I’ve been testing various skills on my Video Worker, which is part of a larger ICM-based Suite I’ve been building over the last few weeks.
The idea started with a question: "Can I take the power of an ICM workspace and make it genuinely useful to someone who has no interest in becoming technical?"
I had already built a collection of workers for myself — Conductor, Research, Writing, Transcription, Design, Builder, Video and Marketing.
They grew out of my own needs as a teacher and content creator. Over time, I realized the real value wasn't any individual worker.
It was that they shared context.
Research could inform the Writer. The Writer could feed Design. A transcript could become content. Video could work from the same source material. Marketing could understand what had actually been created rather than starting from a generic prompt.
The problem was that the architecture that made this possible was also the thing most non-technical users would never want to deal with.
So I built a GUI over it and called it Quiet Scribe.
The organizing concept is a nonfiction book.
Not because everyone necessarily wants to become an author, but because a book is a great way to force knowledge into structure. Once that knowledge is captured and organized, the same context can become articles, posts, visuals, videos, a course, a website, or marketing material.
My first demo failed pretty badly. 😂
The interface looked great, but exposing the workers through the GUI revealed architectural gaps I hadn't noticed while using them directly.
That sent me into three days of auditing the workers: folder structures, instructions, retained context, unfinished changes, handoffs, and things we'd learned through use that had never made it back into the architecture.
The second demo was probably 80% better.
More interestingly, improving the underlying architecture improved what the workers could do together.
The Video Worker in this clip, for example, can now take a long demo/transcript, work from timestamps, identify different storylines, and develop shorter video treatments from the source material.
That's really the experiment for me: "Can I hide most of the complexity of ICM without losing what makes ICM powerful?"
Open one interface. Explain what you're trying to accomplish. Let the Conductor orient you. Move between specialized workers while the underlying knowledge and context remain connected.
And if someone eventually wants to understand what's underneath, the architecture is still there to explore.
This video came from the latest demo (I probably should put it on YouTube and share it from there hahaah)
Still plenty to improve — and still finding bugs every time I push it — but it's getting much closer to the experience I originally imagined.