Vercel CEO: Agents split from models as two killer apps emerge

Vercel CEO Guillermo Rauch outlines a strategic shift in AI: decoupling models from agents. With 6 million daily deployments (half triggered by coding agents) and over 1 trillion tokens flowing through its AI gateway, Vercel has become a central platform for agent infrastructure. Rauch identifies two killer use cases: coding agents that generate massive volumes of software, and internal corporate agents that transform how companies access and act on data. To enable the latter, Vercel launched Eve, a framework where agents' instructions and skills are defined in natural language, and Vercel Sandbox, which cages agents while letting them apply intelligence. The sandbox prevents sensitive data (like proprietary codebases) from being used for model training. Rauch cites a conversation with Airbus about the risk of an IDE leaking decades of aerospace C++ code.

Rauch argues that agents force companies to open up their data, which is incompatible with traditional SaaS models that trap data. He sees a move away from single-lab lock-in (building everything on OpenAI or Anthropic) toward a plug-and-play ecosystem where every layer (model, harness, data platform, sandbox, gateway) is interchangeable. He notes surprising growth for Gemini despite less media attention. For internal agents, Rauch gives an example: a sales rep can ask, "Which accounts added the most seats in the last two weeks?" without waiting for a custom dashboard. Eve brings that natural-language querying to company data, turning APIs into productivity tools. The long-term implication: agents will dissolve data silos, forcing SaaS giants to evolve or be bypassed.