Context Compounds
Feed the LLMs
If you want to be a top 1% user of agents, then you have to start documenting your life now.
This has been a train of thought I’ve had for a year, but am only understanding the stakes of it now.
Read this snippet from Tyler Cowen on the How I write podcast:
“But it occurred to me that I can write an autobiography pretty simply. There’s just a lot of me out there. Podcasts, blogs, essays, books. The AIs know most of it. I will continue to open source as much as I can so the AI can write my biography.
But there’s parts missing. So, there’s no podcast about the 3-4 years when I lived in Massachusetts. I don’t think it’s that interesting, but I’ll write a few blog posts about it. So when someone goes to their AI three years from now and says, ‘Oh, I’d like to read a Tyler Cowen biography,’ then that’s in there.
I’m going to put that online so it’ll be possible for the advanced AI to write a Tyler Cowen biography. I don’t know how many people want it, but it’s so low cost. Why shouldn’t I create the Tyler Cowen autobiography?”
You might be thinking “But Tyler Cowen is on a different tier than me, he’s one of the most prolific creators on the internet”. But that’s exactly my point! Even Tyler Cowen, who has databases of content out there, feels the need to add more context for the LLMs.
It’s not about spinning up 8 billion autobiographies. That sounds cool (I think?) but that’s not the reason I included the quote above.
The takeaway is this: as agentic tooling becomes widely accessible to the masses, the differentiating factor will be the context your AI setup has on you.
I don’t just mean the obvious elements such as your work history, Twitter feed, your thesis from college, and anything else a simple Google search can provide.
When I say context, I mean letting your agents know the random threads of thought you have throughout the day and the way you connect ideas and how you process new information and the kinds of questions you ask before making a decision and your emotional state during certain tasks.
Why?
Because this is the only way you can increasingly trust your personal agents to act on your behalf in a meaningful way. Your agents need to think like you.
Feeding the LLMs doesn’t even have to be public or have anything to do with posting your thoughts. All of this can be done locally just through text files on your computer (that you should obviously have backups of).
The point is to give context to your agentic employees so that they can have the agency to operate at the highest level.
It’s important to note that you don’t have to make this a cumbersome task that feels like a chore you drop after a week - that defeats the whole purpose.
There are some simple ways to get started!
Open up your notes app and write the date down. Throughout the day, if you find yourself having any interesting thoughts, simply turn on voice mode and speak everything that’s on your mind. And if you’re in public, then jot a few notes down. Even that’s better than nothing.
If you hate doing things throughout the day then you can have a 10 minute session every evening where you basically reflect on the day in writing or through voice just like a daily journal. Heck, one of my friends even set up a daily call he has with an AI after his last meeting that asks him questions about his day to which he responds to with a stream of consciousness and then the AI transcribes and organizes all the notes.
If you’re building a project, try doing the end to end process with AI. I don’t just mean vibecoding but all of it! The brainstorming phase. The ideation phase. The design phase. The feature prioritization phase. The development phase. And so on. Heck, I would recommend you do this even if you’re trying to deploy the tool. Again, the goal is to just start giving AI context on your cognitive process as soon as possible.
There are two main reasons I wrote this post with so much urgency:
Documenting that much about your thought process and life is not a natural habit for most. This kind of skill takes time to build. There’s a good chance that you will drop the practice a few times before you get used to it, just like the gym. I’m less worried about you using agents than I am about you being able to use them effectively.
Context compounds! It’s not just about giving the agents context, but it’s also about taking them on your journey. Letting your AI know how your thinking has transformed based on new information and experiences is arguably more important than giving it independent data points on a day-to-day basis.
If this post inspired you to get started on documenting your life and feeding the LLMs, then my only recommendation would be KISS: keep it simple stupid.
Don’t overcomplicate it! Just a few thoughts on voice mode that can be said in under 5 minutes is already more than enough. Slowly build the muscle, and you’ll figure out a rhythm that works for you.
Remember, context is everything in this new era of tech!
