Terry KS As AI agents become increasingly capable of making autonomous financial decisions, enterprises are facing a new and urgent challenge: how to maintain human oversight when machines begin handling real economic activity on their own.
In an exclusive interview with VSDaily, Rodrigo Coelho, CEO of Ampersend, shared how the company is tackling what he describes as the “agent control gap” — the missing governance layer between AI agents and money.
According to Coelho, this challenge barely existed a year ago because AI agents were not yet capable of autonomous transactions. But as enterprises rapidly deploy increasingly sophisticated AI systems, particularly across Southeast Asia’s fast-growing stablecoin ecosystem, the risks associated with uncontrolled agent spending are becoming impossible to ignore.
“Ampersend solves a problem that did not exist a year ago,” Coelho said. “What stops an agent from spending where it shouldn’t, and who sets those rules?”
He explained that many organizations initially attempted to solve the issue by embedding spending restrictions directly into agent code or reviewing transactions after settlement. However, these methods proved ineffective because the agent responsible for spending could not reliably govern itself, while post-transaction reviews only identified problems after financial losses had already occurred.
Ampersend was created to address that gap by introducing a real-time governance layer between AI agents and payment infrastructure. Through wallet-level policy enforcement, enterprises can define spending limits, transaction caps, allowlists, and compliance requirements that are automatically enforced before any transaction is approved.
Coelho said one of the most common reactions from customers after implementing Ampersend is a sudden realization of how vulnerable their systems were previously.
“The first thing we hear is, ‘I didn’t realize how exposed we were,’” he said.
According to Coelho, enterprises often only discover the extent of the problem after an AI agent unexpectedly continues running or repeatedly triggers API calls that generate excessive compute and service charges. In some cases, a single looping workflow can accumulate tens of thousands of dollars in costs before anyone notices.
The turning point for many users, he noted, comes when they witness a transaction being blocked in real time because it violated a policy established by a human operator.
“That’s the exact moment the value clicks,” he explained. “Once you see policies enforced automatically at the wallet level, application-level checks and after-the-fact reviews stop being acceptable.”
Beyond preventing financial losses, Coelho believes governance infrastructure will become essential for regulatory compliance and enterprise-scale AI deployment. Compliance teams increasingly need proof that every AI-driven transaction was authorized and policy-compliant before execution.
“Without that, regulated institutions simply cannot deploy agents at scale,” he said.
Rather than replacing existing financial systems, Ampersend positions itself as a bridge that enables legacy enterprises to safely adopt agentic commerce while continuing to use their existing payment infrastructure.
“We sit between agents and payment rails,” Coelho explained. “The goal is to let enterprises keep their financial infrastructure while adding the control layer they need before agents touch any of it.”
For Coelho, the deeper issue extends beyond technology and into accountability itself. He argues that the AI industry is currently advancing capabilities faster than governance frameworks are being developed.
“The industry is shipping capability at machine speed and treating governance as a problem to solve later,” he said. “That order is wrong.”
He added that regulators, enterprise boards, and users are increasingly demanding systems that preserve traceable human authorization over autonomous systems.
“‘The agent acted on its own’ will never be an acceptable answer to an improper spend,” he said.
Looking ahead, Coelho believes the future of AI-driven commerce will be defined by two major shifts. The first is the emergence of universal governance layers underneath every enterprise AI deployment, similar to how businesses currently manage employee spending through approval workflows and policies. The second is the transformation of the wallet itself into the primary enforcement point for compliance and authorization.
“The wallet becomes the enforcement point for policy checks, sanctions screening, spend limits, and governance,” he said. “That fundamentally changes how the agent economy is built.”
While Ampersend currently focuses on payment governance, Coelho sees the company evolving into a broader control platform for AI decision-making by 2030. The same architecture could eventually govern software deployment, system access, email communications, and operational approvals handled by AI agents.
“The wallet becomes the enforcement point for any agent decision that requires human authorization and trust,” he said.
Ampersend is also positioning itself at the convergence of AI and Web3 technologies. Coelho noted that blockchain infrastructure already provides many of the foundational elements needed for agent governance, including programmable signing, verifiable identity, and immutable audit trails.
Importantly, he emphasized that the company remains payment-rail agnostic.
“We use stablecoins today and we’re integrating fiat next,” he said. “But the control plane remains the same regardless of the rail underneath.”
By the end of the decade, Coelho hopes Ampersend will be remembered as one of the companies that helped preserve human accountability during the rise of autonomous economic systems.
“We want Ampersend to be remembered for keeping humans in control as agents took over real economic activity,” he said.