JBS Weekly
This week I built a custom content pipeline for a client who had a real problem: he wanted to run a faceless YouTube channel on psychology, attention, and modern life, but he has a full-time job and zero hours left over for production.
The channel idea was solid. The bottleneck was everything between the idea and the finished video. So we built a tool that handles most of it.
Here's how it came together and what I had to fix along the way.
🛠️ This Week’s Build
The client came to me with a clear vision for his channel: animated essays about the systems that shape how people think and behave. Topics like attention, manipulation, identity, distraction. He had real opinions on all of it. What he didn't have was time. His full content production process would have taken a full week on top of his day job. That's not sustainable for one episode, let alone a channel.
I built him a custom web app that walks him through the entire production process from a single interface. He enters a topic, picks a content pillar from a predefined list, and clicks to start a new episode. From there, the app moves him through six stages: topic research, a perspective interview, script generation, visual scene planning, thumbnail creation, and a voiceover and music brief. At the end, it generates an upload package he can push to YouTube himself.
A few things worth noting about how it's built. The front end is intentionally lightweight. Almost every action triggers a web-hook that hands off to a self-hosted n8n workflow on the back end. Each stage has its own workflow, so nothing is tangled together and costs stay low. Gemini handles the research. ChatGPT image handles the visual and thumbnail prompts. Blotato handles the actual image generation and video editing.
The perspective interview step is the most important part of the whole thing. The client was very clear that he didn't want AI generating generic content from whatever it found on the internet. He wanted his take. So the app runs him through a chatbot-style conversation, pulls out his actual experience and opinions on the topic, and uses that as the foundation for everything that follows. The script reflects his voice, not a research summary.
Mid-build, I hit a UI problem. Every stage was returning raw JSON directly in the interface. Fine for a machine to parse, hard for a person to read. I reworked the display so the JSON collapses to the bottom and the main view shows clean, human-readable markdown. I also added a regeneration box for each section so he can flag specific changes without rerunning the whole stage.
The result: a process that used to take a full week now takes about two hours. He can split it across a lunch break and an evening. The content still reflects his perspective, he just isn't doing the mechanical work anymore.
📰 AI News This Week
The White House Is Controlling Who Gets GPT-5.6 First
The Trump administration asked OpenAI to stagger the release of GPT-5.6, requiring government approval before each new customer gets access. Sam Altman has said this is the safest path to broader release, expected a few weeks out, and that it won't be the permanent model. This is a meaningful precedent: the government now has a direct hand in deciding which businesses can access frontier AI models and when.
Joe's Read: This affects any company that was expecting to adopt GPT-5.6 on day one, since you may need to wait for approval depending on whether your use case qualifies as government-sanctioned.
Microsoft Expands Copilot in Excel with Finance-Specific Capabilities
Microsoft added new finance-focused skills to Copilot in Excel, including trusted data connectors and improved audit trails. This isn't a general-purpose update. It's targeted at financial workflows where traceability and data sourcing matter. If your ops team lives in Excel for budgeting, forecasting, or reporting, this is worth a closer look.
Joe's Read: This directly affects finance and ops teams using Excel for complex workflows who have been waiting for Copilot to do more than basic summarization.
Apple Raises MacBook and iPad Prices Due to AI-Driven Chip Shortage
Apple hiked prices on several MacBook and iPad models by $100-$400 because memory chip costs have more than doubled since October 2025. The shortage is driven by AI data center demand, and analysts expect prices to keep climbing into 2027. This is AI cost showing up somewhere most people weren't looking: your next hardware refresh budget.
Joe's Read: This hits ops and IT decision makers at small and mid-size companies who are planning hardware purchases in the next 12-18 months and need to revise their cost assumptions now.
🧰 Tool Worth Trying This Week
Open Engine (Multi-Agent Task Handoff System)
Open Engine is a copy-paste framework for building AI workflows where one model's output becomes the next model's input, without a human manually transferring the work in between. It uses a structured task record with attached sources, limits, and receipts so context doesn't get dropped when tasks move across tools or agents. The smallest version you can build is one task, one agent, one blocker rule, and one receipt.
Caveat: This is not a plug-and-play product with an interface. It's a structured guide and template system, so you'll need to be willing to set up the underlying workflow yourself, likely in a tool like Linear or a similar task manager.
🗺️ From The Field
This build reinforced a simple rule: do not automate the part where the judgment happens.
In this pipeline, the client’s judgment happens during the perspective interview. That is where he explains what he believes, what he disagrees with, and what angle he wants the episode to take.
Everything after that is production. Research. Structure. Script support. Visual planning. Thumbnail prompts. Upload packaging.
That distinction matters. When you separate judgment from production, you know what to protect and what to automate.
🤔 Joe’s Take
AI slop usually starts with a bad assumption: that faster content is the same as better content.
It is not.
The speed only matters if the system preserves the thing people came for in the first place: the creator’s point of view.
That is why I liked this build. The client was not asking for a shortcut around having an opinion. He wanted a way to turn his opinion into finished work without losing a week every time.
That is the kind of leverage AI should create.
⚒️ Tools I Use
n8n — This week's client build runs almost entirely on n8n: each stage of the content pipeline hands off to its own self-hosted workflow, which keeps the stages independent and costs low. If you have a production process with multiple steps that someone is currently moving by hand, this is the tool that stops that.
VoiceInk — A local AI dictation tool for Mac that transcribes your voice with near-perfect accuracy and runs entirely on your device, meaning nothing you say ever touches a cloud server.
Blotato — Blotato handled the image generation and video editing side of this week's client build, sitting at the end of the pipeline where the visual and thumbnail prompts get turned into finished assets. If you're building any kind of content production workflow, it covers the distribution side too: scheduling, platform-specific formatting, and native publishing to 9 platforms with no per-post fees.
Beehiiv — What you're reading right now is published on Beehiiv. If you're thinking about starting a newsletter or moving off a clunky platform, this is the one I'd recommend. 20% off your first 3 months with my link.
Google Workspace — Gemini handled the research stage in this week's content pipeline build, and it's included in a Google Workspace Business Standard plan alongside NotebookLM Plus and the full enterprise suite. Better value than a standalone Gemini subscription when you're already paying for Google anyway. 14-day trial and 10% off your first year.
Descript — The client's content pipeline ends with a voiceover and music brief, and that audio still needs to be edited once it's recorded. Descript handles that side: you edit the transcript and the media follows, with automatic filler word removal and captions built in. 50% off your first two months on the Creator Plan.
💭 Final Thoughts
Most people think about AI content tools as something that generates output. This one is designed around capturing input first. That distinction matters because the quality of what comes out is directly tied to what the person put in. If you're building any kind of production pipeline, figure out where the human's judgment is irreplaceable, and build the tool around protecting that step.
PS: If you want a content production process that runs in two hours instead of a full week while still sounding like you, book a discovery call and we'll build the solution together in a live session.
Cheers,
Joe