Hyoseo Lee
Home About Projects Blog Thinking
dev-log

lesson from WW-extension implementation

Topic Product Launch & AI Development
Area Solopreneurship
Summary
A reflective dev-log on successfully deploying a multi-component service using modern cloud infrastructure and AI code generation, highlighting the challenges of managing LLM context efficiency and navigating the psychological shift from developer to entrepreneur when pricing a product.

How it has been going

To say the conclusion first, The service is deployed successfully.

it has three big parts now, which are Backend, Website, and Extension.

Extension is just a simple vanila JS, posted on chrome webstore. Website is developed with vite + react + typescript, and deployed by vercel. Backend is fastapi based system, that uses psql database. most part of database is managed by Supabase, including users and Authentication. for Authentication, I use oAuth JWT method, with azure oauth. Lastly, it’s payment system are from stripe, and every design by stitch.

So basically, this project is just combining all trending cloude service and make them work.

And I guess that is all for this project, and that is not what I want to talk about now.

What I learned so far

I can say there are two big area that I learned about. one is about Claude code, and Second is monetization.

Claude code

During this project, I used claude code a lot. 100% of code were generated by it and 99% of documents were written by it.

But I still think I don’t know how to use this tool. This is like treating a person. he works really well, but I don’t know how to make work him most efficiently.

I can feel the difference between sessions and I tried my best to remove the gap between them by documentation.

And the best method that worked is documentizing a lot.

but it should have non-duplicated informations, because while changing, some might left and that makes ambiguous points on documents.

And Another thing I found out is there are specific jobs that deletes the tokens.

actually documentation is not a process that use tokens a lot, design does.

it’s maybe because I was using stitch without figma(I know nothing about figma).

When I use stitch, I just ask him to make design, select all I want and extract to .zip file.

and give the zip file to claude code, and say “use this design sample to this page”. and then claude code starts to scan the file I gave, and generates.

and that process uses a lot of tokens.

Usually, just coding and documenting needs about 3 hours of work to spend all 5h usage. but when I was doing this prcess, I finished by 5h limit less then an hour.

There might be other factors, but It did use so much usages.

I guess that’s all for claude code

Monetization

As I said on earlier blogs, this is the first project that I actually aimed for monetizing.

thanks to stripe, I could implement the payment system really easily. it literally does all job for me.

But I had so much problems related to this monetizing problem.

which is “How much should I value my product?”

This is actually a Entrepreneur’s problem.

And this is more difficult to software products, because the original cost of the production and benefit from the product is really unclear.

they really depends on the scale of the system, area of product, and target customer.

So while choosing price, so many problems come into my head. “does people know the cost of gemini api?”, “how much cost will be optimal?”, “Can I change the price later?”, “What if they feel some betrayal for price change?”, “what if they think I’m evil company?”, “How much user and net balance I’m expecting?”, “How much money I want for a month?”, “will people actually pay for my service?”, “Is 100 sign in bonus credit too much?” and tons more.

There are no correct answer, or not even a standard. I cannot compare this to other ai services.

currently the credit spend is depends on the actual gemini-api token usage. so It is impossible for losing money.

I also thought about monthly subscription business model, but I thought that doesn’t fit well to coop terms.

I don’t know actually. what I wanted to say was there are whole new world of concerns that I never experienced.

Adding monetization is not just adding feature to the system, it is a huge jump from developer to actual entrepreneur.

So what

I think I talked pretty unorganized today but that is actually what’s going on my head.

I just hope I learn a lot from this, and earn some poket money form this project ^^