Vercel has been building the infrastructure for AI agents to act.
AI SDK gives developers the primitives to build agents.
AI Gateway gives those agents access to hundreds of models through one interface.
Sandbox gives them somewhere isolated to execute code.
Workflows give them durable execution.
Connect gives them access to external tools and services.
Put those pieces together and you get something very different from a chatbot.
You get software that can decide → call tools → execute actions → continue working.
And that’s where I think the next missing layer becomes interesting:
What happens when the action itself costs money?
Imagine a Vercel-built agent working on behalf of a user.
It needs to access a paid dataset.
Or call a specialized API.
Or purchase a digital service.
Or pay another machine for a specific task.
The agent can already reason about what it needs.
It can already use tools to get it.
But commerce introduces another requirement:
the agent needs a way to transact.
This is where I’d introduce @GOATNetwork to @vercel.
GOAT Flow isn’t just another checkout sitting on a website.
Its current architecture exposes machine-readable payment information through QuickPay, including agent-facing discovery and pricing surfaces. Agents can inspect the available payment terms and use x402/QuickPay flows to make payments programmatically.
That creates an interesting bridge between Vercel’s agent infrastructure and an agent-native commerce layer.
Vercel builds the agent.
GOAT Flow gives the agent a way to transact.
And the distinction matters.
A traditional checkout assumes:
Human → website → checkout → payment
An agent-native flow can look more like:
Agent → discover resource → read payment terms → authorize payment → receive verified result
GOAT Flow’s x402 flow is designed around exactly this kind of machine-readable payment challenge: the resource can return the payment requirements, the payer transfers the required asset, and GOAT Flow verifies the payment before fulfillment.
There’s another reason this fits Vercel particularly well.
Vercel’s own production data shows how quickly AI applications are becoming agent-shaped: its AI Gateway reports that agentic workloads represent 59% of token volume, with production teams often routing workloads across dozens of models.
As these systems make more tool calls and interact with more external services, the economic layer becomes increasingly important.
An agent that can make 10 API calls isn’t fundamentally different from an agent that can make 10 API calls and pay for the ones it needs.
That’s the step I’m interested in.
Not putting a crypto button on an AI app.
Making payment another tool an autonomous application can use.
GOAT Flow already supports the pieces businesses need around that: Checkout, QuickPay, API Payments, card + crypto payments and x402.
Vercel has spent years making the web programmable.
Now its agent stack is making the web actionable by software.
The next question is whether that software can participate in the economy without constantly handing control back to a human.
Build the agent.
Give it tools.
Give it a way to transact.
That’s where I see an interesting Vercel × GOAT Flow connection.
👉https://t.co/tppoUR4q4E