I built a Framer template using only AI agents. It taught me everything I already knew.
The first shot was genuinely impressive. Claude Fable one-shotted the foundation — clean layout, real structure, looked ~80% done in minutes. I sat there thinking I'd be shipping by the end of the day.
Then I let it build the custom components, and everything got familiar.
I got lost fighting its design choices — re-explaining taste it couldn't see. Somewhere in the middle Fable got switched off, which definitely wasn't helping. Though maybe I'm just looking for something to blame.
Small changes took forever. The last 20% ate every minute the first 80% saved.
The takeaway is the one I already knew going in: Framer 3.0 didn't change how AI works. A new tool doesn't close the old gap — agents get you to a fast demo, but taste and the last mile are still yours.

