Followloop routes AI tasks to cheaper models

The core idea is solid: most everyday AI work does not need a frontier model, and hiding that routing behind a simple workflow is where the value lives.

• –The savings claim is interesting, but it is still self-reported usage data, so the important question is failure rate when the router misclassifies a task
• –The MCP angle matters because it targets existing Claude Desktop, Cursor, and Claude Code users without forcing a workflow change
• –The safety-screened MCP library may be the stronger moat than routing alone, since trust and discovery are still messy in the MCP ecosystem
• –This looks more like practical cost infrastructure than model innovation, which is exactly the kind of product AI teams actually adopt
• –It competes more on convenience for everyday users than on raw routing sophistication, which is a sensible wedge