Obsidian makes AI 2x faster

I started using Obsidian recently (literally today), and I've already noticed a significant difference when using it alongside AI. I'm using it as a knowledge base for a project I'm developing with the help of AI. Before switching to Obsidian, I kept all the decisions we made in a single file called "planejamento.md". This file contained all the planned features for the application, processing logic, data policies, technologies that might be used in the project, questions the AI itself would ask me to clarify things that hadn't been decided yet, and a lot of other information. The file had already grown to over 2,000 lines, which was starting to slow down the AI's responses because it had to read the entire file every time it executed a task. So I decided to split it into several smaller files and connect them through Obsidian. I asked the AI to plan the transition from one large file to multiple smaller, organized sections, along with two main files: "MASTER.md" and "WORLDMAP.md". "MASTER.md" works essentially like Codex's "AGENTS.md". It contains instructions on how the AI should behave before executing each task. Its main purpose is to guide the AI through analyzing my request, identifying the topic we're discussing, and checking "WORLDMAP.md" to determine which files are actually needed to complete the task and achieve the best possible result. "WORLDMAP.md", on the other hand, acts as a navigation guide. It contains summarized information about all the other files in the vault, organized into tables divided by category. It also includes links to every file, along with a brief description of what each one contains. Just switching from one massive file to multiple smaller ones has already made my AI roughly 2x faster on simple tasks. For example, at my current stage of development, ChatGPT is asking me questions to define specific features of the application. The process works like this: the AI presents me with three questions, each containing two or three possible answers, along with a recommendation based on the project's existing context. I can simply choose one of the suggested options or provide a completely different answer. After that, the AI saves the questions, available options, recommendations, and my answers, along with the date of the change and a record of what was modified. I'm currently using GPT-6.1 Sol with medium reasoning effort for these tasks. Before using Obsidian, each execution took around 5–6 minutes. Now that I've separated everything into smaller files, the exact same process takes 3 minutes at most, simply because the AI only needs to read the files that are actually relevant to the task.