AI agents need badges, not borrowed logins
AI agent identity is the idea that autonomous software agents should authenticate with their own scoped, revocable credentials instead of reusing a human user’s account, so organizations can grant, audit, and revoke agent permissions independently while keeping human identities and responsibilities clearly separated.
Vercel’s acquisition of Better Auth is a clear bet that this concept will define the next wave of AI infrastructure. Better Auth, a TypeScript authentication framework with around 4.7 million weekly npm downloads, is best known for login, sessions, and permission tooling for web apps. But its more interesting work is Agent Auth, an open protocol that gives AI agents identities separate from the people deploying them, with delegated and revocable permissions. Today, most agents “borrow” their operator’s login; every system they touch sees the human, not the agent. That is a security nightmare in waiting. If AI agents are going to write code, ship deployments, and query internal systems at scale, they need their own badge at the door—not a copy of yours.
Vercel’s scale makes AI agent identity impossible to ignore
The move only makes sense when you look at Vercel’s numbers. Its cloud infrastructure handles 6 million deployments daily, with roughly half already triggered by coding agents. That means millions of automated actors are pushing code and touching production systems every single day. Add more than 1 trillion tokens flowing through its AI gateway daily, and you get a picture of AI agents not as experiments, but as routine participants in software delivery.
In that context, leaving agents to operate under borrowed human identities is not a minor oversight; it is systemic risk. Guillermo Rauch has spelled out the core problem: when an agent runs under your identity, every service sees you, not the agent, and there is no clean way to limit or selectively shut down a specific agent or subagent. That is incompatible with serious AI infrastructure security. As coding and internal corporate agents move from prototypes to production challenges, AI agent identity stops being a nice-to-have and becomes a prerequisite for governance.
Agent Auth protocol: turning AI agents into first-class citizens
Agent Auth is Vercel’s answer to this gap: an open protocol that gives AI agents their own identities, with scoped, delegated, and revocable permissions that stay separate from their human deployers. In practical terms, it pushes agent authentication closer to how we treat microservices and humans today: each actor has its own account, its own roles, and its own audit trail. That is the only sane path if enterprises want to know which agent did what, when, and under which authorization.
This is not happening in isolation. Another major provider has launched a feature that gives its assistant its own presence in Slack under connected accounts, rather than acting through whoever invoked it. The implementations differ, but the idea is the same: treat AI agents as distinct actors with their own identity, not as shadow extensions of human users. Once agents are first-class citizens in the identity system, you can finally give one agent read-only access to a CRM, another full deployment rights, and revoke either without touching human access. That is what real AI agent identity looks like.
Why enterprises should care: auditability, revocation, and AI infrastructure security
Enterprises already know the pain of over-privileged human accounts. AI agents without precise identities repeat that mistake at machine speed. Today, if you deploy an internal automation agent using your own login, every action it takes—opening pull requests, updating business apps, hitting internal APIs—lands in logs under your name. You cannot easily prove which agent did what, you cannot limit a rogue subagent without shutting down the whole setup, and you cannot apply least privilege to each agent. That is the opposite of good AI infrastructure security.
By contrast, a protocol like Agent Auth promises scoped, delegated, and revocable permissions for each agent. Combine that with Vercel’s Eve framework, where companies define agent instructions and skills in natural language, and Vercel Sandbox, which restricts what data agents can access or export to protect sensitive information, and you start to see a coherent security story. Role definitions, data boundaries, and independent revocation all converge on one goal: agents that can be trusted inside compliance-heavy environments because they are both powerful and accountable.
What this acquisition signals for the AI stack
Better Auth did not arrive here by accident. It began as a response to gaps in existing authentication libraries while its creator was building an analytics project. The framework grew quickly, entered an accelerator, raised a seed round, and later took over stewardship of Auth.js when both projects aligned. But the founder admits the agent identity problem is “big and moving fast” and beyond a small team’s capacity alone. Joining Vercel brings scale, infrastructure, and distribution to push Agent Auth into mainstream AI workflows.
Importantly, Vercel says the Better Auth framework remains open source under an MIT license, with the same open contribution model and ecosystem support. That matters: identity groundwork for AI agents should not be a closed black box. The acquisition signals a broader shift in the AI stack. As infrastructure platforms converge with AI labs and enterprises adopt multi-model strategies, authentication and authorization become shared foundations, not afterthoughts. Treating agents as first-class identity citizens is no longer speculative architecture; it is the new baseline. The organizations that internalize this now will deploy more AI, with more confidence, and fewer late-night incident calls.
