I haven't triaged my own inbox in months. I haven't manually organized a research note in over a year. When tax season came around, I handed Claude three years of documents and walked away. It found patterns I'd missed, flagged deductions I'd overlooked, and organized everything into a format my accountant actually thanked me for—and asked how I'd done it.
This isn't hypothetical. It's Tuesday.
If you use Claude primarily for code, you're using a general-purpose reasoning engine as a specialized tool. You're leaving most of its value on the table.
Drowning in Context
Knowledge workers are drowning. Not in work—in information about work.
According to Forrester, knowledge workers lose 30% of their workday searching for information. Post-pandemic research from Nakash and Bouhnik found that some workers now spend up to 1.5 working days per week just gathering and organizing information. Gartner reports it takes an average of 18 minutes to locate a single document.
We tried to solve this. For a decade, we built elaborate second brain systems—Obsidian vaults, Notion databases, Roam graphs, Evernote notebooks. We developed methodologies with acronyms: CODE (Collect, Organize, Distill, Express), PARA (Projects, Areas, Resources, Archives), Zettelkasten. Thousands of people took courses on how to build these systems.
Here's the uncomfortable truth: we solved the wrong problem.
Second brains are excellent at storage. They're terrible at thinking. You can have the most meticulously organized vault in the world, and you still have to do all the reasoning yourself. The bottleneck was never storage. It was synthesis.
A Category Error
When AI coding assistants emerged, we categorized them the way we categorize most software: by their primary use case. GitHub Copilot is a coding tool. ChatGPT is a chatbot. Claude is a coding assistant.
This was a category error.
What makes Claude good at code isn't a narrow capability tuned for programming. It's a general capability: the ability to take unstructured context, reason over it, and produce structured output. Feed it a messy codebase and a feature request, and it produces working code. Feed it a pile of research papers and a question, and it produces a synthesized answer. Feed it medical records from three different providers and ask for a timeline, and it produces one.
The mechanism is identical. Only the domain changes.
Code was the first killer app—not because AI is uniquely suited to programming, but because programmers were the first users with the technical sophistication to push the boundaries. They discovered what the technology actually was—a general-purpose reasoning engine—before the marketing caught up.
The rest of the world is still waiting for permission.
Consider this yours.
The Reframe
Stop thinking of Claude as a coding tool that can do other things. Start thinking of it as a general-purpose reasoning engine that happens to be packaged for developers.
The same context window that can hold an entire codebase can hold:
- Your inbox (thousands of emails, full conversation threads)
- Your financial records (bank statements, tax documents, receipts)
- Your medical history (records from multiple providers, lab results, prescriptions)
- Your research (papers, articles, notes, bookmarks)
If you can describe the context and the desired output, Claude can likely do it. That's not a coding skill. That's all knowledge work.
The question isn't whether the technology is ready. The question is whether you've updated your mental model.
My Playbook
Let me show you what this looks like in practice.
Research & Knowledge Management
Claude lives in my Obsidian sidebar. When I'm researching a topic, I don't just search my vault—I ask Claude to synthesize across it. It connects ideas I'd filed in different folders months apart. It identifies gaps in my understanding. It suggests questions I hadn't thought to ask.
When I save a new article or paper, I don't just file it. I ask Claude to extract the key claims, identify how they relate to my existing notes, and suggest where they should connect. My vault went from a graveyard of abandoned notes to an active thinking partner.
Email Triage
I point Claude at my inbox periodically. It reads everything, identifies what actually needs my attention, drafts replies to routine messages, and extracts action items into a structured list. What used to be a 45-minute daily ritual now takes about 10 minutes of review and approval.
The key insight: most emails don't need me—they need information or a standard response. Claude handles those. I handle the ones that actually require human judgment.
Financial Analysis
Tax season used to mean a week of gathering documents, categorizing expenses, and second-guessing whether I'd missed something. Now I export my records, hand them to Claude, and ask specific questions: "What deductions might I be missing for home office expenses?" "Are there any unusual patterns in Q3 spending?" "Organize these documents for my accountant."
It's not replacing my accountant. It's making me a better client.
Medical Records
I've collected medical records from four different providers over the past decade. None of them talk to each other. Getting a coherent timeline of treatments, medications, and test results used to require hours of manual compilation.
Now I hand Claude the stack of PDFs and ask: "Create a chronological health timeline. Flag any patterns or concerns I should discuss with my doctor." I walk into appointments with questions I wouldn't have known to ask.
The Meta-Level
But the real shift happened when I stopped using Claude for individual tasks and asked it to look at everything.
"Look at my docs and pull together interesting info."
It came back with a meticulous knowledge base: projects, personal, financial, health—each section filled with synthesized information I'd scattered across years of notes. Connections I'd never made. Patterns I'd never noticed. A structure I wouldn't have thought to create.
Claude didn't just work within my system. It helped design the system.
Building Your Command Vocabulary
Once I saw what was possible, I wanted to systematize it. I noticed I was typing the same prompts repeatedly—same preamble, same instructions, same output format. So I built custom commands.
Think of it like a personal CLI for life. Unix commands each do one thing well: ls, grep, cat, sort. My Claude commands work the same way:
| Category | Commands | Purpose |
|---|---|---|
| Capture & Research |
/note, /todo, /research
|
Input goes in, structured output comes out |
| Daily Rituals |
/status, /eod, /standup, /prep <meeting>
|
Woven into the rhythm of work |
| Maintenance |
/cleanup, /organize
|
Keeping entropy at bay |
/status gives me a current state across all projects. /eod wraps up my day—summarizes what happened, identifies loose threads, sets up tomorrow. /prep <meeting name> pulls relevant context and talking points before I walk into a call. /research <topic or url> does a deep dive and returns structured findings.
These aren't productivity hacks. They're a vocabulary. And like any vocabulary, once you have the words, you can express thoughts you couldn't before.
The progression looks like this:
- Discover Claude works beyond code
- Start using it for specific domains
- Notice repetitive prompts
- Build custom commands
- Now you have a personal operating system
Trust But Verify
I'm not going to pretend this is magic.
Claude still hallucinates occasionally. Specific facts need verification. Dates and numbers deserve a second look. For anything high-stakes—medical decisions, legal documents, financial filings—Claude is a powerful first pass, not a replacement for professional advice.
Trust but verify. Let Claude do the synthesis. Apply human judgment where it matters. This isn't fundamentally different from how you'd treat any capable assistant—you'd still review their work on important matters.
The difference is the breadth. Most human assistants specialize. Claude doesn't have to.
What Are You Leaving on the Table?
If you're only using Claude for code, what else could you be doing with a general-purpose reasoning engine that can hold 200,000 tokens of context—roughly 150,000 words, or three novels' worth of your life?
The knowledge workers spending 30% of their day searching for information—that's solvable. The second brain systems that store but don't think—Claude thinks. The administrative overhead of email, scheduling, document organization—most of it is pattern recognition and text transformation, which is exactly what these models excel at.
The agent is ready. The capability is here. The bottleneck is the mental model that says "this is a coding tool."
It's not. It's a reasoning engine. And reasoning is what knowledge work is.
The question isn't whether AI can help with the rest of your work. It can. The question is whether you're willing to find out what you've been leaving on the table.
Start with one domain. Build one command.
The general-purpose agent has arrived. What are you waiting for?
Top comments (0)