Detailed Analysis
A Reddit thread titled "Keeping it organised" surfaces a common friction point among power users of Claude Code as they scale from casual chat-based interactions to more structured, agentic coding workflows. The original poster describes a familiar arc: starting with Claude's chat interface, where "Projects" provided built-in organizational scaffolding, and then transitioning to Claude Code, where the lack of an equivalent structure led to disorganization. The user manages two separate businesses, each with multiple departments, and is asking the community for practical folder-structure and workflow advice—specifically whether to create separate directories per business/department and launch Claude Code sessions within them, and whether to future-proof custom "skills" or prompt configurations by storing them in an external tool like Obsidian rather than locking them into Anthropic's ecosystem.
This question reflects a broader maturation happening among Claude Code's user base. Early adopters of coding agents often treat each session as ephemeral and self-contained, but as usage scales across multiple projects, clients, or business units, the absence of native organizational primitives (akin to "Projects" in Claude's web/desktop chat app) becomes a real productivity bottleneck. Claude Code, as a terminal-based agentic tool, inherits the file-system-driven structure of software development environments—meaning organization is largely left to the user's own directory conventions, CLAUDE.md files, and git repo boundaries, rather than any built-in workspace concept. This is a meaningful UX gap for non-engineers or solo operators using Claude Code for business operations rather than pure software development, since they don't have the same intuitive mental models around repos, branches, and folder hierarchies that professional developers do.
The Obsidian question is particularly telling of a broader trend: users increasingly want to decouple their "prompt engineering capital"—custom instructions, reusable skills, personas, and workflows—from any single vendor's platform. This mirrors anxieties seen across the AI tooling space, where switching costs and vendor lock-in are top of mind as the LLM market remains fiercely competitive between Anthropic, OpenAI, Google, and others. Storing reusable "skills" (a term gaining traction with Anthropic's own "Agent Skills" feature, which lets users package instructions, scripts, and resources into portable, model-agnostic folders) in a tool like Obsidian—a local-first, markdown-based note-taking app—offers portability, version control via plain text, and independence from any specific chat interface. This desire for LLM-agnostic knowledge management is becoming a recurring theme as businesses build increasingly complex, custom automation stacks and want assurance that switching providers later won't mean rebuilding their entire prompt library from scratch.
More broadly, this thread illustrates how the "vibe coding" and agentic AI movement is pushing non-traditional users—small business owners, solopreneurs, consultants—into workflows historically reserved for software engineers, and exposing the gap between developer-oriented tools and the organizational needs of business operators. As Claude Code and similar agentic coding tools (GitHub Copilot Workspace, Cursor, etc.) gain adoption beyond pure engineering teams, there's growing pressure on vendors to either build better project/workspace management directly into these tools or to formalize best practices (structured folder templates, standardized CLAUDE.md conventions, modular skills libraries) that help non-technical users replicate the clarity they once had in simpler, project-based chat interfaces. The community-sourced nature of this thread—users turning to Reddit rather than official documentation—also underscores that Anthropic's own guidance on scaling Claude Code across multi-business, multi-department contexts remains thin, leaving practitioners to invent their own conventions.
Read original article →