Guillermo Rauch explains that Vercel is developing V, an agent designed to perform real work across a company rather than answer isolated prompts. Its operating model combines reusable skills, approved tools, company context and a persistent identity so the agent can understand responsibilities and act within explicit boundaries.
Riley Brown and Guillermo Rauch discuss why useful enterprise agents must connect to proprietary data and existing systems without receiving unlimited access. Permissions should be scoped to each task, sensitive actions should remain auditable, and the agent should use the same kinds of controlled interfaces that employees use for email, analytics, search and internal applications.
Guillermo Rauch argues that the strongest agents will be proactive as well as conversational. Event triggers and schedules can start recurring work, while evaluations and feedback loops can measure results and refine instructions over time. He also favors model-agnostic infrastructure so companies can choose the best model for each job without rebuilding the surrounding agent system.
Watch on YouTube
