Vercel's Andrew Qu on why agents are a new kind of software

In a discussion with Latent Space, Vercel's Chief of Software, Andrew Qu, offered his perspective on why AI agents should be viewed as a fundamentally new type of software. He explained that building effective agents requires rethinking traditional abstractions around skills, sandboxed environments, and making websites agent-readable. Qu detailed how Vercel's internal agent framework, Eve, was developed to address these needs, emphasizing that agents operate differently from conventional applications—they must be able to browse, reason, and act within constrained, secure contexts. For developers and solopreneurs building AI workflows, this suggests that designing for agent interaction involves more than just API access; it requires creating structured, interpretable interfaces and safe execution spaces. The practical angle for builders is to consider how their own software can be made agent-friendly: by exposing clear skill interfaces, enabling sandboxed testing, and ensuring web content is parsable by language models. This shift could redefine best practices for AI-integrated products.