The Agentic Stack Just Got Its Payment Layer

A week that changed everything

Last week I wrote about the agentic infrastructure stack forming — Cal.com for scheduling, AgentMail for email, Signbee for signatures. It was a thesis. Three pieces of a puzzle that felt like it was heading somewhere.

Then Stripe dropped MPP.

The Machine Payments Protocol is an open standard that lets AI agents pay for services autonomously. No API keys. No sign-up flows. No billing dashboards. An agent calls an API, the server responds with a price, the agent pays, and the resource is delivered. One HTTP round-trip.

The thesis got its fourth piece — and suddenly it stopped being a thesis.

An AI agent can now research a prospect, send personalised outreach, book a meeting, draft and sign a contract, and process payment. No human. No GUI. End to end.

What MPP actually is

I spent yesterday evening reading through the MPP docs and the pattern is elegant. It uses HTTP 402 — the “Payment Required” status code that's been sitting in the HTTP spec since 1997, waiting for its moment.

The flow works like this:

1. Agent calls your API — POST /api/v1/send
2. Your server responds 402 Payment Required with payment options
3. Agent pays via a Tempo wallet (USDC on-chain)
4. Agent retries with a payment credential
5. Server verifies payment, delivers the resource, returns a receipt

No API keys. No OAuth. No account creation. The agent just pays and goes.

The services directory already has 50+ integrations — OpenAI, Anthropic, Firecrawl, fal.ai, Browserbase, AgentMail. Proper infrastructure. And right now, there's not a single e-signature service listed.

That's about to change.

Signbee + MPP

We're integrating MPP into Signbee. The plan is straightforward: add the 402 payment challenge flow to our /api/v1/send endpoint so any agent with a Tempo wallet can send a document for signature without needing an API key or account.

Think about what that means. An agent running autonomously can:

1. Discover Signbee through the MPP services directory or llms.txt
2. Call the API to send a contract
3. Pay per document — no subscription, no billing page
4. Get a webhook when the document is signed

Zero onboarding. The agent finds us, uses us, pays us. That's the agentic business model in its purest form.

And then DoorDash happened

As if Stripe MPP wasn't enough for one week, DoorDash announced they're building infrastructure for AI agents to dispatch human workers for real-world tasks.

Read that again. Agents hiring humans.

The agentic stack just got a physical layer. Agents could already handle digital workflows — emails, scheduling, contracts, payments. Now they can dispatch someone to pick up a package, deliver documents, or run an errand. DoorDash is becoming the API between AI and the physical world.

And here's the quiet genius: every completed task generates training data for the robots that will eventually replace the human doing it. It's a self-funding data flywheel disguised as a gig economy extension.

The speed is the story

What strikes me most isn't any single announcement. It's the pace.

A week ago, the agentic infrastructure stack was three pieces and a blog post. Today it's four pieces with a payment protocol, a physical-world layer forming, and 50+ services already integrated.

We launched Signbee on Product Hunt this week. The AgentMail co-founder reached out. Cal.com reposted our stack tweet. Stripe validated the entire thesis with MPP.

I keep thinking about something I wrote last week: “We're roughly one year away from an AI agent that runs a full sales cycle.” After this week, that estimate feels generous. The infrastructure is shipping faster than anyone — including me — predicted.

What's next for Signbee

Short term: MPP integration. We want to be the first e-signature service in the MPP directory. When an agent needs a document signed, it should find us the same way it finds OpenAI for inference or AgentMail for email — through the protocol, not through a Google search.

We're also adding idempotency keys to the API — a missing primitive that the HN community flagged. Essential for agents running unsupervised with retry logic. If the same request comes in twice, you get the same document back instead of a duplicate.

The agentic era isn't coming. It shipped this week. We're just trying to keep up.