I run 24 AI-powered Facebook pages across 5 languages. Each page generates its own content — zodiac readings, dream interpretations, astrology forecasts — and posts automatically, every single day. Today I activated image workflows for my 5 Vietnamese pages, launched ad campaigns for 4 of them, and fixed two broken pipelines along the way. Here's the full story.
The Setup: 5 Pages Stuck at Video-Only
My Vietnamese pages had been running since late March, but only with video content. Each page had two video workflows posting 6 times per day. The image workflows existed but were turned off — standard procedure in my content pipeline. I don't activate image posting until a page has enough video content to establish itself.
The 5 pages cover different niches — zodiac by stars, zodiac by Chinese astrology, universe predictions, past-life readings, and dream interpretation. All in Vietnamese, all targeting the VN market.
Testing: 3 Passed, 2 Failed
I triggered all 5 image workflows simultaneously. Three worked perfectly on the first try — content generated, image created, posted to Facebook, database updated. Clean runs.
But two pages failed at the Facebook posting step with a cryptic error: "Unable to fetch video file from URL." Except these weren't videos — they were images.
The Bug: A Wiring Problem
When these workflows were cloned from a template, the internal variable references got mismatched. The image generation step saved its output as image_path, but the Facebook posting step was looking for Generate Image. Since that variable didn't exist, Facebook received an empty URL and complained it couldn't fetch a "video file."
This is the kind of bug that only surfaces when you run the full pipeline end-to-end. Unit testing individual steps wouldn't catch it — the steps all worked fine on their own. It's the wiring between steps that was broken.
The same issue affected the database update step. It referenced Get Unposted Row instead of the actual output variable row.row_id. Fixed both, re-tested, and got clean passes across all 5 pages.
Scheduling: 15 New Time Slots Across 198 Existing Ones
With my scheduling system, every workflow across all 24 pages must fire at a unique time with at least a 5-minute gap between any two. That's 198 active time slots already occupied.
I used my scheduling algorithm to find optimal image posting times for each page:
- Calculate the gaps between existing video posts for each page
- Place image posts in the middle of the largest gaps
- Check every proposed time against all 198 existing slots
- Shift by a few minutes if there's a collision
- Process pages one at a time, adding each new schedule to the "occupied" set before calculating the next
The result: 15 new time slots, zero collisions, every page now posting 9 times per day (6 video + 3 image) evenly distributed across Vietnamese peak hours.
Ad Campaigns: 4 Pages, 16 Ads, Live in Minutes
With the content pipeline running, I launched Page Like campaigns for 4 of the 5 pages. Each campaign uses the same proven formula from my AI ad system:
- 100 THB/day budget per page
- 4 latest reels as ad creatives
- Targeting: Vietnamese women, 25-65
- Advantage+ audience optimization
All 4 campaigns with 16 total ads went live through a single API call each. The fifth page stays organic for now while we evaluate its content performance.
The Bigger Picture
What happened today is a typical scaling day in my multi-language operation. The pattern is always the same:
- Test the new capability end-to-end
- Find and fix the things that break (there are always things that break)
- Calculate optimal scheduling to fit into the existing system
- Activate and launch ads
The entire session — testing 5 workflows, debugging 2 failures, scheduling 15 new time slots across a 198-slot system, and launching 4 ad campaigns with 16 ads — took one sitting. That's the power of having an AI agent that understands your entire system.
Automation at scale means things will break. The question isn't whether they break — it's how fast you can diagnose, fix, and redeploy. When your AI knows the codebase, the database schema, the API endpoints, and the business logic, "fast" means very fast.
This kind of scaling — going from five pages with video to five pages with video plus images plus ads, all in one sitting — is what happens when your AI has real access to your systems. That's the setup every Jarvis customer gets: a dedicated server where the agent knows your codebase, your database, and your business logic from the start.
— Pond