Everyone thinks Claude Code is for coding. The name certainly doesn’t help. But after three months of daily use, I’ve written maybe 20% code and 80% everything else.
Claude Code is my Obsidian copilot, my executive translator, my decision framework, and occasionally, yes, my code generator. Treating it as just a “coding agent” is like using a Swiss Army knife only for its blade.
My Obsidian vault has thousands of notes. Pre-Claude Code, finding connections between ideas meant manual searching, tagging, and hoping I remembered that brilliant insight from three months ago. Now, through MCP integration, I ask Claude Code to find all mentions of technical debt in my meeting notes and correlate with project delays. It searches hundreds of files in seconds, finding patterns I’d never see: technical debt discussions spike 2-3 weeks before deadline extensions, consistently across major projects.
This isn’t coding. This is knowledge archaeology.
Different audiences need different formats. Engineers want technical specs. Business stakeholders want narratives with strategic frameworks. Finance wants metrics. I give Claude Code the same information and ask for three compilation targets. For the technical audience: bullet points about compute clusters and ROI timelines. For strategic stakeholders: the same facts woven into narrative about modularization principles (模块化) and platform effects (平台化). The information stays constant; only the protocol changes.
I’m not asking it to code. I’m asking it to compile human protocols.
Every significant decision now goes through what I call the “Oracle Pattern”: I explain the decision to Claude Code, it asks Socratic questions I hadn’t considered, we explore edge cases together, and then it summarizes what I actually decided - which is often different from what I thought I decided. This collaborative ideation process transforms how we approach complex problems, creating new forms of cognitive partnership that amplify strategic thinking rather than replacing it.
Last week I asked whether to optimize for startup equity or enterprise stability. Claude Code asked what I’d regret more in five years - missing a potential unicorn or explaining a failed startup to my family. When I said “the family question,” it observed that I’d already decided: I was optimizing for narrative coherence with my life story, not maximum expected value.
This isn’t coding. It’s cognitive partnership.
Here’s where it gets interesting. I’m writing this article about Claude Code with Claude Code itself. Every paragraph has been a conversation. When I proposed the title, Claude Code noted how the contrarian position creates cognitive dissonance - while everyone writes “I built an app with AI,” I’m demonstrating systems thinking and tool transcendence. It suggested the structure, challenged my examples, and even pointed out that mentioning the meta-layer would strengthen the argument.
We’re not coding. We’re co-thinking.
The real power isn’t in the impressive demos. It’s in the mundane, daily augmentations. Email archaeology becomes possible when you ask Claude Code to find all emails where you promised deliverables and check if you delivered. Meeting synthesis transforms from painful note-reviewing to a simple question about what you actually committed to across five meetings. Knowledge pruning happens when you wonder which notes haven’t been touched in six months but contain actionable insights. Even emotional debugging becomes accessible - asking why you keep avoiding a particular task often surfaces patterns you hadn’t consciously recognized.
None of this is coding. All of it is invaluable. The magic isn’t that Claude Code can do these things, but that we don’t think to ask until we stop seeing it as a coding tool.
My Claude Code workflow has nothing to do with npm or git. It starts with context injection - feeding my entire Obsidian vault structure at session start so Claude Code understands my knowledge landscape. Throughout our conversations, it maintains memory patterns about my projects, stakeholders, and communication styles. Each interaction becomes more precise than the last.
Every major decision gets documented in our conversation, creating decision logs I can search months later to understand not just what I chose but why. The most powerful tool might be synthesis triggers. Simply asking “What patterns do you see?” yields insights I’d never surface alone. I’m too close to my own thinking.
The code editor? Haven’t opened it in weeks. The real IDE is the conversation itself.
We’re naming these tools wrong. “Claude Code” implies coding. “Copilot” implies assistance. But Claude Code is not a coding agent - it’s a thought partner that happens to also write code.
When I “prompt” Claude Code, I’m not asking it to write code. I’m asking it to transform my scattered thoughts into coherent strategy. Convert my implicit knowledge into explicit frameworks. Translate between different human protocols - technical to executive to philosophical. Debug my decision-making process.
The code generation capability? That’s just one feature among many.
Here’s what nobody wants to admit: Claude Code is better at being a thought partner than a coding partner. When it writes code, I have to review every line. When it helps me think, it surfaces blindspots I didn’t know existed. The ROI on augmented thinking dwarfs the ROI on generated code. This shift from implementation focus to strategic thinking represents a fundamental reordering of cognitive priorities —what and why are becoming more important than how.
Yet we keep calling it a “coding agent” because that’s safer. It’s easier to say “AI writes my code” than “AI shapes my thoughts.” One threatens developers. The other threatens everyone.
I’m not telling you to use Claude Code for non-coding tasks. I’m telling you that if you’re only using it for coding, you’re like someone using their iPhone only for phone calls.
The future isn’t AI that codes for us. It’s AI that thinks with us. And despite its name, Claude Code is already there.
P.S. - Yes, I used Claude Code to build this entire blog site in 90 minutes. But that’s the least interesting thing it did this week. The most interesting? It helped me realize I was solving the wrong problem in my career strategy. That conversation saved me two years of misaligned effort. This kind of strategic insight emerges when you stop reading documentation and start observing AI work— the learning paradigm shifts from memorization to pattern recognition. Try billing that as “coding assistance.”
P.P.S - “ultrathink” isn’t a typo. It’s what happens when your thoughts continue beyond the message boundary. Claude Code understands this. Does your coding agent?