Vercel Agent uses staged, scoped permissions for production work

Vercel Agent begins a production investigation with read-only access, which lets it inspect recent deployments and build logs and compare a failed deployment with the last known-good one. If that leads to a proposed redeploy, the agent presents a plan listing its tasks and the scopes each task needs, and the user approves the whole plan in one decision.

Access is defined by scopes, the named capabilities the Vercel authorization system recognizes. The agent runs as a Vercel App that the user authorizes within a team, and that authorization is capped by the user's existing permissions. Vercel OAuth exchanges approval for shorter-lived tokens with specific scopes, and Vercel APIs check each request against them.

Any plan that asks for access beyond the read-only baseline waits for approval as a whole, including its read-only tasks. Plans also need approval when they include supported external actions, such as opening a GitHub pull request. Vercel notes that read-only operations can still expose sensitive information, so automatic approval does not make the work harmless.

Agents working on production can now run read-only investigations without a prompt for each step, while writes and external actions wait for a single plan-level approval. Builders should expect approval to cover every task in a plan, including the read-only ones. Scopes and the approving user's own permissions set the outer limit on what an agent can do.