What changed
On August 28, Vercel added a dashboard flow for creating and deploying eve agents. The builder creates a private Git repository, deploys it as a Vercel project, and gives the team an agent they can chat with right away. You define the agent's identity, pick a model through AI Gateway, add a Next.js web chat or Slack channel, and connect services such as Linear, Notion, or a custom MCP server. Vercel's changelog explains the new builder.
Why it matters
This is more than a shorter setup command. It makes the first mile of an agent a product workflow: choose the behavior, wire the interface, provision the runtime, and receive source code that remains available for review and customization. The underlying eve framework is filesystem-first: Markdown defines instructions and skills, TypeScript defines tools, and one agent can be surfaced through web, Slack, API, cron, or other channels. The eve framework overview describes that model and its durable execution, sandbox, connection, and subagent primitives.
For product teams, that reduces the gap between an internal experiment and a deployable service. It also makes platform trade-offs visible earlier. Model selection, channel access, MCP permissions, deployment ownership, and approval behavior are no longer distant infrastructure decisions; they are part of the initial product shape. Vercel's original eve launch post shows the broader goal: agents that ship with durable execution, sandboxed compute, human approvals, and multiple delivery surfaces.
Builder takeaway
Treat dashboard scaffolding as the start of a product boundary, not the finished agent. Keep instructions and tools in version control, define least-privilege connections, test web and Slack surfaces separately, and add explicit approval steps for writes or other irreversible actions. The useful pattern is the handoff: a fast generated baseline with a real repository and deployment that engineers can review, test, and evolve.