Invest into context

I have a 14,000-word document I maintain manually, to add into ChatGPT queries I make related to my role as CTO. It covers a lot of ground about Pactum:

• An extremely basic description of who we are and what we do
• Company strategy, long-term goals, financial model, and progress towards those
• Our target market, ICP, personas, value propositions
• High-level documentation of our products
• Org structure of the whole company (at a high level) and of Product/Engineering (granularly)
• Engineering metrics, architecture, principles, hiring structure, etc

That's only half of it. The other half describes myself:

• My goals and how they relate to company goals
• Things I've learned about myself, what I am like, and how I like to work
• Feedback and recommendations I've received, and how I strive to improve
• Results from personality tests I've taken (specifically, Gallup Strengths)

This might seem like a pretty insane amount of sensitive information to give into an LLM. But assuming you're compliant and trust the vendor as much as Google, or it is in fact Google, privacy is not a big issue (and none of the chat apps I use train on my data). So other than normal vendor compliance, I don't have to worry about this too much.

The value, on the other hand, is massive. Since I've pre-written a huge amount of context, I rarely have to explain much to an LLM to have a useful chat. Writing a single sentence, and copy-pasting this context file, launches me into a conversation that is probably lower-context than I'd have with a colleague, but beats any discussion I could have with a friend or coach. And of course, LLMs are different conversation partners: available 24/7 and easily steerable.

I rarely use specific text or even advice from these chats. Rather, it is a tool to iterate my own thinking on a tough subject (that by definition requires lots of context).

Here are real examples from my usage, which I would never do if I had to write out the full context every time:

• How do we grow usage of our product?
• I am struggling to understand what to expect of $ROLE. Help me think about that.
• I want to make $CHANGE, but not in a top-down mandated way. How do I involve people in the process, so we determine the direction together? Given what I am like, what am I likely to miss?
• How do companies like ours measure product performance?

To be clear, it is currently completely static, does no RAG or searching Confluence or anything like that. I have curated what I think is basic and important information myself -- and this level of understanding is often just assumed at a company, so not even written down.

I use it daily, sometimes in combination with deep research.

I've probably spent a total of two hours building the file by this point, mostly by extracting text from pre-existing documents using an LLM, and sometimes manually adding context where necessary. I've asked an LLM to tell me what the gaps are, so I can proactively add more relevant information. Recently I created some very light file structure and scripts to automate bits and pieces of it.

If you build your own, I'd love to hear how it goes for you.