🚀 Just shipped an open-source template for AI services.
✅ Includes full auth and per-request payments.
✨ Automatically charges cost + margin.
🤖 Demo uses OpenAI API to mimic ChatGPT 👇🏻
I agree.
Unless you're making throw-away prototypes, AI coding productivity boost is limited by your ability to properly review code.
Otherwise, you're shooting yourself in the foot in the long-term.
@wesbos You can check for nested deps of your own repos with:
npm explain plain-crypto-js
pnpm why plain-crypto-js
yarn why plain-crypto-js
bun why plain-crypto-js
Or you can do a raw search across your repos:
grep -rI "plain-crypto-js" <parent dir>
@YifanBTH Use this command if you don't have rg installed:
grep -rI "plain-crypto-js" <dir>
These days, I mostly use a thin wrapper around the native fetch() rather than axios.
@YifanBTH I think there's an optimal amount of "change complexity" that an AI should make before a human reviews it. That way, the human can provide just enough guidance to take a good long-term path.
It feels like we are all team managers now.
Team members are different AI models.
Each has its own pros, normally speed vs quality.
And you can easily "hire" more.
I’m probably at peak AI FOMO right now. Subscription history over the past 4 months:
chatgpt plus + claude max
|
chatgpt pro ⬆️ + claude pro ⬇️
|
chatgpt pro + gemini pro ⬆️ + claude pro
|
chatgpt plus ⬇️ + gemini pro + claude max ⬆️
Someone please invent a way to make my FOMO life easier.
Lesson learnt.
Don't use Claude Code for research. It will eat up all the tokens, which are much better used for coding.
Use ChatGPT 5.1 "Research" mode instead.
We already have battle-tested components and frameworks we can stack together.
I think that the bigger change that's coming is actually the opposite: there will be more and more AI generated glue, which is _not_ as battle-tested, so the key is creating a development framework to keep long-tern development velocity high and product reliability good despite the massive increase in AI-generated code.