Vercel CEO Guillermo Rauch discussed the evolution of AI agents from prototyping to production, highlighting two killer use cases: coding agents and internal corporate agents. He noted that last year was about experimentation, but the focus has shifted to practical challenges like secure data access, auditing, and creating trails of agent activity. To address these, Vercel developed frameworks like Eve, which allows agents to be instructed in natural language, and Vercel Sandbox, which cages agents to enforce data policies. Rauch emphasized the risk of data leakage, citing a conversation with Airbus about the danger of coding IDEs training on proprietary codebases. He described how internal agents, such as one used by a Vercel sales rep, can overcome data bottlenecks that limit employee productivity. The interview was conducted after Vercel's ShipNYC conference, where the company discussed its role in AI infrastructure and competition with major labs.
Actualites avec Curation IA