My experience: Claw's sales demo was magic. The pilot was a series of workarounds.

Just got off the pilot with Claw AI. The sales demo was pure magic—it watched my screen, understood my workflow, and auto-generated perfect Zapier tasks. I was ready to throw money at them.

The actual pilot? A different story. The agent kept getting stuck on permissions, misinterpreted simple commands, and required so much manual correction that I was basically building the automations myself. The "AI" felt like a fragile wrapper around a bunch of scripts I could've written in Make.

Has anyone else hit this wall? The gap between the polished demo and the real, messy workday use is huge right now. I'm still hopeful, but it's back to manual Zapier builds for now.

dk