300+ Viral Products Run Wild Every Night: 0-Person TikTok Factory Opens for Business!!

At NGS, we mainly work on TikTok. Scrolling through FastMoss, the page is filled with other people’s viral products. In theory, if you replicate all these market-verified hits, at least one is bound to blow up.

Version 1.0 used an n8n workflow, which worked. But it was a hassle to configure and required occasional manual maintenance.

Then we upgraded to 2.0. Last month, during our community livestream, we demoed this Feishu template – much lighter than n8n. The entire pipeline, from finding viral products to deconstructing them and producing new ones, was functional.

But there was one unresolved issue: every step required a human. Find the benchmark, download the viral product, fill in a few fields, click “Generate Prompt,” then click “Generate Storyboard,” confirm the storyboard, then click “Generate Video.” Each step still needed someone to click to move forward. In other words, AI was doing the work, but humans were supervising – efficiency was still bottlenecked.

The Bottom Line First

So this is the 3.0 approach. I used Matrix to build this 0-person-supervision TikTok viral product factory running in batch mode. I tested several playbooks: selling functionality, selling emotions, selling aesthetics, etc.

Check out the final videos:

First one is fine, second one laughs a bit too hard, third one is the most refined but has a small bug – did you notice it? Haha. Still usable after some editing.

But it can do more than this. Some people have used Matrix to fully automate Google Ads with AI, and the results were decent. Others are running TikTok shopping ads where the video is AI-generated and self-published, and the ad spend is also automated.

It’s not just about efficiency; the quality of work is also there. So “0-person supervision” marks the efficiency frontier where AI transforms from a tool into an employee, into a 1-person or 0-person company.

01

The so-called “0-person supervision” doesn’t mean hands-off.

Key Insight

Let me pour some cold water first: not everything can be handed over to AI.

The essence of “0-person supervision” is to give the SOP to AI and keep the judgment for yourself. So for AI automation, you need to first figure out which links can be completely outsourced, which nodes must retain human oversight, and then fully hand over the parts that can be handed over.

For example:

• Product selection research – Can be handed over. Let AI scrape data and dig into pain points on TikTok, Reddit. These information-gathering tasks are ten times more efficient when done by AI than by humans. But the final decision on which products to pick cannot be handed over – it’s your comprehensive judgment on category, supply chain, and profit margins. Keep this layer for humans.
• Finding benchmarks, generating prompts, creating storyboard images, generating videos – These steps have clear rules and can run on an SOP, which is exactly what AI excels at. But the final check on whether the generated video is suitable for posting, whether the AI generation feels too artificial, whether the product is distorted – that final look is also kept for humans.
• Publishing, data collection – Scheduling videos to post and monitoring video metrics are also mechanical actions. Best to hand these over too.

Mark This

So, to achieve a high degree of automation, assuming AIGC costs are not a concern, human intervention is needed at only two nodes: product selection, and final video review. Everything else is delegated to the AI CEO to allocate.

Here’s the framework for the entire pipeline:

I mainly used Matrix here, because it integrates Codex and ClaudeCode to collaborate – honestly, it’s hard not to be tempted. Haha.

02

Plan First, Execute Later – Start by Running an MVP

The starting point for Matrix isn’t “building departments” – it’s “writing a single goal.”

After logging in, the first step is to select a workspace. I jumped straight into this Agent company.

Then select the business direction.

After selection, move to the next step which is the company goal. I just wrote it as a task overview.

After setting up the company, my CEO took over.

Before running, I first set up a rough multi-dimensional table, then threw in the PDF methodology from NGS’s livestream about replicating viral products, and gave it to the CEO to learn. Then I told it my idea: build a pipeline from product selection to replicating TikTok viral products, using Feishu only for result presentation.

The CEO didn’t start working right away. It first checked my multi-dimensional table, then organized a plan.

But the plan was a bit heavy – 8 tables, detailed field planning including storyboard IDs, various obscure parameters, multiple review rounds, the works. I asked it to cut it down to a minimum MVP.

So it integrated into 4 tables: Category Research Table → Product Library Table → Viral Source Replication Table → Video Results Table. Need to run the MVP from 0 to 1 first, then iterate later.

I also noticed a detail: if you don’t reply for a while, the CEO will proactively follow up. It feels like a heartbeat mechanism, continuously responding.

03

CEO Supervises for You, Department Collaboration from Product Selection to Final Video

Once the plan is confirmed, you can give the order anytime.

Note!!

Before that, go connect third-party tools: click on Assets, add Reddit and TikTok first. If you can’t find something, just open the browser to operate. Wouldn’t have known otherwise – Matrix’s skill tree is fully loaded! It even integrates over 1000 tools…

I asked it to research three categories: prank toys, pet supplies, fashion accessories. For each category, it research 3 viral products. The CEO delegated the work itself.

In the main chat window, you can see the team lead reporting progress in real time. But if you click into the “TikTok Viral Benchmark Library” department, it’s also outputting simultaneously, showing it opening the browser to work on TikTok. Honestly, its browser operations are the fastest I’ve ever seen – clicking away furiously.

After it finishes, all departments’ outputs are uniformly backfilled to Feishu. And when you click into each department’s chat, you can see the corresponding execution process and output report.

At this point, you need to step in for product selection and make a judgment. Not every benchmark is worth replicating – for example, complex human actions, physical relationships, or scenes that defy AI common sense will likely result in AI-generated failures, just wasting cost.

After I approved it in Feishu, the CEO continued delegating tasks to the “TikTok Viral Content Workshop” department, and set up a scheduled task on its own, because generating images and videos takes some time. This used built-in GPT Image-2 for image generation and Seedance 2.0 for video generation.

At this point, I’m completely hands-off – just waiting for acceptance. No need to supervise the departments. The CEO is watching itself. When the scheduled task triggers, it automatically proceeds to the next step.

Building on this functional demo, further refining the video publishing and data collection, and iterating into a system that can batch produce hundreds of videos – it’s not a dream.

04

The “0-Person Company” in the AI Era: The Key is a Closed Business Loop

What surprised me most wasn’t just achieving automation – it was discovering that Matrix also has an Agent Revenue & Wallet module.

You mean it can actually make money for me?

Before, when we talked about Agents, it was mostly about generating content. But Matrix has closed the loop on “commercialization.”

AI doesn’t lack generative capability; it lacks the ability to organize that capability into a production system. Before, the competition was about Prompt engineering. Now, it’s about whether you can organize AI into a company that can actually make money.

Mark This

AI’s boundaries are actually very clear: it’s great at finding information, executing, and running 24/7. But deciding what’s worth doing and why – that judgment has to be yours. It can never replace your judgment on the essence of business.

Judging what the real pain point is, judging when to scale up or cut losses.

This is the watershed for OPC in the AI era.