As product managers, we all have our ways of messaging, company knowledge, task management, interacting with stakeholders. This is just the life of a PM. What we’re adding into the picture is a copilot - an LLM with a lot of context. But all the context is something I decided it can know. It’s giving me a lot of help, but all of it’s still going through me. In the beginning, when LLMs came out, product people were really excited that it could write a lot of stuff for us. But I think that’s the least exciting use case. If you give your copilot a lot of context, it can do way more high level tasks like thinking and strategic decisions and soft skills. Writing suddenly becomes an afterthought. What does that actually mean? Three levels of context: Company level context: Everything that has nothing to do with a particular initiative but is true across teams. What would you give a new PM on their first day? That’s company context. Initiative level context: What I’m actually working on right now. The opportunity assessment, research, stakeholder feedback. I keep updating this over time. Custom instructions: How I want my copilot to behave, what values and principles I want it to use when pushing me in different directions. Every major LLM platform has this same pattern. Claude projects, ChatGPT projects, Gemini, Microsoft Copilot - they all have knowledge, instructions, and your initiative lives in a chat thread. I recreated my Riverside copilot to demo what this looks like in Cursor: First thing: instructions. I can create a rule to always apply, no matter what conversation about what file I’m having. This is very similar to what I would put as custom instructions in a project. You go to Cursor settings, and it’s a relatively friendly UI. My project knowledge is markdown files. I’ve created company level context with things about what it means to be a PM, like the process. There’s stuff about me, like feedback I got from my manager. There’s the high level stuff as well: vision, strategy. I’ve also had an AI interview me in the past and ask me questions, so I created files from that. This is some pretty juicy context here. Then I have initiatives. In each initiative folder, I create a markdown file called “thinking.” It’s my scratch pad. I like to think in terms of an opportunity assessment, so my scratch pad is organized according to these headings that eventually become a PRD, but it’s more like how I organize my messy thoughts: what we know, what we should research, solution ideas, risks and question marks to validate. The next step is looping Cursor in using Agent mode. After I share specific context, I ask “Can you please add this to the document, keeping the headings the same and using my original words, but keeping it succinct and just fitting in what we have? And don’t make up anything I didn’t tell you.” It’s in agent mode, so it’s helping me write this document. I actually prefer making changes myself after the conversation, or dictating them in again. Personal preference. But you can have it help with success metrics, create user research plans, make new files. The project knowledge just keeps updating in a very tight loop. ➡️ Build your Cursor copilot with three layers of context in markdown files: company knowledge that never changes, initiative docs that evolve daily, and instructions for how you want to be challenged. Your chat threads become disposable - the real intelligence lives in documents that get smarter with every conversation.