Vibe coding without understanding your code won’t bring you anything. A few days ago, I deployed a web app that links price movements on @Polymarket with real-time news and displays them directly on the chart: polymarketnews(dot)com. I was facing an excessive token consumption on OpenAI and on the API I use to fetch real-time news. I asked Claude many times, but it couldn’t identify the issue. By digging into the code myself, I realized that instead of storing articles that had already been fetched, the code was calling the APIs again every time to re-fetch the same articles and re-analyze them with OpenAI. So I made a few changes, now the code only fetches articles within the most recent time window (last 5 minutes), analyzes only these new articles with OpenAI, and then merges them with the existing ones. Result: a 50× reduction in API usage. Now I pay much less for OpenAI, and I can even reduce the search interval to get closer to real-time news. Costs before: > ~$500/month for the news API > ~$10/day for OpenAI Costs now: > ~$500/month (same cost, but closer to real-time fetching) > ~$0.5/day Honestly, I’m starting to wonder if vibe coding isn’t designed to increase your API costs just to boost their own revenue hahaa. Anyway, the key takeaway is you still need real technical skills, combined with vibe coding, to reduce time and effort by 99%.