Vercel dashboard now scaffolds and deploys AI agents backed by a Git repo

Agents straight from the dashboard

Vercel has added an agent builder to its main dashboard. According to a changelog entry published on the Vercel blog on August 28, selecting "Add New → Agent" starts a short guided flow that produces an "eve agent" — a deployed AI assistant you can talk to as soon as the process finishes.

The detail that sets the feature apart is what the flow leaves behind. Vercel says the builder scaffolds the agent's files into a new private Git repository and deploys that code as a brand-new project under your team. You are not merely configuring an agent inside someone else's console; you end up with a working deployment backed by ordinary source code that you can open, edit, and extend whenever the defaults stop being enough.

What the builder asks for

Before the agent goes live, the builder walks you through the decisions that shape it. Per the Vercel blog, the options include:

• Describing the agent's identity in plain written instructions, covering who it is and how it should behave.
• Choosing from any of the models offered through Vercel's AI Gateway, rather than being tied to a single provider.
• Giving people a place to converse with the agent, either a Next.js web chat or a Slack channel.
• Connecting the agent to outside tools and data, through built-in integrations such as Linear and Notion, or through a custom MCP server.

The entry point is the same "Add New" menu teams already use for creating projects, which keeps the agent path close to Vercel's existing workflows.

Why it matters

The release extends the playbook that made Vercel popular — going from code to a running URL with minimal friction — to a new category of software. Instead of assembling an agent stack by hand, a developer can get a live, chat-capable agent on the platform they may already use for deploys.

The Git-backed design is the most consequential part. A generated agent that exists as normal code in a repository is a starting point rather than a ceiling: teams can modify the scaffold, run it through standard review, and treat it like any other project in their toolchain. That keeps the fast path from becoming a trap if requirements grow beyond the builder's checkboxes.

The integration choices also signal where agent tooling is heading. Bundling connectors like Linear and Notion covers common team workflows on day one, while support for custom MCP servers leaves room to wire up anything that exposes its capabilities through the protocol — a pragmatic hedge as MCP becomes a shared language for tool access.

Slack support matters for adoption too. The people who need an agent's help do not have to visit a bespoke web app, and the Next.js chat option remains available when a standalone interface is the better fit.