The Copilot Paradigm: Should Blockchain Be Made Fit for Humans or Just Be Left Alone?     

With the advent of agentic artificial intelligence, the debate has taken a different turn. Should blockchain actually be made more accessible, or left to its own devices, conducting financial business autonomously and with dazzling efficiency while common mortals do what they do best: go fishing?

"Blockchain developers have not been very good at designing systems for regular humans," says Marc Vanlerberghe, chief strategy and marketing officer at the Algorand Foundation. Crypto has always struggled with the challenge of "how do you cross that bridge?" he said in an interview.

Now, rather than simplifying crypto for people, a growing cohort of system engineers believes the next financial system should be designed for artificial intelligence agents instead.

Recent improvements in AI-agent capabilities bring with them a hope that autonomous financial products, operated with a few sentences of natural language, are within reach. Agents can now browse the web, call application programming interfaces (APIs), execute code and interact with smart contracts in a single workflow. Instead of operating as simple chatbots, they are increasingly turning into autonomous software entities capable of making decisions and carrying out financial transactions with limited supervision.

"They could kind of humanize and talk to you in a normal, human language and not have a bunch of technical jargon," said Luke Youngblood, the maker of Moonwell, a lending app that uses AI agents. The agentic economy, for its supporters, brings with it a new way of interacting with the financial system – one that is more intuitive, integrating responsive, hyper-personalized customer service with the financial products needed to boost an economy. 

To its detractors, an agentic-based financial edifice is a disaster waiting to happen, with hackers and bad actors rigging a machinery prone to manipulation and hallucination, and all too happy to go down a rabbit hole. Humans will always need to have the 'off' switch within reach.

Financial copilots: an irresistible business case?

Agentic AI will help groups that have so far been marginalized get access to financial services, says Eric Piscini, CEO of Hashgraph. He draws a parallel with Kenya, where the advent of mobile phones opened up banking services to women, a demographic left behind until then. This allowed many to become economically active or start a business. 

"The same thing will happen with AI agents because AI agents will remove the barrier to entry for consumers to engage with financial services," he said. "You're mixing a consumer facing app with a backend infrastructure. Those two things collide and benefit from each other. The business case is going to be so strong that you won't have a choice."

In this view of the TradFi future, teams of AI agents will work together as 'financial copilots' for the average consumer. Acting without the biological constraints of humans, they work 24/7 and process millions of data points. "If agents work and achieve even a fraction of their potential, they make our lives easier because they do the things that we have to do but don't necessarily enjoy doing," Lane Rettig, former head of research at NEAR Foundation told Sandmark.  

For consumers, this could mean finding the best rates for borrowing, optimizing a personal wealth portfolio and helping them keep to saving goals, much like a financial advisor. For businesses, it can make systems more efficient, opening up the possibility for new business models. 

Ripping up the rails

When it comes to designing a financial system for autonomous agents, the existing rails have to be replaced by others newly built from scratch. For instance, identity verification based on biometrics no longer works. While this takes extensive input from software engineers and security experts, it also removes a hurdle that has kept many consumers locked out, like those that don't have the right documentation in place. 

Crypto was built to allow for anyone to create a wallet and conduct transactions as long as they have the right keys. Its permissionless nature makes it ripe for agentic adoption. "Blockchain and crypto allow us to kind of give our agents payment rails that don't require them to open up a bank account for the agent itself," said Moonwell's Youngblood. "It opens us up to a world where people might have multiple agents in the future doing different things for them."

Building the agentic financial system on blockchain adds an infrastructure to support these functions. With the potential to facilitate speedy cross border payments and operate with low transaction costs, it can create an environment that traditional financial systems are not designed to support. 

Micropayments, for instance, have largely been excluded from TradFi rails by the economics of transaction fees. Blockchain could reduce this cost, making micropayments a critical part of agent-to-agent settlement and opening up the possibility of pay-per-use fee structures. 

Balkanized communities, hallucinations and AI malice

Crypto exchange operator Coinbase introduced the x402 protocol in mid-2025, in an attempt to support such agent-to-agent payments. The Algorand Foundation is one of its most important ecosystem partners and formally joined it in February 2026.

Although promising, it has its limitations. Currently, said Algorand's Vanlerberghe, the merchant and the shopper have to be in exactly the same 'context' for a payment to go through. This requires both to be on the same chain and in the same currency for every step of a transaction. 

The x402 protocol groups multiple transactions together, allowing for processes like swaps. Despite this, Vanlerberghe said, the environment of "balkanized communities" could restrict the agentic economy from gathering traction. 

"If agentic commerce is to take off in a big way on blockchains, it's going to take more than one blockchain to be able to handle the traffic," he said. "Interoperability is going to be absolutely critical."

The existing frameworks of the internet itself are set up to stop automated attacks, creating a hostile environment for agents to operate in. Many online financial platforms lack APIs that can connect with an agentic workflow. Although agents are learning to respond to authenticators like CAPTCHA, these systems are set up to recognize human behaviour, rejecting any interaction that seems automated. 

Building trust is therefore a critical part of designing products for the agentic economy. Agents will need to shed their vulnerability to hallucinations and abuse by malicious actors, for instance by the injection of spurious or manipulative prompts. 

Gaming the algorithm

In January, Surf AI and Princeton University published research in which 17 leading AIs were tested for their ability to make sound financial decisions in a real-world environment prone to negative influences and attacks.

The investigators found that AI models achieved only 12–28% accuracy on tasks that entry-level human analysts handled routinely. With full access to specific tools, even the best AI performer reached 67.4%, far worse than the 80% human baseline. Agents with access to detailed on-chain data models still fell back on generic search engines to solve research questions 55.5% of the time. They were misled by SEO-tweaked misinformation and social-media manipulation even when correct answers were directly accessible through purpose-built tools. 

Separately, the University of Georgia exposed vulnerabilities in Google's agent payments protocol, AP2. Researchers demonstrated that prompt injections could trick agents into sharing sensitive data of users or cause them to buy specific products or services. They warned that agentic financial systems were "fundamentally unprepared" for an environment full of potential hackers. 

In the short term, those working on agentic-AI finance systems see blockchain as a way of at least keeping track of any mistakes. An agent’s every move is logged, providing a forensic trail that shows exactly where it has gone rogue. 

"We can't prevent hallucinations from happening,” Vanlerberghe acknowledged. "But what we can do is make sure that when it happens, there's auditability and there's transparency, so we can identify it and then potentially correct it."  

This doesn't, however, prevent the error from happening in the first place. 

Know your customer, the agent 

Elements that could help AI agents be more credible include the construction of identity protocols and standards that pertain to the agents themselves, said Rettig of the NEAR Foundation.

"In the case of humans, you have KYC, you have passports, identity cards, identity numbers," he said. "Obviously, that will not be the case for agents. They need cryptographic identity."

Rettig said that currently, the setup for every agent is bespoke, requiring the manual creation of a number of APIs. Agents often have to revert to human oversight to prompt or create specific actions. "All this process needs to be made more friendly for agents," he said. "Once an agent has a digital certificate, a public key, an identity that's kind of globally accepted and recognized, then they will be able to act more autonomously on my behalf."

In early February, the Ethereum foundation launched ERC-8004, a protocol standard that will allow agents to create reputations onchain. While this is still restricted to agents running on the Ethereum network, the National Institute of Standards and Technology, launched an initiative to create a more globally recognized set of standards for AI agents.

Empowering or unsettling, there's a choice to be made 

The more urgent question, therefore, will soon become less about whether agents can act on behalf of humans, and shift to deciding to what extent they should be left to do so alone.

"We are on the edge of agents being able to transact and exchange value without humans in the loop," said Piscini. "That's happening today. That's why it's very urgent to actually put frameworks in place so that we don't let those agents do whatever they want, we don't let them exchange value on our behalf without us being in the loop."

Smart contracts could enforce spending caps, whitelist counterparties or require multi-signature approval for large transactions. Reputation systems could score agents based on past behaviour, while cross-chain identity standards could help agents recognize trusted peers. But each safeguard introduces friction and nudges the system back toward human oversight.

Ultimately, the reach of robotic autonomy may be a matter of personal preference and trust. "There's a wide spectrum of people that want control and then people that might be willing to give up control," said Youngblood. "We need to build up a lot of guardrails to protect people if the human is completely out of the loop."

Supervising an algorithm

While crypto's original ambition was to remove financial intermediaries, the agentic economy may replace them with algorithmic ones, shifting humans from active operators to supervisors of increasingly autonomous digital counterparts. Whether that trade-off feels empowering or unsettling may determine how quickly this machine-native economy takes hold.