Unlike other AI agents that perform transactions on behalf of humans, these AI agents act as if they were the human users themselves.
For example, AI agents today ask permission before booking and paying for hotels as they are an extension of their human users.
With Skyfire’s Agent Checkout, the user instructs the agent, such as on budgets and approved shopping websites, and the agent does the rest autonomously. This includes creating an account, verifying an email, adding credit card information and making a purchase.
Skyfire also gives the AI agent its own identity and credentials, so when it transacts, the system treats it just like it would any human user.
Built on top of a new open standard called KYAPay — or Know Your Agent Pay — the platform can even handle microtransactions below $5 in value.
“The KYA protocol really is about opening up our product so that anybody can adopt it,” said Amir Sarhangi, CEO and co-founder of Skyfire, in an interview with PYMNTS. “It allows for coordination across many different sectors.”
Upping the Scrutiny
The new protocol comes at a time when governments, banks and businesses are reassessing and strengthening their know your customers (KYC) systems. For example, Wise implemented recommendations put forth by a European regulator; and the U.K. is cracking down on banks for AML violations.
In fact, 7 in 10 financial institutions are now using AI and machine learning to fend off bad actors, according to the PYMNTS Intelligence and Hawk collaboration, “Financial Institutions Revamping Technologies to Fight Financial Crimes.”
For its part, the U.S. has indicated that fighting fraud would continue to be a bipartisan effort.
Skyfire has launched with about 30 partners, including APIFY, CarbonArc, BuildShip and Forter.
Sarhangi said he sees a future where AI agents can research, subscribe to tools, access gated data and even pay for content or services, all on their own.
“Today, agents get blocked” from doing tasks because something in their workflow needs human intervention, Sarhangi said. “An agent finds a particular tool it needs access to — a human needs to get involved to unlock it.”
See also: Skyfire CEO Says AI Agents Need New AI Payments Rails
AI Agents as a New Type of Customer
The core of Skyfire’s system is the KYA — or Know Your Agent — protocol, which Sarhangi said provides a way for businesses to identify and trust who or what is attempting to access their services.
“The first thing they need to do is say, ‘Who is this agent that’s trying to access my services? Does this person already exist in my database as a customer? If they don’t, I want to be able to create an account for them and then be able to take a payment,’” Sarhangi said.
Skyfire’s system allows for direct integration with representational state transfer application programming interfaces (REST APIs) and supports OAuth2 and OIDC protocols, meaning businesses don’t need to rebuild their authentication systems. Developers can onboard using Skyfire’s software development kits (SDKs) and reference implementations, which Sarhangi said take as little as 15 minutes to integrate.
According to Sarhangi, initial demand is strongest from developers building AI agents that require access to tools like customer relationship management (CRM), data platforms and authentication services.
Agent Checkout also introduces controls to manage the risk of giving agents spending authority.
Skyfire is rolling out “just-in-time decisioning” that lets users approve high-value transactions in real time.
“When an agent hits a particular size transaction … the human can enable it to ask for a decision point at the moment of that transaction,” Sarhangi said. “It’s able to send you an SMS (text message), you approve it, and it continues on its work.”
A Balancing Act
Sarhangi acknowledged that fully autonomous AI raises trust concerns, particularly in enterprise or B2B payments.
It’s an issue that is garnering much attention, as C-suite executives grapple with balancing the new technology’s possibilities with responsible use. That’s a concern brought up by Trustly Chief Legal and Compliance Officer Kathryn McCall, who spoke to PYMNTS during a conversation last month.
There needs to be a balance between human oversight and AI automation as agents take over more tasks.
“Can you trust an agent to just do whatever? No,” Sarhangi said. But “I think that has gotten better over the last six months.”
Take hallucinations. “Every month you’re getting significant improvement,” Sarhangi said. “You don’t even hear about hallucination as much anymore.”
He does see a near-term opportunity for Agent Checkout in handling predictable, repetitive money flows like B2B supplier payments and subscription-based services. “B2B payments are probably the most predictable,” Sarhangi said. “A lot of them have very predictable patterns in terms of what’s being bought on a monthly basis.”
As the AI economy evolves, Sarhangi believes a new category of customers is emerging: AI agents themselves.
“This is where I get really excited — we’re going to actually be creating a completely new category of revenue,” he said. “Agents as a new customer category — and that’s where I think every CFO should be focused on.”
Read more:
