Agentic Payments Protocol: Infrastructure for Ai Commerce

Much of the conversation about how Ai will reshape commerce now revolves around agentic commerce—shopping orchestrated by agents—and bots that can authorize payments on their own.

What gets less airtime is the web-wide plumbing—protocols and interoperability for payment workflows—that would let those agents traverse sites and complete checkouts with near-zero friction.

An agent transaction protocol is a shared set of message formats and rules that lets software agents request prices, confirm identity, present user authorization, initiate a payment, and exchange receipts with merchants and payment systems. In agentic commerce, it’s needed because today’s checkouts assume a human clicking through forms, while agents need a consistent, machine-readable path from “intent” to “paid” across many sites.

These protocols typically establish trust by binding an agent to a verifiable identity and requiring cryptographic proofs—such as signed requests, nonce-based replay protection, and auditable receipts—so merchants can confirm who is acting and under what authority. Security approaches also lean on least-privilege permissions, encrypted transport, and revocable credentials to limit the blast radius if an agent or device is compromised.

Secure interoperability is what turns agent-driven payments from clever automation into something merchants can trust at scale.

The core actors are the consumer-side agent (acting for a person or business), the merchant side (a website, merchant agent, or checkout service), and the protocol infrastructure that routes messages and standardizes how mandates, payment instructions, and status updates are expressed. Payment providers, risk systems, and protocol operators can sit alongside these actors to handle settlement, fraud checks, and interoperability across platforms.

A common “mandate model” is how the user delegates spending authority to an agent: the user creates a mandate that encodes constraints (amount, merchant, category, time window) and conditions (approval rules, dispute handling), then the agent presents that mandate when it requests a charge. Mandates can be rotated, paused, or revoked, and merchants or payment systems enforce them by validating the mandate’s scope before accepting or capturing funds.

For merchants, agent-friendly protocols can reduce checkout friction, lower abandonment, and make it easier to support new buying channels without custom integrations for every agent. For consumers, they can make recurring and routine purchases safer and more controllable by turning “trust me” interactions into explicit, inspectable permissions. For developers, a shared protocol can replace one-off scraping and brittle automations with stable interfaces, testable flows, and clearer error handling.

Because the protocol sits above the payment rails, it can describe requests that settle via cards and bank transfers as well as newer options such as wallet-based payments and Web3 systems. In practice, that means an agent can negotiate what the merchant accepts, then choose an appropriate rail—traditional or onchain—while still returning a standardized confirmation and receipt to the rest of the workflow.

For example, a small business could authorize an operations agent to restock office supplies under a monthly cap, letting it compare prices across vendors, place an order, and pay without a human re-entering billing details each time, while still producing a receipt and an audit trail for accounting.

Different teams are proposing adjacent standards, and the names can be confusing: a payments-focused protocol such as Ap2 is typically scoped to authorization, payment initiation, and settlement-related messages, while a more general agent coordination protocol such as Acp is designed for broader task negotiation and inter-agent communication. The practical differentiator is where the standard draws the line—payment-specific guarantees and controls versus general-purpose agent collaboration.

Businesses preparing to integrate usually start by making key checkout and post-purchase steps machine-readable—product details, pricing, tax and shipping rules, refund policies, and receipt formats—then exposing a secure endpoint that can verify mandates and return deterministic payment status. Many also plan for operational needs up front, like dispute workflows, logging, and clear fallbacks when an agent’s request is ambiguous or outside policy.

The ecosystem’s near-term challenge is fragmentation: multiple protocols may compete before interoperability hardens into common patterns. Longer-term questions include liability when an agent misbehaves, standardized dispute and refund handling, and the balance between privacy and the verification merchants need to manage fraud.

Circuit & Chisel, a New York software startup, is tackling that gap. In September, the company raised $19.2 million to introduce Atxp, an agent transaction protocol built to help an Ai agent navigate the web and pay, enabling agent-to-agent (A2a) interactions between agents and merchants with fewer hiccups.

The firm was co-founded by two former Stripe leaders. CEO Louis Amira previously ran crypto and Ai partnerships at the payments giant, and David Noel-Romas, Stripe’s former head of crypto engineering, now serves as Circuit & Chisel’s chief technology officer. Both departed Stripe last year.

Stripe invested in the startup, alongside Polygon Labs and Samsung Next, the investment arm of Samsung Electronics.

Samsung Next outlined its thesis in a September blog post about the deal.

Payments are a foundational layer in technology, and Circuit & Chisel has the team, product, and timing to position Atxp as the go-to rails for autonomous software.

Amira, based in the New York metro area, discussed the company in a Feb. 23 interview.

Editor’s note: This interview has been edited for clarity and brevity.

Why Leave Stripe to Build This Agent Payments Protocol?

Louis Amira: We parted on excellent terms. We stood up Stripe’s crypto team, and even after the Ftx collapse raised internal questions about whether to keep investing, the founders held firm and we continued. Later, after Stripe acquired Bridge and other companies, a few investors asked what farther-out concepts we wanted to pursue. We had some ambitious ideas. We built for a long stretch and launched, but the market wasn’t quite ready. Stripe could have underwritten that future, but it was likely too early for them. Once they saw what we were building and why, they wanted to back it as investors and stay close to the protocol’s trajectory.

Where Circuit & Chisel Fits as the Agentic Ecosystem Matures

I thought I understood the multi-year outlook when we left a little over 12 months ago, but I’m less certain now. We’ve spoken with roughly ten peers working on similar problems and with strategic players such as Coinbase, Solana, Stripe, and other investors. At this stage, few are prepared to make outsized bets because the path remains unclear.

Many Protocols Are Emerging: What Happens Next for Agent-Driven Payments?

We have strong conviction about our roadmap for the coming months. A development that once sounded far-fetched is starting to materialize: not just people, but agents are signing up and effectively installing Atxp themselves so they can use our tools. Those agents configure credentials and then run transactions through the system. Looking ahead, agents are very likely to become our primary customer—a notion that drew laughs a year ago but now feels far more plausible.

Shopping and Bills: Do People Keep Steering Agents or Fade Into the Background?

I expect your agent to take on more each year without requiring your input. Routine items—like the same bill you have seen for months—get handled automatically, no archiving required.