Blog Outline: Text Renaissance

Working Title

“AI Made Text Great Again: The Text-Based Workflow Renaissance”


1. Why Text?

The Unix Philosophy

  • Everything is a file, text is the universal interface
  • Composability: small tools that do one thing well
  • Version control: git diff your documents, not binary blobs

The AI Renaissance

  • LLMs are text-native: they read text, generate text, transform text
  • Cloud tools (Overleaf, Google Docs) lock you out of this
  • Local text files + AI = the workflow Unix designers dreamed of

2. Trigger: Overleaf → Local LaTeX

The Move

  • Used Overleaf for years - convenient but limiting
  • No LLM integration (can’t use Claude Code or Copilot)
  • Collaboration limits on free tier

The Setup (Simpler Than Expected)

  • sudo apt install texlive-full - one command
  • VS Code + LaTeX Workshop + LTeX+
  • SyncTeX: click PDF → jump to source

What Changed

  • Claude Code can now edit my .tex files directly
  • Git history of my resume changes
  • Local fonts, local control

Practical Example: Awesome CV

  • Professional resume/CV template
  • XeLaTeX for custom fonts
  • Modular structure: education.tex, experience.tex, skills.tex

3. Extending to Slides: Markdown Wins Here

The Research

  • Compared 5 tools: Marp, Pandoc Beamer, python-pptx, reveal.js, Slidev
  • Key finding: Marp & python-pptx best for LLM integration (24/25)

When to Use What

ToolBest For
MarpQuick slides, multi-format output (PDF/PPTX/HTML)
Pandoc BeamerAcademic, math-heavy content
python-pptxFine-grained PPTX control

The Connection

  • Same philosophy: text source → rendered output
  • Same benefits: git, AI editing, automation
  • Markdown is simpler than LaTeX, good for ephemeral content

4. The Bigger Picture: A Text-Based Knowledge System

Note-Taking: Obsidian

  • Markdown files in a folder (you own them)
  • Links between notes, graph view
  • Plugins: LaTeX math, templates, daily notes
  • Sync via git or Obsidian Sync

Flashcards: Anki + LaTeX

  • Spaced repetition for learning
  • LaTeX support for math formulas
  • Can generate cards from markdown notes

The Full Loop

Reading/Learning
    ↓
Obsidian (notes in markdown)
    ↓
Anki (flashcards, LaTeX math)
    ↓
Writing (LaTeX for formal, Markdown for slides/docs)
    ↓
All versioned in git, all editable by AI

5. How AI Lowers the Barrier

The Old Problem

  • Gilles Castel’s workflow required mastering:
    • Vim motions and modes
    • UltiSnips for snippets
    • LaTeX syntax and packages
    • All simultaneously, steep curve

The New Reality

  • No need to choose from a ton of tools or spend a weekend learning them

  • Objects you operate on become intuitive: shape, color, table, equation

  • AI handles the syntax, you focus on content

  • Ask Claude: “Write a LaTeX table with these columns”

  • Ask Claude: “Why won’t this equation compile?”

  • Ask Claude: “Convert this markdown to Beamer slides”

  • Learn incrementally, AI fills the gaps

Not Replacing Learning, Enabling It

  • Still need to understand what you’re doing
  • But the “stuck” moments are shorter
  • Can focus on concepts, not syntax lookup

6. Getting Started: Practical Steps

Minimal Setup

  1. Install TeX Live: sudo apt install texlive-full
  2. Install VS Code + LaTeX Workshop
  3. Install Marp CLI: npm install -g @marp-team/marp-cli
  4. Pick one project: a resume, a presentation, notes

Don’t Boil the Ocean

  • Start with one piece
  • Add tools as you need them
  • Obsidian can wait, Anki can wait
  • The point is text files - the tools are secondary

7. Conclusion: AI Made Text Great Again

  • The Unix philosophy: text is the universal interface
  • Cloud-based AI services are convenient but create walls
  • AI assistants need text to work with you
  • The dream that seemed distant is now within reach
  • Started with one move: Overleaf → local. The exploration continues.

Appendix: Resources