Vercel's announcement, while framed as a simplification, arguably introduces a new layer of cost complexity for developers. The move to token-based pricing, while common for raw model inference, becomes murkier when a platform layer adds its own token fee on top. This could make total cost forecasting difficult, as users must now track two separate token rates—Vercel's and the underlying provider's—for a single task.
The change seems to signal that AI agent platforms, having moved past the initial novelty phase, are now grappling with sustainable unit economics, potentially passing infrastructure and orchestration costs more directly to the end user.