Giwan

korean indie hacker. learning saas the hard way, in public. between bets — figuring out what's next

[Giwan @tvprogramlover]

Everyone was tweeting "Omarchy Omarchy Omarchy."

Thought it was just a skin, but the keyboard workflow hits different. Feels right at home. ⌨️🔥

[Giwan @tvprogramlover]

Free or paid, the review process should go on equally for every developer.

Or maybe this is just my own assumption...

[Giwan @tvprogramlover]

I’m currently uploading free apps on Cafe24 (Korea’s Shopify) to check the platform's potential for distribution and monetization.

At first, the app review was super fast. But then I got a phone call from them asking, "Why publish for free? Are you doing charity work?" 🧵👇

[Giwan @tvprogramlover]

Ever since that call, their app review process has gotten noticeably slower.

Look, I get it—corporations need to make money. As a solo dev, so do I. But changing the review speed or attitude based on whether an app is free or paid just doesn't feel right.

[Giwan @tvprogramlover]

But for big decisions, I still fall for the stereotype that "only a frontier model can answer this." So I asked DeepSeek V4 Flash how to save tokens while getting frontier-quality results—and it gave me a https://t.co/nsDRRixfDE script. Distillation outside of training!

[Giwan @tvprogramlover]

Ever since canceling my Claude Max 5x plan, I’ve been hyper-aware of token costs. Mid-tier models handle most of my daily work like coding, web tools, and blogging just fine.

[Giwan @tvprogramlover]

@FCMJD @jordanurbs However, for elite-tier coding architecture and heavy agent workflows, top-tier models (like Claude / Fable5) remain worth it.

[Giwan @tvprogramlover]

@FCMJD @jordanurbs The Value Verdict:If you're burning through 4x $200 plans, high-efficiency models (like DeepSeek v4 flash) match Opus 4.8 quality in blind tests for automation and SaaS workflows, offering massive cost savings.

[Giwan @tvprogramlover]

@FCMJD @jordanurbs If you are dealing with massive enterprise codebases, Claude is essential—especially for Claude Code (its agent harness, vulnerability checks, and ability to read entire code flows rather than just small chunks are unmatched).

[Giwan @tvprogramlover]

@FCMJD @jordanurbs For indie SaaS and automation (like Chrome CDP for crawling/blogging), frontier models like GPT and Claude handle high-Level tasks effortlessly with minimal direction. Models like DeepSeek and Qwen need more specific prompting, but still get the work done well.

[Giwan @tvprogramlover]

@FCMJD @jordanurbs No, I haven't used it before, but I want to try it next time.

[Giwan @tvprogramlover]

@FCMJD @jordanurbs At Venice rates DeepSeek V4 Flash is $0.14/$0.28 per 1M, so $200 ≈ 1.3B blended tokens/mo. That's the "can't exhaust it" part, and it's real.

[Giwan @tvprogramlover]

Distribution is the hardest part of shipping projects.
Building a community from 0 is unrealistic.
Existing marketplaces are better, but the App Store and Play Store are already overcrowded.
That’s why I’m hunting for new "blue ocean" marketplaces to actually get noticed.

[Giwan @tvprogramlover]

@ctatedev Since dynamic elements (npm packages, runtime objects, etc.) drop back to the embedded QuickJS-ng (reintroducing a JS engine anyway), how do you handle this in a native executable?

[Giwan @tvprogramlover]

Just got my 2nd paying customer for my SaaS! 🚀

It’s only 9,900 KRW, but it means the world to me. It’s a huge sign of hope that I can actually build my own business. 🥹

Onwards and upwards! 💪💻

[Giwan @tvprogramlover]

P.S. You can use the DeepSeek API with zero transaction fees if you pay via Alipay.

[Giwan @tvprogramlover]

Can't justify $100/mo on the Claude Max 5x plan anymore. Switching my stack to Opencode + DeepSeek v4 Flash—the price-to-performance is incredible.

I used to chase AI trends, but honestly, open-weight models are enough for the small projects that cover my living expenses.

[Giwan @tvprogramlover]

@swyx Oh, the first one is my submission! I'm so honored to be featured here.

[Giwan @tvprogramlover]

Given the security issues with AI, this video is a must-watch. I haven't seen it yet, so I need to check it out ASAP!
https://t.co/T3w6l2PxIn