July 5, 2026
Readme wars: knowledge edition
Knowledge Should Not Be Gated
Techies are ditching the lockbox and screaming: just let the bots read the files
TLDR: The article argues that AI works surprisingly well when knowledge is stored in simple readable text instead of hidden in complex systems. Commenters turned that into a bigger fight about missing sources, corporate data risks, and whether the internet has quietly made knowledge harder to reach.
The big idea in this piece is almost hilariously simple: instead of stuffing a company’s knowledge into complicated systems only software can decode, just write it down in plain text files that humans and AI can both read. That’s the vibe behind Google’s Open Knowledge Format, and the community reaction was not calm. For some readers, this was a long-overdue “finally!” moment. One commenter basically turned it into a battle cry, saying the old hacker dream was that information should be free, while another compared modern AI to a search engine that gives answers with no visible sources — a spicy complaint that hit a nerve.
But the thread didn’t stay idealistic for long. It swerved straight into paranoia, privacy panic, and good old internet gloom. One of the hottest takes warned that handing private company data to AI services is basically volunteering your business secrets for mining, asking why more corporations aren’t freaking out. Another commenter went even broader, lamenting that the open web of the 1990s has been replaced by “approved sources” and surveillance vibes. Meanwhile, not everyone bought the article’s villain. One pushback argued that software kits and tools were never the gatekeepers here — they were just the boring plumbing that made tricky systems usable.
So the real drama? A simple file format became a proxy war over who controls knowledge, whether AI should show its homework, and whether convenience has quietly built a maze around information that used to be easy to read.
Key Points
- •The article argues that AI knowledge systems have often depended on infrastructure such as embeddings, vector databases, retrieval layers, SDKs, and graphs.
- •RAG and Graph RAG are described as useful responses to earlier limitations in context windows and model costs.
- •The article says these retrieval systems transform human-readable documents into machine-oriented formats that are harder to inspect directly.
- •It highlights a growing practice of storing agent context in markdown files like CLAUDE.md, AGENTS.md, DESIGN.md, and MEMORY.md.
- •The article presents Google’s Open Knowledge Format and Andrej Karpathy’s LLM Wiki pattern as signals of a shift toward plain-text, standardized knowledge formats for LLMs.