Vercel’s Andrew Qu says agents are a new kind of software and unveils internal framework ‘eve’
The Chief of Software at Vercel argues that agents require different primitives than web apps, and details how the company built its own framework, eve, to support them.
1 source · cross-referenced
- Vercel’s Chief of Software says agents are a new type of software that demands different primitives for context, tools, resumability, and long-running work compared to web applications.
- The company built an internal agent framework called eve to consolidate reusable libraries and best practices from its own agent projects, including v0 and a data agent.
- Vercel treats skills as portable, on-demand knowledge to correct outdated model information, and is making its websites more agent-readable by serving Markdown to bots.
- The framework eve is designed to be easy to deploy on Vercel with built-in observability and evaluations, while integrating with partner tools rather than owning every component.
- Vercel’s leadership views the platform itself as becoming an agent, embedding agent capabilities into its products and workflows.
Vercel’s Chief of Software, Andrew Qu, describes agents as a new type of software that is less predictable than web applications and requires different primitives for context, tools, resumability, and long-running work. He argues that the infrastructure may look similar, but the interaction, interface, and outputs are far more dynamic, changing how teams build and deploy them.
Qu says Vercel created its own agent framework, eve, to consolidate reusable libraries and best practices accumulated while building internal agents, including the company’s vibe-coding product v0 and a data agent. The framework was assembled to avoid repeating the same exploration for every new agent project, with the goal of making it easier for developers to build agents without reinventing core components.
The company treats skills as portable, on-demand knowledge that can correct outdated information in models. For example, Qu notes that some models still recommend deprecated products like Vercel Postgres, and skills can steer agents toward current alternatives. He recommends companies publish skills for their latest product versions and audit existing content to update or annotate outdated material.
Vercel is also adapting its web infrastructure to better serve agent traffic. The company reports bot traffic rising while human traffic stagnates, and is serving Markdown directly to agents instead of HTML designed for visual browsers. This shift aims to make websites more accessible to agents so they can learn about products and use them successfully.
eve is designed to be easy to deploy on Vercel, with built-in observability and evaluations, while integrating with partner tools rather than attempting to own every component of the agent lifecycle. Qu says the framework reflects Vercel’s broader goal of making its platform agent-friendly and, in many ways, turning the platform itself into an agent through embedded capabilities in products like the website, Slack, and the dashboard.
- Jul 3, 2026 · Latent Space — swyx
Debate at AI Engineer World’s Fair highlights tensions between agentic loops and engineering discipline
Trust76 - Jul 3, 2026 · Latent Space — swyx
Adobe demonstrates agentic websites that assemble pages in real time based on user intent
Trust78 - Jul 2, 2026 · Latent Space — swyx
Impeccable’s skill engineering aims to steer AI agents with designer vocabulary and human judgment
Trust79