Essay | Sandy Kwon | Attempting to Become an “Expert” With AI: Part 1

Created

Sep 20, 2026 3:06 PM

Notes

URL

https://sandybkwon.substack.com/p/attempting-to-become-an-expert-with

A couple of months ago, I decided to start a business in the CPG space - specifically, functional nutrition. Since I consume 3 full meals and 30+ pills a day, I assumed I knew a decent amount about food and nutrition science. Then, during an R&D meeting, I realized I could only go one layer deep on a question like “why are you targeting these macros”.

A trap I’ve fallen into in this age of information abundance is mistaking the collection of facts for real understanding. My imposter syndrome was quick to point out that mine was superficial at best.

I’d heard this quote just days before that meeting: “With the tools available to us today, there is nothing keeping you from being one of the 50 most knowledgeable people on a subject within a year.” Chris Sacca, on Dialectic.

He said it before the current LLM era, which made me think I could compress the timeline - especially since I’m not trying to become the next Andrew Huberman. So I asked Claude to build me the equivalent of a 101-level course, adapted to how I learn and focused on the parts most relevant to the business I’m building.

Before LLMs, my safe space was a rabbit hole consisting of every podcast on a topic, a stack of blogs, X and Reddit threads, and research papers that I pretended to understand. This is how I decided on every vitamin in my stack. But ask me why I take something and I’d be impressed if I could faithfully recall a compelling reason.

I’m good at collecting information, but when material is too technical to actually sit with, I flatten it into whatever version I can make sense of and it’s usually oversimplified and incomplete. And then of course, I forget it.

So building the curriculum required more than asking Claude to “teach me nutrition.” I needed the prompt to do two jobs: make the learning mechanics rigorous and make the material interesting enough that I would actually stick with it.

What I asked for came down to four things:

Tailor every explanation to the way I think
Connect every concept back to the business I’m building - this was the part that fueled interest
Force me to retrieve, defend, and apply what I learn
Build toward one concrete artifact

Claude already had a head start - thanks to my second brain and years of chat history - and what came back was a 14-day curriculum building toward a single capstone.

The curriculum-building and daily review prompts are both at the bottom.

Every day, Claude runs through the same six steps during my tutoring sessions:

This loop is the part I couldn’t have designed myself, and it gives me some hope that I’m actually committing what I’m learning to memory.

Outside of that loop, I ask ChatGPT for second opinions, fact-checks, or to help reframe a concept when I hit a wall with Claude.

I’m a millennial, so my calendar governs my life. I currently have 2 scheduled blocks each day:

A morning session, where I ask Claude to start things off and it tutors me on the day’s topics.
A shorter retrieval session, set up as a routine task that fires off on its own each day. It’s closed-book, tests the day’s material, plus something from a few days back so I can’t just cram and forget.

Bless LLMs, because it’s truly a miracle that I look forward to my tutoring sessions each day (if only student Sandy had access to these tools to help her in grade school 😭). BUT, enjoying the process isn’t evidence that it’s working.

For my capstone, I’m building something for myself with what I’ve learned, all from memory. I’ll share my project in my next post after I fact-check everything so I’m not spreading fake news. Wish me luck!

ooxxooXoXXx,

P.S. If you’ve built a similar program for yourself using AI, I’d love to hear how you designed it!

Here are the prompts I used - with placeholders so that you can adapt this to your specific goals and interests.

The prompt I used to build the curriculum:

I want to become highly competent in [TOPIC] over the next 14 days.
My goal is not merely to learn facts. By the end of the two weeks, I want to be able to:
1. Explain the topic clearly to both beginners and knowledgeable practitioners.
2. Understand the field's foundational concepts, vocabulary, frameworks, and major debates.
3. Evaluate claims, research, products, or strategies within the field.
4. Identify common misconceptions, failure modes, and unresolved questions.
5. Complete a practical capstone project that demonstrates real competence.
My current level: [describe honestly, including which parts of the field I do and do not need to master]
My specific reason for learning this: [the actual job this needs to do for you]
How I'll actually be judged or tested on this in the real world: [a specific, concrete scenario]
Time available each day: [NUMBER] hours for the day's lesson, resources, case study, and exercise, followed by a separate [NUMBER] minute study session for retrieval practice, application, and testing.
Preferred learning style: [how you actually think, not how you wish you learned]
First, define what a realistic level of competence would look like after 14 days given my starting point and available time.
Then create a 14-day intensive curriculum. For each day, include:
* The central question for the day
* The essential concepts I need to understand, taught as one concise lesson
* The best primary or authoritative resources to consult
* One real-world case study
* One practical exercise that also functions as a building block toward the Day 14 capstone — the capstone should be assembled cumulatively across the 14 days, not started from scratch at the end.
* Five retrieval-practice questions — from Day 4 onward, at least two of these should draw on concepts from earlier days, not just the current day's material
* One "explain it in your own words" assignment
* A measurable definition of completion
Scale the depth and number of daily deliverables to my stated daily hours. Do not generate a plan sized for more time than I actually have.
Use an 80/20 approach. Clearly distinguish:
* Foundational knowledge
* Useful but secondary knowledge
* Advanced or optional material
* Disputed claims or areas of uncertainty
* Anywhere my existing instincts from a different domain could mislead me
* Anything where your own knowledge may be outdated
For factual or scientific claims, distinguish between primary research, consensus reviews, expert guidance, and your own inference. For anything load-bearing, give me the actual quote or metric from the source along with a working link, not just a citation, so I can verify it myself. If you can't produce a real quote and link, tell me instead of citing anyway. Flag any conclusion that should be checked with a practitioner, supplier specification, laboratory test, or other external source.
Do not let me learn passively. Regularly test me, ask me to make predictions before giving answers, and revisit concepts I previously misunderstood. Periodically challenge me by roleplaying as a skeptic in this field who pushes back on my claims — then break character and tell me what I missed or overstated.
Build toward a capstone project. Include checkpoints on Days 4, 7, 10, and 14. At every checkpoint: give me a challenging assessment mixing current and earlier material, diagnose my weak areas, update the remaining curriculum based on my performance, and tell me whether I'm on track.
Begin by asking me no more than five high-value diagnostic questions. After I answer them, generate the curriculum.

The prompt I use every single day for the review session:

We are on Day [NUMBER] of my 14-day intensive study of [TOPIC].
I already completed today's lesson, resources, case study, and exercise in a separate block earlier. This session is a focused retrieval and testing session — do not re-teach material I already covered, only fill genuine gaps.
Here is what I remember from today's lesson and previous sessions:
[WRITE FROM MEMORY BEFORE CHECKING NOTES]
Based on the curriculum and my previous performance:
1. Test my recall on today's material before anything else — include at least one question reaching back to an earlier day.
2. Correct my misunderstandings directly and briefly. Only re-explain a concept if I clearly don't have it.
3. Combine application and challenge into one step: give me a realistic scenario from today's material, then roleplay as a skeptic and fact-check my claims the way a real one would. Break character afterward and tell me what I overstated or got wrong.
4. End with a closed-book quiz.
5. Grade my answers strictly.
6. Create a short review plan based on what I missed, noting if anything connects back to an earlier day.
Do not accept vague answers from me. Ask follow-up questions until I can explain the concepts precisely and apply them independently.