In this video I walk through exactly how I set up an AI agent pipeline that pushes out 200, 300, even 400 pieces of content per week, almost entirely on autopilot. No expensive team. No Buffer subscription. No manually logging into five different apps after every upload. Just a system that watches for a new file, grabs it, transcribes it, summarizes it, and hands off usable content to every platform where your buyers are spending time.
I have been iterating on this for a while and the version I show here uses N8N, Google Drive, Amazon Web Services S3, AWS Transcribe, OpenAI, and Airtable. You do not need to know how to write code to build it. You need patience in the early stages, because it will frustrate you at first, and then it will run in the background while you do something else entirely.
What You’ll Walk Out With
- A plain-English breakdown of the six tools that power this content machine
- The exact polling and transcription loop that makes everything automatic
- Why AWS Transcribe beats Whisper for longer videos and what that means for your workflow
- The modularization principle that keeps one broken credential from killing your entire system
- The real math: how one video becomes 20 to 40 pieces of content across every major platform
- How to package this as a paid service for small businesses and content creators
- A realistic look at what this system actually costs to run each month
- Want to know which online income stream fits where you are right now? Answer a few questions at finder.platformproof.com and get a personal recommendation.
Why Most Creators Are Still Manual-Posting in 2025
Here is what most people do. They record a video, upload it to TikTok, then open a new tab and upload to YouTube, then Facebook, then Instagram, then maybe LinkedIn if they have energy left. That process eats two to four hours every single time. After a few weeks it becomes the thing that does not get done.
The creators who are everywhere, people like Gary Vaynerchuk or Alex Hormozi, are putting out 200 pieces of content per week. But they are doing it with a full team behind them. That team is expensive. Most people watching this cannot afford that team, and should not try to build it yet. What you can do instead is build an AI agent that does most of that distribution work for free, or close to it.
The core promise of this system is simple: you create once, and the machine handles the rest. You do not have to think about repurposing again once the workflow is live. It runs until something breaks, and I will cover how to build it so that when something does break, it breaks in a small, contained way rather than taking everything down with it.
The Numbers That Make the Case for Being Everywhere
Let me give you some platform numbers that reframe how big this opportunity actually is. Facebook has over 3 billion active monthly users. YouTube is at 2.5 billion. Instagram sits at 2 billion. TikTok at 1.6 billion. Snapchat 800 million. Twitter 600 million. Pinterest 470 million. Threads 200 million. Add those up and you are looking at more than 11 billion monthly users across platforms.
Now here is what that number means practically. If you just wanted to reach 100,000 people in a month across all of those platforms combined, that is 0.009% of total monthly users. To get to a million impressions, that is still only 0.01% of the total. If your goal is to make $10,000 a month, you do not need the whole internet to know who you are. You need an almost invisible fraction of the people already scrolling every single day.
The other thing worth pointing out is that your customers are not all in one place. Some of the people who would buy from you spend most of their time on Snapchat. Others live on LinkedIn. If you are only posting to one platform, you are leaving those people out entirely. Being what I call omnipresent, showing up consistently across multiple platforms, is not about vanity. It is about being where buyers actually are, regardless of which app they happen to prefer.
The Six Tools This System Runs On
Before I walk through the workflow, here is a quick map of the tools involved and why each one is in the stack.
N8N is the automation platform that connects everything. I used Make for a while and it worked fine, but I switched to N8N because it gives you more control. There is a steeper learning curve, and that is worth acknowledging upfront. If you have never used an automation tool before, plan for a few sessions of trial and error before things click. Self-hosting N8N, which I do on a service I refer to as NHN, lets the workflow run 24 hours a day without stopping.
Google Drive serves as the intake folder. When I upload a TikTok video, it automatically gets saved to a specific Google Drive folder. That folder is the trigger point for the entire workflow. No manual file transfers needed.
Amazon Web Services S3 is the storage layer. N8N downloads the video from Google Drive and then uploads it to an S3 bucket before transcription begins. Buckets in S3 work like folders. You will need to create one for your videos and a second one for your transcription output files. To keep storage costs down, I recommend deleting the files from S3 once transcription is complete, since you will have permanent copies in Airtable anyway.
AWS Transcribe handles the speech-to-text step. I am going to cover in detail below why I chose this over the more commonly used Whisper.
OpenAI is used specifically for summarization. Once the full transcript is ready, I send it to OpenAI’s chat model inside N8N using a built-in node. The output is a clean summary that becomes the raw material for every other piece of content I want to create from that video.
Airtable stores everything. Each row in my Airtable base holds the video title, the full transcript, and the summary. From there, I can branch out into blogs, social posts, image captions, LinkedIn articles, whatever format I want to create next. Airtable is the central hub that makes all of that branching possible.
How the Workflow Actually Runs, Step by Step
Here is the sequence from start to finish, the way it runs every time I upload a new video.
The N8N workflow is set to poll my Google Drive folder every 20 minutes. This is called polling. The workflow keeps asking the same question on a loop: is there a new file here? When the answer is yes, it kicks into motion. The interval is configurable. You could set it to check every 10 minutes or every hour depending on how quickly you need the pipeline to run.
When N8N finds a new file, it downloads it from Google Drive. Files are saved using the unique ID that Google Drive assigns to each upload. That ID is what prevents the system from accidentally processing the same file twice. After downloading, the workflow immediately uploads the video file to the S3 bucket that I set up for incoming content.
Next, N8N triggers an AWS Transcribe job on that file. Transcription is not instant. AWS Transcribe processes the audio and returns a transcript file, but it takes time depending on the length of the video. To handle this, I set up a loop inside N8N that checks for the completed transcription every 5,000 milliseconds and tries up to five times. You can adjust both of those numbers. If your videos are long, bumping the retry delay up or increasing the max attempts gives the transcription job enough time to finish before the loop gives up.
Once the transcript is ready, N8N pulls it and hands it to an OpenAI node for summarization. For short videos this works directly. For longer videos, there is a chunking step you need to add. N8N has a built-in data loader and character splitter that breaks the transcript into pieces of a set size before sending it to OpenAI. I have my chunk size set to 1,000 characters. The summarization chain processes each chunk and then combines everything into one coherent summary. Without that chunking step, a long video will break the summarization chain because it sends more text than the model can handle in one shot.
The final step is pushing everything into Airtable. The workflow creates a new row with the video title, the full transcript text, and the AI-generated summary. At that point the pipeline has done its job. The transcript and summary are sitting in Airtable ready to be turned into whatever content formats you need next.
The last thing I want to add to this workflow, and I mention it in the video, is a cleanup step at the end that deletes the original video file and the transcription file from S3 once everything is saved in Airtable. Storage in S3 costs money. If you are running this at any kind of volume, those costs add up. Deleting completed files keeps your S3 bill manageable.
Why AWS Transcribe Instead of Whisper
Most tutorials that cover AI transcription in automation workflows point you toward OpenAI Whisper. And Whisper is good. I used it for a while. The problem is the file size limit. There is a cap on what you can upload into Whisper, somewhere around 25 megabytes. For short-form content that is usually fine, but if you are repurposing YouTube videos or webinars or anything longer than about 10 to 15 minutes, that limit becomes a wall.
When you hit that wall, you have to extract the audio from the video before uploading. That adds another step, another point of failure, and more complexity to maintain. The whole point of this system is to reduce the manual work and the moving parts, not increase them.
AWS Transcribe handles video files up to about four hours long with no issues. My typical videos run around 30 minutes. The four-hour ceiling gives plenty of room and I have never had a file size problem. I am AWS certified and I have worked with AWS for years, so I had some familiarity coming in. But even if you have never touched AWS before, setting up a Transcribe job through an N8N workflow is not especially complicated. The AWS documentation is thorough and most of the heavy lifting happens in the N8N node configuration rather than in the AWS console.
Yes, AWS Transcribe costs more per minute than Whisper. But the cost difference is not dramatic enough to matter unless you are running transcription at very high volume. The tradeoff of skipping the audio-extraction step and having no file size ceiling is worth the slightly higher per-minute cost for most people building this kind of system.
The Modularization Principle That Protects Your System
This is something I feel strongly about, so I want to take a minute to explain it properly. When people start building AI agent workflows, the temptation is to put everything into one giant flow. One workflow that does 8 or 10 or 12 different tasks in sequence. On paper that looks efficient. In practice it is fragile.
Here is the problem. If a credential expires midway through that giant flow, or one API call times out, or one node throws an error, the whole thing stops. Depending on where in the chain the failure happened, you might lose all the work that was done before the break, or you might not know which step caused the problem, or both.
The principle I use instead is modularization. Build each workflow to do one specific thing. One workflow downloads the file from Google Drive and uploads it to S3. A separate workflow handles transcription. A separate workflow handles summarization. A separate workflow handles the Airtable entry. Each one is independent. If the transcription workflow breaks because of an AWS credential issue, the Google Drive download workflow keeps running fine. You fix the broken piece without touching anything else.
Think about it this way. Jan from accounting only handles accounts receivable. She does not also do accounts payable, data entry, and loading boxes on the shipping dock. Jan has one job. She might not do it perfectly every time, but she knows exactly what her responsibility is, and when something goes wrong in shipping it does not affect her work at all. That is how your AI agent workflows should be designed. One job per workflow. Small, contained, replaceable.
The other benefit of modularization is portability. If I decide to start a project in a completely different niche, I can pull out the specific workflows I need, leave the ones I do not, and spin up a new system without rebuilding from scratch. The transcription workflow works regardless of whether the content is about dog training or digital marketing. The pieces are interchangeable because they are not tangled together.
From One Video to 40 Pieces of Content
Here is where the math gets interesting. Once you have a transcript and a summary in Airtable, you have the raw material to create a lot of different things. Let me list out what I can realistically make from a single video using this workflow.
A blog post from the summary. A Substack post to reach a newsletter audience. An Instagram reel using the original video. A native image post for Instagram with a caption pulled from the summary. A LinkedIn video upload with a written post. A LinkedIn image carousel. A Twitter or X thread. A Pinterest pin with an image and description. A Facebook video post. A Facebook text post.
That is already 10 to 12 pieces of content from one video. Add variations, different angles on the same topic, and you can reasonably get to 20 to 40 pieces per video. Now multiply that by three videos a day. That is 60 to 120 pieces of content per day. Over a full week, you are looking at somewhere between 420 and 840 pieces of content, and you are doing very little of the actual production work yourself.
Gary Vaynerchuk and Alex Hormozi are producing at those volumes with full teams. This system gets you to similar output numbers with one person and a modest monthly software spend. That is the real case for building it.
How to Package This as a Paid Service
If you are not a content creator yourself, or you want to build a business around this system rather than using it for your own content, there is a clear service model here. Small businesses and individual content creators need exactly what this workflow produces. They do not have time to repurpose their own content, and they cannot afford a full social media team. You can be the solution to both problems.
The packaging I would suggest is three tiers. The entry-level package covers three platforms. A client picks Instagram, LinkedIn, and Twitter, for example, and you deliver repurposed content to those three places from every video they send you. The mid-tier package expands that to seven platforms. The top tier covers 20 platforms and includes every format each platform supports.
Because you built this using the modularization principle I described above, adding or removing platforms for a specific client is a matter of plugging in or pulling out a workflow module. You are not rebuilding the whole system for each client. You are swapping interchangeable pieces. That is what makes this a scalable service rather than a custom one-off project every time.
I may put together a full blueprint for this service model. If that is something you want, check the description of the video for a link. I will share it there when it is ready.
Not sure which online income path fits where you are right now?
Answer a few questions and get a personal recommendation at finder.platformproof.com.
What This Actually Costs
I want to be straight with you about the cost side because I think a lot of tutorials gloss over this.
AWS Transcribe charges by the minute of audio processed. The exact rate varies by region and whether you use standard or enhanced transcription. For most creators running a few videos per day, the monthly cost is noticeable but not prohibitive. If you are doing high volume, say 30 or 40 long videos per week, you will want to do the math before committing to AWS Transcribe exclusively.
N8N self-hosted is free to run. You pay for the server it runs on, not for the software itself. A basic virtual private server that can handle N8N comfortably costs somewhere between $5 and $20 per month depending on the provider and specs you choose.
OpenAI API costs for summarization are low on a per-video basis. Most summaries are short enough that the cost per call is a fraction of a cent. Unless you are processing hundreds of videos per day, OpenAI is not a meaningful line item.
Google Drive and Airtable both have free tiers that will handle moderate volume. If you are processing a lot of content or storing large files for a long time, you may eventually need a paid plan on one or both, but most people building this for the first time will not hit those limits quickly.
AWS S3 storage costs are genuinely minimal if you follow the cleanup step I mentioned and delete files after transcription is complete. The storage costs only accumulate if you are leaving files in the bucket indefinitely.
Total monthly cost for a typical solo creator running this system: somewhere between $30 and $80 depending on your volume and your VPS choice. That is the realistic number. Compare that to the cost of hiring even one part-time social media assistant and it is not a hard decision.
Find Your X
Building a content distribution system is one piece of the puzzle. The other piece is knowing what you are trying to sell or promote before you scale up your output. If you are not sure which online income model fits your specific skills, situation, and goals, that is where I can help. Take the assessment at finder.platformproof.com and get a clear recommendation based on where you actually are, not a generic answer that could apply to anyone.
Frequently Asked Questions
Do I need to know how to code to set this up?
No. N8N uses a visual node-based interface. You connect pieces together by dragging and dropping rather than writing code. Some steps require you to paste in credentials or configure API settings, but there is no programming involved. The learning curve is real but it is about patience and familiarity with the tools, not about writing code.
Why not just use Make instead of N8N?
Make works and I used it for a while. I switched to N8N because it gives you more flexibility and control over how workflows are structured, and because self-hosting N8N removes the per-operation pricing that Make uses. If you are already comfortable with Make and your workflows are running smoothly, there is no urgent reason to switch. If you are starting from scratch, N8N is where I would begin today.
Can I use this with YouTube videos instead of TikTok uploads?
Yes. The trigger in this workflow is a new file appearing in a Google Drive folder. Whatever you save to that folder, whether it is a TikTok video, a YouTube recording, a podcast audio file, or anything else, the workflow will pick it up and process it. The source of the content does not matter to the system. Only the Google Drive folder trigger matters.
What happens if the AWS Transcribe job takes longer than the retry window?
The workflow will fail at the transcription step. You can prevent this for most videos by adjusting the retry settings. In N8N, increase the retry delay from 5,000 milliseconds to something higher, or increase the number of attempts from five to ten or twenty. For very long videos, a higher delay paired with more retries gives the job enough time to complete. You can also set up an error handler in N8N to notify you by email or Slack when a transcription step fails so you can rerun it manually.
Is there a way to test this without spending money on AWS?
AWS has a free tier for new accounts. AWS Transcribe is included in the free tier for a limited number of minutes per month in the first year. If you are building this for the first time, you can prototype the full workflow within the free tier before you start incurring charges. Just be aware of the free tier limits and set up billing alerts so you know when you are approaching them.
What does the chunk size setting in the summarization step actually do?
When you have a long transcript, sending the entire text to OpenAI in one call can exceed the model’s context limit and cause the summarization to fail. The chunk size setting breaks the transcript into smaller pieces, in my case 1,000 characters per chunk, and sends each piece to OpenAI separately. N8N’s built-in data loader and character splitter handle this automatically once you configure the chunk size. The final summary is assembled from all the individual chunk summaries. The quality is still good even for very long transcripts.
How do I keep S3 storage costs from piling up?
Add a cleanup step at the end of your workflow that deletes both the original video file and the transcription output file from S3 after they have been saved to Airtable. In N8N, this is an AWS S3 delete node. Once the transcript and summary are in Airtable, you do not need the files in S3 anymore. Deleting them keeps your S3 bill minimal regardless of how many videos you process per month.
Can I sell this as a service without being a developer?
Yes. The service model I described in this post does not require you to write code or maintain software in the traditional sense. You are configuring and managing automation workflows, not building applications from scratch. The skills you need are workflow configuration in N8N, basic AWS account management, and client communication. Those are learnable without a programming background. The value you are selling is the result, not the technical knowledge behind it.
Read Next
If you want to see another version of this kind of automation in action, this one focused on social media posting while you sleep, read this next.
This AI Setup Runs My Social Media While I Sleep
Sources
- Facebook monthly active users: 3 billion (as referenced in video)
- YouTube monthly active users: 2.5 billion (as referenced in video)
- Instagram monthly active users: 2 billion (as referenced in video)
- TikTok monthly active users: 1.6 billion (as referenced in video)
- Snapchat monthly active users: 800 million (as referenced in video)
- Twitter/X monthly active users: 600 million (as referenced in video)
- Pinterest monthly active users: 470 million (as referenced in video)
- Threads monthly active users: 200 million (as referenced in video)
- N8N automation platform: n8n.io
- AWS Transcribe documentation: aws.amazon.com/transcribe
Helping 1 million working adults make their first $3,000 online with the skills they already have. Alston Godbolt, Platform Proof.