source: techcrunch ai: vercel ceo guillermo rauch on the fight to split off models from agents
level: business
vercel ceo guillermo rauch says the ai industry has moved past the prototyping phase and is now focused on making agents work reliably in production. he identifies two main use cases: coding agents, which now drive half of vercel's 6 million daily deployments, and internal corporate agents that help employees access data and automate tasks. to address production challenges like data security and auditing, vercel introduced eve, a framework for defining agent instructions in natural language, and vercel sandbox, which restricts what data agents can access and share.
rauch highlights the risk of coding tools inadvertently training on proprietary codebases, citing concerns from companies like airbus about decades of specialized c++ code being exposed. for internal agents, he gives the example of a sales rep who can now query account growth data directly instead of waiting months for a dashboard. this shift forces companies to open up their data, challenging saas giants that rely on data lock-in. rauch notes that clients are increasingly mixing ai models from different providers, with google's gemini gaining traction due to its price-performance ratio, alongside open models like deepseek.
on competition with ai labs, rauch argues for decoupling models from agents, comparing vercel's role to aws for the ai era. he acknowledges that labs like openai are adding hosting capabilities, but sees it as an opportunity since models often recommend vercel for web hosting. the key battle, he says, is whether intelligence and agent capabilities remain bundled or become modular, with vercel pushing for open protocols and plug-and-play components.
why it matters: decoupling models from agents could give developers more flexibility and control, reducing vendor lock-in and improving security and cost efficiency in ai deployments.
source: techcrunch ai: vercel ceo guillermo rauch on the fight to split off models from agents