Vercel says agents now drive millions of daily deployments as it pushes Eve and Sandbox to protect enterprise data
Vercel reports millions of agent-driven deployments and over 1 trillion tokens daily through its AI gateway while introducing Eve and Sandbox to safeguard corporate data.
Vercel has told investors and customers it now processes roughly 6 million deployments a day, with roughly half triggered by coding and automation agents, and that more than a trillion tokens traverse its AI gateway every day. The company unveiled new internal controls, led by a framework called Eve and a containment tool named Sandbox, aimed at keeping sensitive data from leaking during agent operations. CEO Guillermo Rauch made the remarks following the company’s recent ShipNYC conference, framing the tools as responses to production realities that emerged after a year of intensive prototyping.
Vercel reports agent-driven surge in deployments
Vercel’s platform, originally built to let developers deploy applications without managing servers, has pivoted toward agent orchestration as a central use case. The company says coding agents account for a large share of token consumption and deployment activity on its infrastructure.
That shift has pushed Vercel into a more prominent role in the AI stack, sitting between model providers and developers’ production environments. The company’s metrics indicate that agent activity is already reshaping how software is built and hosted at scale.
Coding agents and internal agents emerge as primary drivers
Executives describe two primary “killer apps” for agents: developer-facing coding assistants and internal corporate agents that surface and synthesize enterprise data. Coding agents accelerate software creation and produce heavy token traffic, while internal agents are focused on productivity gains across functions like sales and operations.
But running internal agents in production introduces requirements beyond raw model performance, including secure data access, audit trails, and tool-call logging. Those operational needs are what Vercel says motivated its development of new governance and containment features.
Eve framework and Sandbox aimed at data control
Vercel’s Eve framework lets organizations express an agent’s instructions and capabilities in natural language so policies and permissions can be applied consistently. The company positions Eve as a way to formalize an agent’s behavior without sacrificing developer agility.
Sandbox is a companion control that isolates agents in a constrained execution environment, permitting useful interactions while limiting what data can leave or be accessed. Vercel frames the two tools as complementary: Eve defines intent and skillsets, and Sandbox enforces boundaries around data flows and external integrations.
Enterprise sales agent example at Vercel
Vercel described an internal example where a sales representative used an agent to prioritize accounts by recent seat growth, a task that previously required bespoke dashboards or delayed engineering projects. That kind of targeted query is emblematic of how internal agents can remove bottlenecks and put analytics directly into the hands of line workers.
Executives say the same agent technology can power both customer-facing features and internal productivity tools, delivered through consistent APIs. The result, they argue, is faster decision-making and reduced reliance on centralized reporting projects.
Clients diversify across OpenAI, Anthropic, Gemini and open models
On the model side, Vercel reports customers no longer feel compelled to lock into a single laboratory or provider. Teams now mix proprietary models such as OpenAI and Anthropic with emerging choices like Gemini and several open-source alternatives to optimize for price and performance.
This pluralism, Vercel says, reflects production priorities: latency, cost per token, and consistent integration matter more than brand alone once workflows reach scale. The company also highlights growing adoption of open models such as GLM-5.2 and others that meet specific operational requirements.
Platform competition and the push for open infrastructure
As model providers expand features—some now publishing small web-hosted tools directly—Vercel finds itself both cooperating with and competing against large AI labs. The firm frames its strategy as providing modular building blocks so customers can mix models, data platforms, and governance layers rather than being locked into a single vendor.
Vercel is positioning its stack as the infrastructure layer for agent-powered applications, arguing that open protocols and composable components will mirror traditional software engineering practices. The company believes that approach will preserve customer choice and enable broader integration across enterprise systems.
Vercel’s latest announcements underscore a broader industry shift from experimentation to operations, where governance, data control and cost efficiency become the decisive factors for adoption. As enterprises bring agents into production, platforms that balance developer experience with security and auditability will likely determine which tools scale across organizations.
