File Formats in the LLM Era

It was easier for me to format a table directly in HTML using Codex than in Excel.

Throughout my career at Zoho and OSlash, I’ve gained firsthand insights into office suite products (MS Office, Google Workspace) and developed a fascination with file formats (xlsx, docx, pdf).

Makes me wonder - what kind of file formats will succeed in LLM era?

I believe, we’ll transition from machine-optimized formats (pdf, xlsx, docx) to human-readable formats (markdown, .txt, .csv, .json).

• LLMs with Markdown: Instant processing, near-perfect accuracy (8,000 tokens)
• LLMs with DOCX: Requires conversion, risks data loss and formatting issues (25,000 tokens)

Hence, it is faster for a LLM to read code than to read contracts.

This will fundamentally change the modern office suite as we know it.

Today most of the tools such as Excel, Word are format first editor (500+ buttons), in future you might have just one tool infinite context with a dual mode interface:

• Compose mode: plain markdown with AI assistance
• Present mode: full formatting applied automatically (the rendering could be a PDF, HTML, Doc, Table)

Companies who will internalize this insight will also create content that are more human-readable (LLM ready) rather than machine first!

It could be a huge competitive advantage in today’s token driven era!