FlashyCardy

• 01

  Sign in and protected pages with Clerk.

  Logged in users go straight to their dashboard.

• 02

  A database for decks and cards.

  Each user’s data is kept separate.

• 03

  A freemium model.

  Free accounts get three decks, with a warning at two and an upgrade prompt at three. Pro is unlimited, billed through Clerk.

• 04

  Automatic deploys.

  Pushing to main ships to production.

• 01

  The build failed with an error that meant nothing.

  The real cause was a version clash: Clerk didn’t support Next.js 16 yet. I ruled out four other causes before finding it, and the fix was downgrading to Next.js 15.

• 02

  An AI agent broke 14 files at once.

  I asked Cursor to fix one lint error. It fixed it, then rewrote 14 files in the same commit. The build passed, but the app crashed on every page.

• 03

  The fix was live but nothing changed.

  Vercel kept serving an old deployment from cache. Understanding why took longer than the fix itself.

• 01

  Read the dependency versions, not just the error.

  Checking Clerk’s supported versions against Next.js release notes found the clash.

• 02

  Used git history as a debugger.

  I found the breaking commit, rolled back to the last working version, and restored only the files that were still good.

• 03

  Read runtime logs, not build logs.

  Build logs show what failed to compile. Runtime logs show what crashed for real users, down to the file and line.

• Give AI agents small, specific tasks and review every change. Broad instructions produce broad damage.
• One error message can have many causes. Treat it as a clue, not an answer.
• Version compatibility is infrastructure. I now check it before every upgrade.
• Building changes how I design. Once you’ve debugged the thing you designed, you design with what’s hard to build in mind.

AI generated decks for Pro users. You give it a topic, text or a link, and the OpenAI API returns ready made cards. Also on the list: sending users back to their dashboard after payment.