Prologue: BYO(build-your-own) Tutor
The digital world is moving forward at breakneck speed. So fast even, that it's hard to be sure if you know what you think you've learned, and harder still to get clarification through the fog of progress. I found myself wanting to learn about so many things without the time or ability to stop and really conceptualize what it was I was learning at all without missing the next lesson. This is when I decided to use modern tools to solve my modern problem.
With a little help from a friend
I'd recently installed Claude in my terminal and had tried a few interesting things. For want of a knowledgeable, Socratic tutor to help me find my own answers, I built an agentic facsimile of a tutoring space in the Library of Alexandria with Socrates (as Claude knows Socrates to be, at least) and, though there were notable breakthroughs in that process, I felt limited, needing to pull up a new agent and prepare myself for a deep dive when sometimes a few nudges in the right direction would do the job (and use far fewer tokens). The question became: How can I find a way to study these things while utilizing the expertise of a post-deep-search Claude without needing to do it sitting at my screen. How can I get these concepts delivered to me in a way that I can understand them inside and out, in a way that puts me in the positions to ask the best questions when there are the questions to be asked. Then it struck me like a low-hanging bar: You're using an LLM, dummy. Just have it write you a paper.
The Learner's Three
Of course an artifact was the solution, but what voice would give me the the concept in a way that would allow me to ingest it from a position of ignorance, then mold that concept into something I can work in, all while learning what best practices and common pitfalls could be in the implementation phase? What role should Claude take to write to me in a tone and candor that would keep me from feeling like I was reading a textbook.
In this case, I found I was thinking backwards. I didn't need to configure a role for Claude to write for me, instead, I could take a role while he gets to be the one with the answers. Not a role -(I swear this em-dash is organic)- three roles; one of a learner, one of an educated enthusiast, and one in the field.
It was there that the Learner's Three was born
The Chronicles Begin
So here's the flow:
Claude runs a deep search on whatever the subject may be. All rules of vetting sources, etc, apply here to get the best data outcomes. When he's done, I have him write me three papers explaining the subject:
One as if I were an extremely intelligent 10-yr old.
One as if I were an Undergrad student, studying this for my degree.
One as if I were a colleague at work that has just made a catastrophic error regarding this exact subject.
I chose the 10-yr old because that's around the youngest person I can think of explaining complex agentic processes to without losing sizable chunks of the subject matter. I used extremely smart so that Claude wouldn't think to dumb things down too much. Extremely smart is vague enough to err on the side of expecting too much (if I were actually a 10 yr old), so, with this artifact, I should conceptually be given the broad strokes and prepped to understand the subject at hand in depth.
The undergrad artifact should be one that gets very conceptual, very heady. The one that tells me everything I needed to learn and more, maybe even touches on concepts and implementations that are far past the scope I was interested in, which is perfect. Too much information means definitely enough information, especially when it comes to artifacts. Plus, if I want to geek-out on whatever the subject may be, there's space there to go wild.
The final artifact denotes practical experience: People can explain things in the most direct and pragmatic way when you're screwing up things that they need to do their jobs. No time for hand-holding, BS, or cute analogies, they need you to know what you need to know to NEVER DO THAT AGAIN!! And that's the kind of practical knowledge that you'd absolutely want before actual implementation in whatever subject you're trying to grasp.
Onward and Upward
I've run this workflow for a few fun subjects, and I believe I've gotten good insights every time, but that's just my opinion. I'll run one by you, so you can judge for yourselves. Let's go meta right off the digital bat with this one and start with Workflow Engineering, given that's what I've been most interested in throughout this blog.
We'll take this step by step and analyze each paper as they came out(child, student, colleague), each alongside its respective artifact, so, if you like, you could just take the artifact and go home, but you may as well sit for a while while you're around. Either way, I hope you enjoy these artifacts, and the journey that brought them here.
Next stop: Workflow Engineering For Kiddos!
Top comments (0)