AI models have become incredibly powerful and capable. You may have heard of agentic commerce, which refers to AI agents that can shop, spend, settle invoices, trade stocks and crypto, and mu
AI models have become incredibly powerful and capable. You may have heard of agentic commerce, which refers to AI agents that can shop, spend, settle invoices, trade stocks and crypto, and much more.
What’s missing is the answer to a very simple question: When an agent screws up, who is responsible?
Rodrigo Coelho, CEO of Edge & Node, told TheStreet Roundtable that governance, not capability, is the wall between AI agents and real money. He believes that enterprising giving spending power to agents is inevitable, but only after the accountability chain is settled.
"When something goes wrong, someone's gonna be responsible. It's gonna be a human at the end of the day,” he said.
Coelho’s conviction was born from sales calls. Edge & Node sells data tools to financial institutions.
“Naturally the discussions would eventually talk about AI and what they're doing internally," he said.
What he heard, over and over, was the same thing, and it wasn’t a technical issue.
"Where it breaks down is following the existing policies and procedures, because you typically have humans that authorize and execute," Coelho said. "When you go to an agent, one of the main questions is: okay, who's responsible? Is it the company itself? Say the agent does something wrong — is it the developer that's responsible, or is it the human that authorized it?"
Corporate procedures have an answer for every human mistake. Someone had to approve the decision, someone actually executed the work, and someone signed an incident report. An autonomous agent breaks that chain.
This raises another question for businesses integrating AI into their workflows. How does anyone even know which agent is acting, and for whom?
"How does an agent identify itself when it's going out in the world and acting on behalf of an organization or a person?" Coelho said. "There are now identity systems being developed as open standards... but there's no real universal standard as of yet. I know digital ID has been discussed in India and other countries, but there's no universal standard yet."
That gap is where many of the emerging agent-payments stack is being built. Open protocols for agents to authenticate themselves and settle in stablecoins. It’s similar to the travel-rule issue crypto is facing for institutional finance, except the party being identified isn’t a person, it’s software.
Until those answers exist, adoption runs through three offices that have nothing to do with model quality.
"At the enterprise level, a chief legal officer, chief security officer, chief financial officer — these are people that ultimately have to sign off to say they feel comfortable implementing these systems within their businesses," Coelho said. "That's really what's the impedance right now."
Coelho explained that despite these concerns, institutions are moving forward with research and experimentation.
"Financial institutions move very slow," he said. "Typically internally there's an AI team and maybe a blockchain team, and they're starting to do experimentation — starting to look at how they might improve processes, implement these solutions to save money or be more streamlined. What's happening now is sandbox environments, proof of concepts, very limited controls, limited spend."
The path to more integrated agents, as he describes it, will not happen overnight. It will be a slow loosening of the leash as executives become more comfortable with the technology and appropriate guardrails are introduced.
"Maybe there's a human authorization now, but eventually they'll become fully automated — after enough time passes and enough comfort is built and the rules are solidified that they can be turned on and left to run autonomously," Coelho said.
Stablecoins firms have poured money into agent payment rails in 2025 and 2026, a category of commerce that Bloomberg has said “barely exists.”
In light of this, some may argue that the space is over the tips of its skis in this buildout, but it can also be seen as validation of Coelho’s thesis about where agentic commerce is headed: it is not necessarily the smartest agents that will win going forward, but the most accountable ones.
