AI agent identity is the new production safety barrier
AI agent identity is the emerging practice of giving autonomous software agents their own scoped, auditable, and revocable credentials, separating their permissions and behavior from the human users who trigger or supervise them so that production systems can control, monitor, and, if needed, terminate these agents without disrupting human access or other automated workflows. Today’s AI agents open pull requests, review code, create deployments, query internal systems, and update business apps, yet most still operate under borrowed human logins. That is not a minor implementation detail; it is a structural risk. If an agent goes rogue, the only kill switch is often the person’s entire account. Vercel’s acquisition of Better Auth is not about another login library; it is a bet that AI-specific machine identity management will decide which autonomous agent infrastructure is safe enough to trust in production.
From Better Auth to Agent Auth: giving agents a real badge
Better Auth began as a TypeScript authentication framework that already sees about 4.7 million weekly npm downloads, handling login, sessions, user management, and permissions across frameworks. But its most important move was to look past humans. Before the acquisition, the team started building Agent Auth, an open agent authentication protocol meant to give AI agents identities of their own with scoped, delegated, and revocable permissions, separate from their human deployers. This is the core of AI agent identity: stop treating agents as invisible puppets and recognize them as actors that must be governed. The pattern is spreading. Another example is Claude Tag, which gives a model its own presence in collaboration tools instead of impersonating the user who invoked it. The message is clear: in the agent era, access control that cannot distinguish between a person and their tools is broken by design.
Why Vercel cares: 6 million deployments and half are machines
This push is not theoretical; it is driven by scale. Vercel’s cloud infrastructure now supports 6 million deployments every day, with roughly half of those triggered by coding agents, and more than 1 trillion tokens passing through its AI gateway daily. In other words, agents are already first-class citizens in CI/CD pipelines, even if the identity stack still pretends they are humans. According to Vercel’s CEO, the industry has moved from prototyping agents to wrestling with production realities, especially around coding workflows and internal automation. Yet existing identity systems were built for people, not autonomous software, so when an agent acts on your behalf, every service still sees you, not the agent, leaving no clean way to limit a specific agent or shut it down without affecting everything else. That is an untenable foundation for autonomous agent infrastructure.
Turning identity into infrastructure: Connect, Eve and Sandbox
Better Auth’s work on Agent Auth will now feed into products like Vercel Connect and Eve, which are aimed at wiring agents into external services. Eve lets companies define agent instructions and skills in natural language, while Sandbox restricts what data agents can access or export, limiting the risk that sensitive information leaks into training data. This stack is effectively a new identity and permissions fabric for machines: the agent authentication protocol defines who an agent is; Connect gives it controlled pathways into systems; Sandbox shapes what it can see and share. Rauch has emphasized that Better Auth remains open source under an MIT licence, with the same community governance and framework support, even after the acquisition. That matters because machine identity management will only work if the ecosystem can inspect, extend, and standardize the way agents authenticate, authorize, and leave audit trails across platforms.
The coming shift: from human SSO to agent-level governance
The AI ecosystem is converging on a simple truth: if agents are going to run production systems, they must have identities as clear and controllable as their human counterparts. Better Auth’s Agent Auth, now under Vercel’s roof, is one of the first explicit attempts to formalize an agent authentication protocol that treats agents as distinct security principals rather than shadows of user accounts. With enterprises adopting multi-model strategies for coding agents and internal automation, and infrastructure platforms racing to support them, the platforms that win will be those that make agent-level identity and access as routine as human single sign-on. AI agent identity is not a nice-to-have bolt-on; it is the next layer of autonomous agent infrastructure—and it will decide whether agents stay experimental toys or become reliable, auditable colleagues in production.
