How We Solved Agent Building — Andrew Qu, Vercel

Building effective AI agents requires moving beyond simple "mega prompts" toward a file-system-based architecture that allows agents to manage their own state, tools, and execution paths. Vercel’s internal development journey—from a rudimentary data science assistant to a sophisticated, scalable framework—demonstrates that agents perform best when they possess deep, company-specific knowledge and access to a secure sandbox environment. The introduction of EVE, a framework inspired by Next.js, simplifies this process by providing a convention-based structure for defining skills, tools, and channels. By leveraging existing infrastructure like sandboxes and workflows, organizations can automate complex, vertical-specific tasks, such as data querying or contract negotiation, significantly boosting productivity. This evolution confirms that the most successful agents are not generic, off-the-shelf solutions but purpose-built systems integrated directly into an organization’s unique operational context.