Vercel Bets on AI Agents Beyond Model Lock-In

Vercel CEO Guillermo Rauch says the next big fight in AI will not be only about which company has the best model. It will be about who controls the agent layer around those models.

In a new interview after Vercel’s ShipNYC event, Rauch explained that companies are moving beyond simple AI experiments and are now trying to put agents into real production workflows. For Vercel, this shift is becoming a major business opportunity. The company is already seeing around 6 million deployments a day, and about half of them are triggered by coding agents, according to TechCrunch. More than 1 trillion tokens also flow through Vercel’s AI Gateway daily.

That scale puts Vercel in an important position. It is no longer just a hosting platform for web apps. It is becoming part of the infrastructure layer where AI agents are built, deployed, monitored, and connected to different models.

From AI Prototypes to Production Agents

Rauch said the AI market has changed quickly over the past year. Earlier, many companies were focused on prototypes. They were testing what AI agents could do, building demos, and experimenting with model capabilities.

Now the focus is different. Businesses want agents that can work safely inside real companies. That means agents need access to internal tools, company data, approval systems, audit logs, permissions, and reliable deployment environments.

This is where Vercel sees a major opening. Coding agents are already creating and deploying large amounts of software. But Rauch also sees another major category: internal company agents. These are AI systems designed to help teams complete business tasks, manage workflows, answer questions, or operate tools across departments.

The challenge is not only making the agent smart. The bigger challenge is making it secure, observable, and useful in production.

Models Are Becoming More Swappable

One of Rauch’s biggest points is that companies are no longer thinking about AI as a one-model decision. Last year, many businesses were choosing a single provider, such as OpenAI or Anthropic, and building around that model.

Now, according to Rauch, companies understand that the AI stack has many separate parts: the model, the agent framework, the data layer, the sandbox, the gateway, and the deployment platform. Once those parts are separated, businesses can switch models depending on cost, speed, quality, and use case.

That is the “split” Rauch is talking about. The model is only one layer. The agent is the full system that uses the model, connects to tools, follows instructions, performs tasks, and produces work.

This matters because it reduces dependence on one AI lab. A company may use OpenAI for one workflow, Anthropic for another, Gemini for a lower-cost production task, and open models for specific internal use cases.

Eve Is Vercel’s Framework for Building Agents

Vercel has also introduced Eve, an open-source framework for building and running AI agents. The company describes Eve as a filesystem-first framework for durable backend agents on Vercel

Eve lets developers define an agent through instructions, skills, tools, and runtime configuration. Vercel presents it as something similar to Next.js, but for agents. Instead of building all the production plumbing from scratch, developers can define what an agent should do and deploy it with built-in support for workflows, tools, approvals, and model access.

This is important because agents are becoming more complex than chatbots. A chatbot mostly responds. An agent may take actions, call APIs, update systems, write code, access company documents, or trigger business processes. That requires stronger controls than a normal prompt box.

Security and Auditing Are Becoming Central

Rauch’s comments also show how seriously Vercel is thinking about security. If an internal agent can access company systems, businesses need to know what it did, which tools it used, what data it touched, and whether a human approved sensitive actions.

This is where audit trails and access controls become essential. In a normal software system, companies already track user actions. In an agentic system, they need similar tracking for AI actions.

That includes questions such as: which model made the decision, what instruction guided it, which tool was called, what data was retrieved, and what output was produced. Without that visibility, businesses may struggle to trust agents in serious workflows.

The AI Labs Are Moving Into Vercel’s Territory

The tension is that AI labs are also expanding beyond models. OpenAI, Anthropic, Google, and others are building more tools around their models. Some of these tools help users create apps, publish websites, run workflows, or connect agents to external systems.

That creates direct competition with infrastructure companies like Vercel. If a model provider also offers the agent builder, hosting layer, and publishing tools, developers may stay inside that company’s ecosystem.

Rauch sees this as a natural move by AI labs, but Vercel’s bet is different. Instead of tying developers to one model provider, it wants to be the neutral layer where different models can be used inside production-ready agents.

Vercel’s AI Strategy Is About Control, Flexibility, and Scale

Rauch’s message is clear: the AI market is moving from model excitement to production infrastructure. Companies still care about powerful models, but they also care about flexibility, cost, reliability, security, and control.

That makes the agent layer more important. Whoever controls that layer may shape how businesses use AI in daily work.

Vercel is positioning itself as the platform where agents can be built, deployed, audited, and connected to multiple models. The company’s argument is simple: models will keep changing, but businesses need stable infrastructure around them.

As AI agents become more common in coding and internal business operations, the real competition may not be just between model labs. It may be between the platforms that decide how those models are used.