Claude Code Is Not Just for Coding
I recently completed a complex and time-compressed technical requirements analysis for a multi-organisational datalake. This involved over 12 stakeholders, three vendors and multiple interviews, conversations with each. Working as a sole-supplier with large amounts of valuable information incoming is a serious responsibility, the sheer volume of potential gold dust that could be missed. I decided that this would be the perfect opportunity to leverage two great AI tools to aid me capture, summarise and build a complex growing picture of needs and platform characteristics; my trusty notetaker Fathom and Claude code.
To make doubly sure that I did not miss any nuance, I would even enlist Gemini CLI to develop its interpretation with the same context - in fact the results of this could form the basis for a post on its own.
For those that haven’t tried it in this vein, Claude Code is an excellent environment for knowledge work and not just coding. Tracking lengthy transcripts, extracting themes of agreement or disagreement, emerging themes and dependencies. Using a healthy amount of context engineering upfront and planning with a smattering of bash for orchestrating file management and task tracking with my todoist account, proved to be a good recipe for both the planning of the questions, the requirements gathering framework and the transcripts. For those that have yet to embrace the terminal in their day to day, it is a highly satisfying experience. There is to me a distinct difference in behaviour of the Anthropic models in Claude Desktop with Claude code being more adept at complex operations such as trawling through long text dialogues, summarising and making subtle deductions.
Managing the context window is also more intuitive in Claude Code. With large transcript documents, it was essential to work in pre-planned mini sessions, ensuring that on wrapping up, Claude Code would auto create a session summary that would be picked up on a new session. As part of the wrap up, a requirements tracker was updated for both communications to my stakeholders but also as a briefing for Claude each time we started meshing between new pieces of incoming information and what was already collected.
Finally, I’d like mention Sameer Modha who has been a guiding light and inspiration in my adoption of Claude Code in recent months particularly for knowledge work such as this.
I’d love to hear from others using AI tools for knowledge work beyond coding, whether its because you would like to learn more from me or just keen to share your thoughts