I build automation. A lot of it. Over the last year I have wired content and publishing systems that post to Instagram, Facebook, LinkedIn and TikTok on behalf of several brands. The AI part, generating the posts, rendering the images, scheduling them, took days. The part that took weeks, and taught me the most, was getting the platforms to let me publish at all.
If you are planning anything that touches a social platform's API, here is the honest version nobody puts in the sales deck.
The approvals are on the critical path, and you do not control the clock
To publish to Instagram or Facebook programmatically you need a Meta app, business verification, and advanced access to specific permissions like instagram_content_publish and pages_manage_posts. Each of those is a separate review. Business verification alone can sit in a queue. Advanced access is another review on top. None of it moves on your schedule.
LinkedIn was the hardest. Posting to a personal profile is straightforward. Posting to a company page needs the Community Management API, which is gated behind a review that can take weeks and is quietly rejected more often than approved. I watched a perfectly valid request come back as a generic error more than once. TikTok's content posting API is the same story: another gate, another wait.
What I actually do now
Start the approvals first. Before I build the pipeline, before I write a line of generation code, I open the app reviews. They are the longest lead time item in the whole project, so they go first, not last.
Piggyback where it is honest to. Instead of getting my own app approved for every platform, I self host a publishing tool that already carries approved platform apps, plug my channels into it, and skip the queue for the parts I can. When you can stand on an approved integration instead of waiting on your own, do it.
Ship the easy channels first. Personal profiles and the lighter review platforms go live immediately, so the client sees results while the company page clears review in the background. Momentum matters more than completeness on day one.
Make the approval the last gate, not the first blocker. The entire system, generate, render, approve, schedule, is finished and tested against a channel I already have. When the approval lands, I flip one switch. Nothing else waits on it.
The real lesson is bigger than social APIs
This is the pattern in almost every AI transformation I have shipped. The intelligence is rarely the hard part now. The hard part is access: the API approvals, the permissions, the legacy ERP that has to accept the order, the accounting system that has to sync, the client server that has to run it. I have put an AI order desk into a decades old ERP, and the AI was the easy afternoon. The integration and the access were the two weeks.
So when I scope an AI project, the first thing I map is not the model. It is every place we need permission from someone else: a platform, a vendor, an IT team, a process. Those decide your timeline. Plan for weeks there, start them on day one, and build everything else so it is ready and waiting.
The AI will be fine. It is the approvals that will surprise you.