny agent that does the heavy lifting for him.

It all starts with context. Max fed Claude a modest knowledge base – a few markdown files that spelled out who he is, what his business does, who his audience cares about, and even a handful of sample scripts that capture his quirky, conversational tone. Think of it as giving the AI a "cheat sheet" so it can judge topics and draft copy that feels unmistakably his.

Next came the "vibe coding" bit, a phrase Max loves. He simply described in plain English what he wanted: research a set of topics, rank them, spin a few hooks, and then flesh out the best hook into a full script. Claude Code took that description and generated the underlying code. Max didn’t just sit back, though – he set permissions, answered follow‑up questions, and gave the final thumbs‑up on every step.

The pipeline itself is split into three clear stages. The first prompt asks Claude to pull potential topics from the web. The second evaluates each one on relevance, scoring them so the most promising rises to the top. Finally, the third prompt generates three hook ideas for the winner and expands the strongest into a ready‑to‑publish script, all while drawing from the original knowledge base.

At first, Max triggered each stage manually from his terminal. Handy, but a little clunky. He soon stitched the three prompts together into a single Claude‑generated function that spits out a tidy JSON file containing topics, scores, hooks, and scripts. "That’s the moment it stops being a collection of prompts and starts feeling like a real agent," he says.

To make the whole thing feel less like a back‑end experiment and more like a product, Max built a lightweight web dashboard. The interface lists the generated topics, their scores, the hook options, and the full scripts. He can inspect the output, rerun the pipeline, or tweak parameters without ever dropping back to the command line. His tip? "Make it look like you spent time on the design, not a default template – it makes you actually want to use it as mission control."

Testing revealed two quirks. The first version only output a script for the top‑ranked topic, and starting a new research run still required a terminal command. Max asked Claude to expand the scope: now the agent produces hooks and scripts for the five best topics, and a new button on the dashboard launches a fresh run with a single click. The result? Fifteen script variations per pass – enough to keep his content calendar full for weeks.

If you’re thinking about building something similar, Max’s advice is simple: pick a repetitive task you can explain in a sentence, build a working prototype with the tools you already know, use it, then iterate based on what you learn. The whole process can be as quick as thirty minutes, and the payoff is a reliable, AI‑powered assistant that never sleeps.

Want to see the full walkthrough? Catch the latest "Zero to Agent in 30 Minutes" episode on Radar, YouTube, Spotify, Apple Podcasts, or wherever you get your audio feeds. O’Reilly members can even join a live session and ask questions in real time.