AI is beginning to move beyond chatbots in financial services, but letting software initiate and manage transactions remains an early-stage prospect.
Money20/20’s Chief Strategy & Growth Officer, Scarlett Sieber, says firms have moved from exploring potential use cases to working out how to deploy the technology safely, at scale, and across disconnected systems.
In an interview with Invezz, Sieber described early examples of autonomous transactions, including her own test of an AI assistant.
Yet she said broad adoption is likely to take years, with the path depending on stronger interoperability, clearer accountability, and evidence that agents deliver benefits to users.
Stablecoins and public blockchains could help connect financial institutions, she said, but the wider ecosystem and regulatory frameworks remain fragmented.
Here are the excerpts from the interview:
Invezz: Financial services have been talking about AI for years. What has actually changed on the ground?
Sieber traces the industry’s current focus back to earlier experiments, including Bank of America’s Erica chatbot, which drew attention at Money20/20 in 2015.
In the first wave of recent AI enthusiasm, she said, many conversations were still about understanding how the technology worked, identifying practical use cases and estimating whether it could cut costs or increase revenue.
The discussion has since shifted toward deployment, including how to manage the risks.
“There’s a lot more that’s happening now where you are seeing real use cases with,” she said.
“The conversation is okay, we now believe the AI use cases. Now how do we do it responsibly, how do we do it at scale, et cetera,” Sieber said.
At Money20/20’s European event in June, she said, the industry saw what she described as its first live commerce transaction using AI.
The focus is now on how to use agents responsibly and prevent fraud. The shift is also visible in the concerns being raised by companies.
Roughly 30% of submissions to speak at this year’s Money20/20 US event included the words AI and fraud, she said.
Invezz: Consumers are already using AI to help with shopping and travel. How close are we to letting agents transact on their behalf?
Consumer adoption remains relatively early, according to Sieber, even as more powerful tools enter the market.
She cited her own experience testing an AI system that could review her inbox and identify services that might need updated payment details.
“So I was like, go look at my email, pull all the places that you see that I might have an issue and tell me what you think I should do about it.”
The system was able to identify recurring payments and suggest actions, but there were still limits on what it could execute.
“So I do think we’re getting there. But this is clearly a beta. It’s early,” she said. “In general, I still think we have a long way to go.”
The issue is not simply whether agents can perform transactions.
The financial system must also have safeguards capable of controlling what those agents can do.
“It’s really about the systems and processes and procedures set up in place to have the right guardrails around AI use,” Sieber said.
Invezz: What needs to change in the infrastructure before AI agents can transact safely?
For Sieber, interoperability remains one of the biggest weaknesses in the current financial system.
“The interoperability piece is still a challenge. It’s still an issue to move across countries and across financial institutions.”
She said stablecoins could help address some of those infrastructure constraints, particularly for cross-border payments.
“I do think things like stablecoins can help with that,” she said.
She pointed to a consortium of 21 financial institutions exploring a stablecoin company and said the intention to use a public blockchain was significant.
Yet even if those banks succeed, they and their customers represent only part of the financial ecosystem.
Invezz: Could stablecoins eventually become the default for AI-agent transactions?
Sieber sees a growing case for that outcome.
“It does feel like that is the direction we’re going in. There’s a lot more traction there for sure,” she said.
She pointed to growing institutional interest in stablecoins and noted that the technology is increasingly being discussed alongside AI, payments, digital assets and identity infrastructure.
“I do see stablecoins being the most likely key piece to the interconnectedness for utilization,” Sieber said.
She also highlighted the changing composition of Money20/20 discussions, with banks, payment companies, technology firms and digital-asset businesses increasingly appearing in the same conversations.
“So I do really believe that there is a real opportunity now for this to be the future, and stablecoins will certainly play a very large part in that.”
Invezz: Autonomous transactions raise obvious questions around liability and regulation. How far away are we from meaningful adoption?
Sieber is significantly more cautious about the timeline for autonomous finance than the broader AI narrative.
“We are already seeing real-time micro examples of transactions happening fully autonomously with AI, but it’s not at any type of meaningful scale. I think we’re probably still a few years away.”
She argued that financial services cannot adopt autonomous systems at the same speed as other industries because mistakes can directly affect consumers’ money and wealth.
“We’re far away from that from a financial services perspective,” she said.
“I’d say we’ve got some good early examples, but it takes time, especially when you’re dealing with people’s wealth and money. It’s going to be some time before we see any truly scalable adoption of, like, truly autonomous decisions being made.”
Regulation is another complication. Sieber said the pace differs significantly by geography, with some markets allowing financial innovation to develop before regulation catches up while others place regulators in the lead.
That fragmentation could make it difficult for companies operating globally to scale agentic financial products consistently across markets.
Invezz: What would make consumers trust AI agents with their money?
Sieber said trust is critical in financial services, but she does not necessarily view it as the main barrier to adoption.
“I don’t think the trust piece is the biggest issue because there are already examples of people trusting AI more than they probably should.”
Instead, she believes consumers will need evidence that giving an AI agent greater control actually benefits them.
“But I think it’s more around proof, right? So for it to work at scale, it’s proof that by doing so it’s benefiting them.”
That could mean saving consumers time, reducing costs, or helping them make better financial decisions.
“I can see trust coming with these AI agents doing financial transactions on your behalf as long as it’s clear that there is a benefit for them,” Sieber said.
But the system will also need clear answers when something goes wrong.
Invezz: What happens when multiple agents are involved in the same transaction?
The accountability problem becomes even more complicated when one agent is working alongside other agents, potentially across several companies.
“Complicated. Absolutely complicated,” she said. “The short answer is clear accountability and lines where a human gets involved to say yes or no.”
The industry is considering different thresholds for human intervention depending on transaction size, risk, and customer circumstances.
Invezz: What does the financial system look like by the end of this decade?
For Sieber, the long-term goal is not to remove humans entirely from financial decisions.
It is to use technology to eliminate unnecessary complexity while allowing people to retain control.
Sieber said the underlying objective is unlikely to change: technology should give people more control over their finances while reducing the number of routine decisions they have to make themselves.
“Utilizing technology within financial services to make it easier for people to conduct their daily lives where they are involved in the key decisions that really matter, and they don’t have to think about the rest.”
She used credit-card rewards as an example of a task an AI assistant could simplify.
A system that understands a customer’s financial profile could help select the right card for a purchase, rather than leaving the customer to compare reward schemes each time.
“I shouldn’t be thinking about which card I need to use to get the best price on something.”
She expects the convergence between banks, fintechs, payments firms, AI companies and digital-asset businesses to continue.
“Ten years from now we’ll see even more of that.” And the pace of change could eventually produce a more radical outcome.
“But God, ten years from now, given the pace that we’ve had in the last few years, who knows, maybe there’ll be agents on the conference stage; you know, there won’t even be people on those stages anymore, the agents telling the stories on behalf of the people. We’ll see.”
The post AI agents in finance remain years from scale, says Money20/20’s Scarlett Sieber appeared first on Invezz
