I'm a non-coding knowledge worker and recently started exploring how I can treat my work as code.
I want to manage my work inputs and outputs as code - modular, interconnected, organized in a system...
And I want to build workflows similar to coding in Cursor = based on the prompt I or my AI will select relevant context and produce the outputs - without having to switch apps.
My v0 is Obsidian as the context storage and Obsidian Cannoli as the interface for interacting with LLMs.
I'm currently exploring v1 where I want to replace Cannoli with a more robust system - maybe VS Code with some LLM plugin, maybe some homegrown system on top of n8n.
This is really cool Peter, it's exactly what I had in mind. This meta-workflow applies to disciplines outside of software development
I'm curious what kind of knowledge work you do?
Cursor is a perfect analogy. Though like other tools, RAG and other techniques have to be applied judiciously to get great results. Technically, Claude and ChatGPT have Google Docs search features, but I still don't trust the right context is always included. Even with Claude projects I often times remind Claude which documents I'm expecting it to use.
Great article. I'm on a similar journey.
I'm a non-coding knowledge worker and recently started exploring how I can treat my work as code.
I want to manage my work inputs and outputs as code - modular, interconnected, organized in a system...
And I want to build workflows similar to coding in Cursor = based on the prompt I or my AI will select relevant context and produce the outputs - without having to switch apps.
My v0 is Obsidian as the context storage and Obsidian Cannoli as the interface for interacting with LLMs.
I'm currently exploring v1 where I want to replace Cannoli with a more robust system - maybe VS Code with some LLM plugin, maybe some homegrown system on top of n8n.
Thanks!
This is really cool Peter, it's exactly what I had in mind. This meta-workflow applies to disciplines outside of software development
I'm curious what kind of knowledge work you do?
Cursor is a perfect analogy. Though like other tools, RAG and other techniques have to be applied judiciously to get great results. Technically, Claude and ChatGPT have Google Docs search features, but I still don't trust the right context is always included. Even with Claude projects I often times remind Claude which documents I'm expecting it to use.