AI agents are often described as software that can take action, not just provide answers. Skyfire Systems is targeting one of the biggest gaps in that idea: payment. If an agent can plan a task but cannot complete the transaction, a person still has to step in at the moment money moves.
Skyfire has created a payment network built specifically for AI agents to make autonomous transactions. The company is trying to make those payments possible while keeping spending inside clear boundaries.
The problem Skyfire is trying to solve
Today, an AI agent might be able to plan a vacation independently, but a human still has to enter credit card information when it is time to pay. That limitation matters because payments are what turn a recommendation or plan into a completed action.
Skyfire co-founder and chief product officer Craig DeWitt framed the issue directly in an interview with TechCrunch: "AI agents can’t do anything if they can’t make payments; it’s just a glorified search." He added: "Either we figure out a way where agents are actually able to do things, or they don’t do anything, and therefore, they’re not agents."
The concern is obvious. An AI agent connected to money can create risk if it acts incorrectly, spends too much, or makes decisions a business did not intend. Skyfire’s answer is not to give agents unlimited access to a bank account. Instead, it gives each agent a controlled way to transact.
How the Skyfire payment network works
Skyfire assigns each AI agent a digital wallet with a unique identifier. Businesses can place a set amount of funds in that wallet, limiting what the agent can spend from the beginning.
The network also allows customers to set spending limits. Those limits can apply to a single transaction and to spending over time. If an AI agent tries to go beyond those rules, the system pings a human to review the request.
Skyfire also provides a dashboard that lets customers see how much their agent is spending and where the money is going. That visibility is central to the pitch: a company can let an agent transact, but still monitor the activity and cap the exposure.
The main controls described by Skyfire include:
- A separate digital wallet for each AI agent.
- A unique identifier tied to that agent.
- A preset amount of funds that a business deposits.
- Limits on spending per transaction and over time.
- Human review when an agent attempts to overspend.
- A dashboard showing spending amounts and destinations.
Skyfire does not build the AI agents themselves. Its role is the payments network those agents can use. The company’s founders say the responsibility for making agents behave responsibly remains with the companies building them.
Why blockchain and USDC are part of the system
Skyfire built its payments network using blockchain technology. Businesses can deposit and withdraw U.S. dollars from the platform, but behind the scenes Skyfire converts those dollars into a digital stablecoin.
The stablecoin used by Skyfire is USDC, which is pegged to the American dollar’s value. Skyfire holds that stablecoin in a wallet connected to the AI agent.
The founders have experience in payments infrastructure. Amir Sarhangi, Skyfire’s co-founder and CEO, previously sold Jibe to Google. Jibe helped pioneer the RCS messaging protocol, which became the standard for Android’s billion users. Sarhangi is now trying to develop an open protocol for payments in the AI era.
The founders were also early executives at Ripple, the cryptocurrency startup. While there, they helped build a cross-border payments network that processed more than $50 billion during their time there.
Skyfire’s revenue model starts with transaction fees. The company collects 2% to 3% of every transaction. It also says verification services could become another revenue source later.
Early use cases are focused on business transactions
Skyfire officially launched its payment network on Wednesday and announced $8.5 million in seed funding. The funding came from Neuberger Berman, Inception Capital, Arrington Capital, and other investors. Arrington Capital is led by Michael Arrington, the founder of TechCrunch, who left the publication in 2011.
The company has already tested the network in beta over the last two months. Some AI agents have been spending company funds through Skyfire during that period, according to the founders.
Denso, a global auto parts manufacturer, created AI agents to source materials without human help. Those agents could find the materials Denso wanted to buy, but humans still had to step in at the end of the month to conduct a wire payment. Skyfire says its platform now enables Denso’s agents to operate more autonomously.
Payman is another company using Skyfire. Payman allows AI to pay humans for different tasks, in a model compared in the source article to Fiverr. With Skyfire, Payman’s AI agents can hire and pay contract workers autonomously, at least in theory.
For now, Skyfire is focused on B2B uses. That means the immediate target is not a consumer agent shopping on behalf of an individual, but business agents handling tasks such as sourcing materials or paying workers.
The larger bet on agent payments
Skyfire’s founders believe AI agents could change how purchases happen on the internet. Today, online buying usually requires people to fill out personal information, verify identity, and complete payment steps directly.
Skyfire’s longer-term vision is that an AI agent could become a secure intermediary between vendors and a bank account. In that model, the agent would not need open-ended access to money. It would transact through a system designed around wallets, limits, review, and visibility.
Sarhangi described the ambition in broad terms: "The protocol we built will be an open protocol that any company, even a competitor, can use." He added: "We want this to be the thing everybody uses when it comes to payments in the AI world."
The company’s challenge is the same one facing the broader AI agent market: autonomy is only useful if it can be trusted. Skyfire is betting that payments for AI agents will require a dedicated network, not just existing checkout flows with an agent placed in front of them.