One-Third of Vercel Deployments Are Now Done by AI Agents Instead of People

Over the past three months, the number of Vercel deployments initiated by AI agents has doubled. Now agents launch 30% of all projects—a year ago that figure was closer to 3%. This is a massive jump, and it has revealed an old problem: the infrastructure was designed for people, expecting someone to configure servers, click the deploy button, and read logs. But now the acting agent is a machine. And this requires a completely different approach.

Machines Code Faster

Projects deployed by agents call AI services (such as OpenAI, Anthropic, or Claude) 20 times more often than projects written by people. This means AI agents write code that uses AI—and often write code for other AI agents. Such software operates at full speed: testing itself, pushing to production without waiting for approval. Claude Code accounts for 75% of such automated deployments on Vercel. This is not just a code editor—it's a full-fledged infrastructure player. Behind it are Lovable and v0 (6%), then Cursor (1.5%). But the key point—over six months, these figures have grown by 1000%. What was once an exception is becoming the rule.

Infrastructure Must Be Different

When a human writes and deploys code, they need control and visibility. When an agent does it, it needs speed. If clicks in a cloud web interface or manual Terraform management stand between code and the production system, the autonomous cycle breaks. Agents need clean APIs for deployment, preview URLs to verify results, and instant rollbacks. This radically changes requirements. Immutable deployments, instant rollbacks, and preview URLs on every commit—these are no longer conveniences for developers but an absolute necessity for machine-driven development. Otherwise, the agent simply won't be able to verify what it wrote.

The second level of complexity: agents themselves become long-lived workloads. They require multi-stage orchestration, routing between models, and cost control. Every API call, every request through the AI Gateway, every execution in a Sandbox—these are expenses. Infrastructure must be able to pause, resume, retry, maintain state, and log everything that happens.

Vercel's Three-Layer Solution

Vercel has offered an integrated stack of tools:

• AI SDK—a unified foundation for developing AI applications. Version 6 adds an abstraction for agents, allowing you to define an agent once and use it everywhere
• Chat SDK—access to dozens of chat platforms from a single codebase
• AI Gateway—a single point for hundreds of models with budgets, monitoring, fallbacks, and retry logic
• Fluid Compute—optimized for unusual phases of AI workloads (latency, concurrent requests, idle time)
• Workflows and Queues—pause, resume, retry, state persistence
• Sandbox—isolated environment for untrusted code
• Observability—monitoring of agent costs and behavior in production

What This Means

For fifty years, infrastructure assumed that humans would configure it, click the deploy button, and read logs. Each generation of software required a new generation of infrastructure. The cloud turned it into APIs. Then infrastructure began to be derived from the application itself. Now a third shift is happening: infrastructure is adapting to the fact that its end user is not a human, but an autonomous machine. And this is not happening slowly. In three months, everything has changed.