Vibe coding an ad vault so you stop testing the same angle twice
Workshop #2: a brand runs 400–1,000 Meta ads a month, then can't tell if a new idea is fresh or a repeat. So we built a tool that clusters every ad by funnel.
🧪 A GemLabs lab. This is the write-up of our second internal workshop — a live build session where we take a real problem a brand has and stand up a working tool in front of the room with an AI coding agent. The point isn’t the tool. It’s that you can describe a messy operational problem in plain English and walk out an hour later with something running. Steal the pattern.
Yesterday was the second GemLabs workshop. The problem on the table was one almost every scaling brand hits and almost nobody has a system for: you run so many ads that you lose track of what you’ve already tried.
Here’s the shape of it. A brand doing real volume on Meta ships somewhere between 400 and 1,000 ads a month. Different creators, different agencies, different months, different people on the team. Six months in, a creative strategist sits down with a fresh batch of ideas — and has no honest way to answer a simple question: have we already run this? Is this angle new, or is it the forty-first variation of “boost your energy” we’ve shipped since January?
Without an answer, two expensive things happen. You burn budget re-testing angles that already lost. And you think you’re being creative when you’re actually circling the same three ideas. The brand feels busy. The ad account feels stuck.
So in the workshop I opened Claude Code and we built the thing live. I call it ad-vault. It pulls every ad a brand has ever run from the public Meta Ad Library, labels each one by its persuasion angle, and lays them out so you can see — at a glance — which angles are worn out and which corners of the funnel you’ve never touched.
This post is the build journal, plus the one view that turned out to be the whole point.
Subscribe to read the rest of this post.