Today Skyfire is launching a production integration with Sequentum, the enterprise web automation company. Starting now, Sequentum’s new Checkout Agent can complete real purchases on real merchant sites using Skyfire for identity and payments, inside a SOC 2 Type 2 environment with full audit trails for every transaction. The first live demo runs on May 6th.
I want to explain how this actually works, because the architecture is a specific bet on how AI agents currently transact on the open web, and it cuts against most of the current thinking.
The API hope
For the last two years, the cleanest version of agentic commerce on paper has looked like this: every merchant exposes a standardized product catalog and checkout API, agents authenticate (somehow), and a transaction completes in milliseconds over a REST call. Beautiful. It’s also not the world we live in, and I don’t think it’s the world we’re about to live in.
Merchants have spent twenty years optimizing checkout funnels for humans. Those funnels are where they test pricing, run loyalty programs, capture email, and gate regional inventory. Exposing a clean agent-ready API means giving that up, or rebuilding it twice. Some large platforms will do it. Most won’t, not soon, and not in a way that covers the long tail of inventory an enterprise buyer actually wants to reach.
So if you’re an enterprise right now and you want an agent to procure consumer packaged goods, building materials, pay regulatory filing fees, or buy from a specialized supplier whose site was last redesigned in 2019, waiting for that merchant to ship an agent API is not a strategy. It’s a brick wall.
The website is the integration layer we already have. Every merchant supports it. Every checkout works through it. An agent that can operate a browser the way a human does can transact against essentially any merchant today, without waiting for anyone to rearchitect anything.
That’s the substrate Sequentum brings. Eighteen years of enterprise web automation, refined against the most change-prone sites on the internet, deployed in a SOC 2 Type 2 environment. Their agents don’t look like bots to a merchant’s anti-fraud stack, because they weren’t built to. They were built to act like careful, deterministic human operators, and they’ve been doing that for investment banks and global brands since before “agent” meant what it means now.
What Skyfire does inside this
The integration is mechanically simple, and that simplicity is the point.
A Sequentum agent configured with a Skyfire API key at the organization level can generate a signed KYA (Know Your Agent) token. That token carries three things: a verified identity for the enterprise platform, agent identifiers, as well as the verified end user identity. The token gets delivered as a custom HTTP header (skyfire-pay-id) on every request to the target site URL.
When the merchant’s security infrastructure receives the request, it receives a cryptographically signed object that specifies which enterprise is on the other end, who has authorized the agent actions, and how long the authorization remains valid. No corporate card number is exposed. No shared credentials sit in a config file waiting to leak. If the token is intercepted, it’s scoped to one seller, one amount, and one short window, so it’s worthless outside its intended transaction.
The reason this matters is that it solves the two problems that have kept enterprise risk teams saying no to agentic commerce. The first problem is credential exposure: you cannot give an autonomous process your corporate Amex and call it governed. The second problem is attribution: when something goes wrong in an automated purchase, somebody has to be accountable, and “the bot did it” is not a defense that holds up in an audit.
KYA fixes attribution at the protocol level. Every transaction is non-repudiably tied to the enterprise and user who authorized the agent. Combined with Sequentum’s per-action logging, you get a transaction record that looks less like a scraped purchase and more like a procurement event, with a complete decision trail and a verified buyer. With every step of the Sequentum Checkout Agent documented, third-party fraud management tools can detect fraud signals throughout.
That’s the thing a CFO can sign off on. That’s the thing an auditor can trace end to end.
What enterprises can ship this quarter
Sequentum’s Checkout Agent is in production today. The Skyfire integration is documented in their Cloud manual; there’s a configuration page under Organization settings, and the token flow I described above is a live feature an enterprise can turn on this week. On May 6th, we’re running a live launch with an agent buying a real product from a real merchant on the real open web, fully logged inside a SOC 2 Type 2 environment.
The use cases we’re seeing early interest around are the ones where the procurement logic is clear, but the execution is currently manual: accessing websites for product discovery, logging into enterprise accounts to pull invoices, accessing services to automate routine actions, and checking out at ecommerce sites. All of these are jobs where a human is currently clicking through a checkout funnel that was built for a human, and none of them are waiting for a merchant API.
I’m not claiming this settles the protocol wars. Agent Checkout standards will continue to evolve, and Skyfire continues to invest in staying at the forefront of API-based protocol checkout. But the question every enterprise builder is asking me right now is not “which standard will win.” It’s “What can I ship this quarter that my risk team will approve?” Sequentum plus Skyfire is an answer to that question, and it works against the web as it actually exists.
The companies that figure out agentic commerce first are not going to be the ones that waited for a clean protocol. They’re going to be the ones that moved through the messy substrate we already have, with the governance controls regulated enterprises require, and built real transaction volume while everyone else was still drafting RFCs.
If you want to see what that looks like in practice, join Sarah McKenna and me on May 6th for a live Webinar discussing the technical specifics. Bring your risk team.
