Founder Playbook · Sub Club by RevenueCat
6 tactics from Nathan Hudson
How to scale Meta with dozens of AI agents | App Growth Annual 2025
Watch the full episode“The most recent article I wrote... I spoke about tracking cost per winner This is a metric that we track closely that considers production costs as well as actual ad spend costs.”
Track Cost Per Winner, Not Just Cost Per Creative — Production + Ad Spend Together
Most growth teams track CPC or CPM; Nathan Hudson argues the right metric is cost per winning creative — the fully-loaded cost of finding an ad that actually works, combining production expense and testing spend. AI lowers production cost but increases the quantity of losers to test through, so the cost-per-winner metric captures the real efficiency gain (or loss) of moving to AI-generated creatives.
“2025 with Meta's Andromeda has been all about creative diversity It being more important to focus on volume of diverse creatives as opposed to just volume and pumping out a arbitrary number of creatives.”
2025 Shifts From Creative Volume to Creative Diversity — Andromeda Changed the Game
Meta's Andromeda model update changed the optimisation signal. Where 2024 rewarded raw creative volume (more ads = more surface area for the algorithm), 2025 rewards a diverse creative set — different hooks, angles, formats, and concepts. Pumping out fifty variations of the same concept no longer works; the algorithm needs genuine concept diversity to explore.
“The good news is that we are now limited solely by our ability to come up with killer ideas It's all about ideiation and creative strategy.”
The New Bottleneck Is Ideation — AI Removes Production Friction, Not Creative Strategy
As AI tools collapse production time and cost to near-zero, the constraint in paid UA shifts entirely upstream to creative strategy. Any team can generate hundreds of videos; the teams that win are those generating hundreds of genuinely distinct concepts. This repositions creative strategists as the highest-leverage role in growth, not production staff or engineers.
“It's never been so easy to just copy competitors You can literally see what ads they're running on Facebook ad library clone them with AI and push them live But I don't think that's going to work... I think that's going to lead to crazy concept fatigue and trend fatigue.”
Cloning Competitors' Ads With AI Will Cause Concept Fatigue — Original Strategy Wins
AI makes competitor cloning trivially easy — pick an ad from the Facebook Ad Library, replicate it with a video generation model, run it. Nathan argues this becomes a race to the bottom: when every app in a category runs the same cloned concept, the concept saturates immediately and CPMs spike for everyone. Genuine concept origination — finding angles competitors haven't run — is the lasting moat.
“I built a simple N agent which takes an idea or creative hypothesis generates a full creative brief founded on user psychology core value propositions and jobs to be done Then generates a prompt and creates that video with Sora 2.”
Build an N8N Agent That Turns a Hypothesis Into a Sora 2 Video in One Form Submission
Nathan built a fully automated n8n workflow: a creative strategist fills in a form with a hypothesis, and the agent generates a JTBD-rooted brief, turns it into a Sora 2 prompt, creates the video, polls for completion, uploads to Google Drive, and logs the result in Google Sheets. The entire loop runs unattended. Non-engineers on the team can trigger it directly, collapsing production to ideation time only.
“I think that we're about to see another big shift with these new models Neither of those things are going to be a problem Creative volume not an issue Creative diversity not an issue The challenge is going to be that we have so many creatives and not enough ad spend to actually test them all.”
Next Problem: Too Many Creatives, Not Enough Budget to Test Them All
Nathan flags the next wave of growing pains in AI-powered paid UA: budget, not production, becomes the binding constraint. When a single agent can produce hundreds of credible creatives per day, even a well-funded growth budget can't test them all to statistical significance. Teams need smarter creative filtering and prioritisation frameworks upstream — to decide which hypotheses are even worth generating, let alone spending to test.