You’ve probably noticed the official ChatGPT and Codex apps are heavy. Dog-slow. You open a Word document, and the formatting renders like a crime scene. You blame the AI. You blame the bloat. But the truth is much weirder.
The most advanced AI labs in the world aren’t building bespoke file parsers—they’re duct-taping a decades-old open-source office suite to their billion-dollar models.
When you ask ChatGPT to read your spreadsheet, it’s not doing some neural-network optical character recognition. It’s passing the file to a hidden, bundled instance of LibreOffice. Yes, the same free, open-source software you used to download as a student because you didn’t want to pay for Microsoft Office is secretly powering the world’s most hyped AI product.
The Hacker News crowd is furious. They look at the massive app size and ask: “Why with all their cash don’t these companies redo their apps in Rust or native? They are all so heavy and dog slow.” It’s a valid complaint. The app is bloated, and it explains why your files look terrible. But here is the contrarian truth: bundling LibreOffice is actually the smartest engineering decision they could have made.
Legacy document formats are an infinite labyrinth of pain, and rewriting them from scratch is a multi-year sinkhole that ends in tears.
Have you ever looked at OOXML? The standard behind Microsoft Word? It’s a sprawling, convoluted mess of backward compatibility, legacy edge cases, and corporate compromise. To perfectly render a 15-year-old .docx file with weird embedded Excel macros and custom fonts, you don’t need a machine learning model. You need the exact same battle-tested, decades-old codebase that has been patching these edge cases since the early 2000s. LibreOffice is that codebase.
If you use AI to create or edit Office documents, this dependency affects you directly. It dictates your file fidelity, ruins your app performance, and expands your security surface. You are trusting a massive, hidden office suite to not break when fed garbage data.
We want our AI overlords to be pristine, futuristic engines of logic. Instead, they are held together by open-source duct tape and sheer pragmatism.
AI isn’t magic. It’s a clever UI slapped on top of the same messy, legacy infrastructure the rest of the software industry uses. They didn’t reinvent the wheel. They just asked an AI to drive a beat-up old car. And honestly? It gets you to the destination.
FAQ
Q: Isn't bundling a full office suite terrible for app performance and security?
A: Yes, it bloats the app and expands the attack surface, but it guarantees instant compatibility with messy legacy files, which is the harder problem to solve.
Q: How does this affect my daily ChatGPT usage?
A: If you upload complex Office documents, expect occasional rendering glitches and slower performance. The AI is relying on a hidden office suite to parse your files, not native AI logic.
Q: Should AI companies just rewrite everything in modern languages like Rust?
A: No. Rewriting legacy file format parsers from scratch is a notorious sinkhole. Borrowing mature open-source infrastructure is the pragmatic, correct engineering call.