@patio11: A thing I find myself doing frequently enough recently to mention pattern: creating or improving internal tools with th…
Summary
The author describes a pattern of creating or improving internal tools to make it easier for LLMs to interface with business operations, using methods like native Markdown support and shared secret API users.
View Cached Full Text
Cached at: 09/12/26, 04:44 AM
A thing I find myself doing frequently enough recently to mention pattern: creating or improving internal tools with the specific intent of making it easier for LLMs to interface with the rest of life/biz.
Make the tool speak MD natively, add shared secret API user, etc etc.
Similar Articles
@rohit4verse: karpathy moved his token budget from writing code to building a knowledge base. markdown files an llm writes and mainta…
A guide on using LLMs like Claude to build and maintain a knowledge base in Obsidian, with automated daily briefings and note-taking.
@lindsmccallum: the one where @benthompson discovers codex
A personal reflection on using AI coding tools (Codex) to build an app, with ten takeaways from the experience.
@DataScienceDojo: Andrej Karpathy's LLM Wiki is a pattern for building a personal knowledge base that actually compounds over time. The d…
Andrej Karpathy's LLM Wiki pattern enables building a persistent, structured knowledge base that compounds over time, unlike stateless RAG systems. The tutorial shows how to create one in under 30 minutes using LLMs to compile and link markdown pages.
@karpathy: One pattern I find useful for working with LLMs is a nice long ramble session. Sometimes the LLM needs more bits to und…
Andrej Karpathy shares a tip for working with LLMs: using voice to ramble incoherently for 10 minutes, which helps the model understand the user's intent and often produces a cleaner reformulation.
@mattpocockuk: One thing missing from my skill set is what compound engineering has - the promise that your process improves over time…
A tweet explores the gap in AI skills for process improvement over time, suggesting a potential tool like '/improve-me' to update coding standards and refine agent behaviors.