A production line for content, not one big prompt
A YouTuber split his content work into seven agent stages and cut two or three days down to about five hours.
28 Jul 2026 · 4 min read
By Louis-François Bouchard (What's AI) on Blog
The short version
- Seven stages, each with its own agent: research, outline, writing, review, diagrams, titles and translation.
- A typical release went from two or three days of work to about five hours.
- The creator keeps the topic, his own experience, the sources and the final edit for himself.
- A separate reviewer with fresh eyes checks the writer's work, for a fixed number of passes.
Who it's for: Creators, marketers and anyone who publishes every week.
Why it's genius
Most people ask one AI to write the whole thing and get something bland. Louis-François Bouchard, who runs the What's AI channel, does the opposite. He breaks his work into a production line where each stage has one job.
A research agent called Nova runs three set rounds of research and writes up its sources. A writing workflow called Brown drafts from his outline, his voice notes and that research. A separate reviewer, with no memory of the writing, flags problems, and an editor fixes them. Then come diagrams, nine rounds of title ideas narrowed to three for testing, and a French translation.
The clever bit is what he keeps for himself: picking the topic, a brain dump of his own experience, approving the evidence and the final read. As he puts it, an empty starting point just “produces polished emptiness”.
Try it
Write down every step your content goes through, from idea to published.
Mark the steps only you can do, like your opinion and the final check.
Give one other step its own AI chat with saved instructions.
Have a second, separate chat review the result against a short checklist.
You are a strict editor. You did not write this draft. Check it against these rules and list every problem, worst first. Do not rewrite it.
Our take
One giant prompt is why AI writing sounds the same
When you ask one AI to research, plan, write and polish in one go, it does all four a bit badly. Splitting the work up means each step gets clear instructions and a clear check. It's the same reason kitchens have stations.
The fresh-eyes reviewer is the part we'd copy first. An AI checking its own work tends to like what it wrote. A second chat that only sees the draft and the rules is far harsher, which is what you want.
The human parts are the point, not a weakness
Look at what stays human: the topic, the opinion, the proof and the sign-off. That's not him failing to automate. It's him protecting the bits that make the content worth reading. He also admits the model makes opening lines “too safe”, and that only real audience tests can pick between close titles.
If you take one thing from this, make it that. Hand the AI the legwork, never the point of view.
You don't need his code to use his idea
His version is properly engineered, with custom agents and code he has to maintain. Most people don't need that. You can get 80% of the benefit with separate chats or projects for each stage, each with its own saved instructions.
Be honest about the cost too. He says his time saving is his own and not a benchmark for anyone else. A production line takes time to set up and look after, so it pays off if you publish often, not once a month.
Where it shines
Anyone publishing every week: newsletters, YouTube, LinkedIn. It is the clearest example we've seen of “AI does the legwork, you do the thinking”.
Where it breaks
It took real engineering to build, and it needs looking after. He also says his time saving is his own and not a benchmark for anyone else.
Where your genie helps
Seven agents is seven things to watch. Ugenie keeps track of where each one is up to and flags the one that's stuck.
Get Ugenie →Tools mentioned
LangGraph
An open-source toolkit from LangChain for building AI workflows as a series of connected steps.
- Cost
- Free and open source. You pay for the AI models it calls.
- Our verdict
- Powerful, but it's for developers. If you don't code, separate chats with saved instructions get you most of the way.
Mermaid
A way to draw diagrams from simple text, which AI models are very good at writing.
- Cost
- Free and open source.
- Our verdict
- A great trick for anyone. Ask any AI for “a Mermaid diagram of this process” and paste it into a Mermaid viewer.
Prices are rough and change often. Check before you buy.
Questions people ask
Can I build an AI content workflow without coding?
Yes. Use a separate chat or project for each stage, each with its own saved instructions, and a separate chat to review the result. You lose some automation but keep most of the quality gain.
Why use a separate AI to review the writing?
An AI reviewing its own work tends to approve it. A fresh chat that only sees the draft and the rules is more critical and catches more problems.
How much time does it save?
Bouchard went from two or three days per release to about five hours of his own time, but he is clear that this is his result, not a benchmark.
All credit to Louis-François Bouchard. We share things we think are genius and add our own take. Watch or read the original for the full thing.