I had an idea Friday night for a tool that tracks when subscription prices change. Simple concept — you paste in a URL, it monitors the pricing page, and texts you if the price drops or a new plan appears.
By Sunday afternoon it was live. Real users. Actual texts going out. Not a prototype. A working thing.
Two years ago that's a two-week project minimum. Here's what changed.
The actual timeline
Friday night (2 hours): I described what I wanted to an AI coding assistant and had it scaffold the whole project. Next.js app, a cron job for scraping, a Twilio integration for texts. The scaffolding was maybe 80% right. I spent the rest of the time fixing the parts it got wrong.
Saturday morning (3 hours): Built the actual scraping logic. This is where AI was most useful — I described the page structure and it wrote parsers that handled different pricing page layouts. I had to test and adjust, but it got me to "working" much faster than writing parsers from scratch.
Saturday afternoon (2 hours): Auth, database, deployment. Supabase for the backend, Vercel for hosting. The AI helped write the database schema and the auth flow. Deployed by dinner.
Sunday (3 hours): Testing, bug fixes, making it not look terrible. Sent it to friends. Fixed the bugs they found.
What AI was good at
Scaffolding. Boilerplate. Repetitive patterns. Writing the fifteenth API route that follows the same pattern as the other fourteen. Converting my plain-English descriptions into working code that was close enough to correct.
What AI was bad at
Architecture decisions. It'll build whatever you ask for, but it won't tell you if your approach is wrong. I had to catch my own bad ideas. It also struggled with the tricky edge cases in web scraping — sites that load prices dynamically, A/B tested pricing pages, regional pricing.
The takeaway
AI coding tools don't replace knowing how to code. They replace the slow parts of coding. The boilerplate, the syntax lookup, the "I know what I want to do but typing it out takes forever" parts. If you know what you're building and why, these tools genuinely cut the timeline in half or more.
If you don't know what you're building, they'll just help you build the wrong thing faster.