Using AI to build knowledge, not just content
One of the more interesting shifts in how people use AI is moving away from treating large language models as fast answer machines. A
recent piece by Andrej Karpathy describes how he now uses AI less to write code and more to organise knowledge: feeding research material into a system that turns raw inputs into a structured, interlinked body of notes, updated over time.
The appeal is straightforward. Most AI interactions are disposable. You ask a question, get an answer and move on. A knowledge base is different. It keeps the value. That makes AI less like a chatbot and more like an ongoing research assistant. Karpathy describes a markdown-based system that incrementally builds and maintains a persistent set of notes from source material, rather than relying on retrieval at query time alone.
As we’ve written before, AI becomes more valuable when it helps teams accumulate understanding, not just accelerate isolated tasks. That’s a more durable use case than asking a model to generate another email, summary or block of code. It points to a shift from output to memory, from speed to continuity.
The caveat is worth stating clearly: any AI-built knowledge system is only as trustworthy as its sources, its structure and the people reviewing it. But the broader lesson holds. The next useful wave of AI may not be tools that simply produce more. It may be tools that help people retain, connect and reuse what they already know.
