**The problem & the claim:** The results you get from AI are determined by the information it has about you. If it knows what you are working on, why you are working on it, what happened last week, anything you are trying to figure out, it can make better decisions and give you far better results compared to if it did not have all of that information. **Defining context:** Every AI tool has some kind of memory now. Claude remembers things about you. ChatGPT remembers things about you. But that memory is built from whatever happened to come up in conversations you’ve had. It’s never a complete picture of something you are thinking about or working on. To get better results, you need more control over what information the AI actually has. If there’s something specific you want an agent to know about a project, you should not have to explain it over and over again each time you create a new session. It’s way too slow and it takes too much effort. **The Solution**: Most people do not have a system for this. But there’s an emerging set of tools that people are starting to use to design and manage the context they share with AI agents, making the whole process faster and more deliberate. I think it’s really cool, and I think it leads to really compelling new experiences with a computer. The tool I use to mange the information I give to AI agents is called [[Obsidian]]. It’s a note-taking app where all of your notes are stored as markdown files on your computer. The thing that’s made [[Obsidian]] very interesting to me and to others is that [[Obsidian]] is not just a user interface for editing a collection of markdown files but it also allows you to easily link files together. If I mention a person in a note and that person has their own note, I can connect them. If a project is related to an idea I had three months ago, I can link them. Over time you are not just building a collection of files, but you are building a web of interrelated ideas, and that keeps growing the more you write and think. Now I also use an AI agent called [[Claude code]] that can read and write files directly on my computer. So if I write a document about a project I’m working on, I can pass that file to the agent as context. Now it knows what I know. Next time I start a session about that project, I pass that same file in again. I never have to re-explain. That alone is pretty awesome. You can start actively documenting things, maintaining files about your projects, your team, your thinking, and any time you want to work with an agent, you just hand it the relevant file. This is so much faster than explaining things over and over again. But until recently, even though my vault had all of these connections between notes, the agent couldn’t see them. I could point it to a file, and it would read that file. And let’s say I linked to a file, it could try and read that other file by guessing its file path, but it didn’t have an immediate understanding of all of the other files connected to what i was writing. Basically the broader web of ideas is invisible to the AI agent. That changed when [[Obsidian]] released [[Obsidian CLI]]. This allows the AI agent to follow links between notes. It can ask “what’s related to this?” and actually get an answer. It can see the web. This changes a lot. These interrelationships between ideas are very powerful but finding them and seeing them was limited to what I can see in my head or what I could observe by just opening a random note and seeing a linked file or by spending time looking at the Obsidian graph and hoping I would notice some kind of connection. Now the AI agent can do that and it can hold more of the web in its head at once. It can see things and make connections much faster and effectively than I ever could. It can instantly surface something I wrote a year ago that’s directly related to what I’m thinking about today. It can find patterns in my own thinking that are impossible for me to see on my own.