How Developers Are Solving AI’s Financial Limitations with Blockchain

One of the limitations of AI, and by extension robots, is their inability to manage transactions autonomously. To address this, developers are now integrating AI with blockchain infrastructure, allowing AI agents to transact securely, manage digital assets, and execute smart contracts.

At ETH Denver 2025, Decrypt spoke with industry experts from the Coinbase Developer Platform, AI and robotics developers OpenMind and Robonomics to explore how AI-powered humanoid robots and other systems leverage blockchain to function as independent economic agents.

“Out of the box, AI does not interact with the blockchain at all,” Lead Developer at Coinbase Developer Platform Nemil Dalal told Decrypt. “If you want it to have a wallet, send money, or transfer funds to different places, it can’t do that. It also can’t sign up for a bank account on its own. This is a major limitation because AI, in many cases, needs the ability to transact.”

Recognizing this gap, the Coinbase Developer Platform developed a framework called Agent Kit, which the company launched in November, Dalal said. Agent Kit allows AI to be given a wallet, enabling it to perform any on-chain action that a human can.

Calling it one of the earliest use cases for AI agents, Dalal noted the potential of AI to handle tasks such as monitoring the internet and conducting business on behalf of users.

“So far, over 2,000 developers have built on top of Agent Kit, contributing to over $100 million in value,” he said. “One of the most popular use cases is tracking something on-chain and executing an action in response—essentially an ‘if this, then do that’ automation.”

As AI’s influence grows across multiple industries, blockchain developers are increasingly exploring its potential for crypto traders and investors. Still, Dalal acknowledged AI hallucinations remain a problem in the industry.

“It may take actions or provide information that is not true,” he said. “When real money is involved, this issue becomes even more problematic.”

Dalal said mechanisms that give users greater control and oversight are needed to ensure AI can safely handle financial transactions without risking unintended actions.

“For instance, if a transaction is below a certain dollar amount, the AI can proceed automatically. If it exceeds that threshold, approval by the user might be required,” he said. “Similarly, if the AI is paying a new party for the first time, it may need to whitelist them, whereas subsequent transactions would not require additional approval.”

Despite these concerns, Jan Liphardt, founder of open-source AI and robotics startup OpenMind, said AI and robotics provide a unique and compelling use case for crypto.

“Blockchains were built for machines, and they’re a critical part of the infrastructure that’s needed for machines to work well with people,” Liphardt told Decrypt.

A professor of bioengineering at Stanford University, Liphardt highlighted the challenges AI-driven humanoid robots face in human-centric environments, such as banks and airlines, which remain unprepared to accommodate them.

“We’re seeing incredible amounts of friction as thinking machines try to navigate the human world, and a big part of that is a lack of identity,” he said. To mitigate this friction, OpenMind designed Iris to use multiple cryptocurrency wallets to facilitate transactions.

“She has two wallets,” Liphardt said. “One is a straight Ethereum wallet, and the other is a Coinbase wallet,” he said, adding that Iris holds the private keys to the wallets, checking the balances every six seconds.

Balancing innovation with safety

As robots become part of human systems, questions emerge about their identity, how they earn and manage money, how they interact with people, Liphardt said.

What rules govern them and how those rules can be changed are also front of mind, with Liphardt adding decentralized ledgers are well-suited to address challenges as they already support many of those use cases.

In addition to managing interaction with the real world, developers have turned to blockchain technology to enforce transparent and immutable governance rules to address concerns about humanoid robots.

“That’s why we write governance and behavior rules onto Ethereum,” Liphardt explained. “When a humanoid robot boots up, it reads those rules. If someone asks, ‘Why should we trust this computer?’ you can simply point them to the governance rules—they’re immutable and publicly accessible. Essentially, we adapted Asimov’s Three Laws of Robotics and encoded them onto Ethereum.”

While this provides some safeguards, Liphardt acknowledged it’s not a complete solution.

To alleviate fear of humanoid robots, influenced by films like “The Terminator,” Liphardt noted that medical robots are smaller and more approachable than those used in warehouses or hazardous settings.

For Vitaly Bulatov, evangelist with robotics startup Robonomics, the developer of the humanoid robot Unitree G1, blockchain technology offers resources that smaller robotics developers may not have access to.

“While we can run small LLMs on the robot itself, we’ll likely need more compute to fully enable this cognitive layer,” Bulatov told Decrypt. “In our case, we are built on Polkadot, so we serve as a gateway to Polkadot itself. We use compute providers on Polkadot to gain additional computational power for this cognitive layer.”

Bulatov said that to integrate autonomous robots seamlessly into our economy, they must be independent agents capable of managing transactions and contractual relationships.

“What we do is essentially give these robots an identity, a wallet, and the ability to enter contractual obligations with people,” he said.

The system captures data—including error messages and video feeds—each time a payment is processed. According to Bulatov, this information is recorded with a decentralized storage provider on Polkadot, allowing for verification of task completion.

“This way, no matter how large the autonomous system becomes—whether it’s millions of robots or an entire smart city—we can track that, based on payment, the technical execution happens,” Bulatov said.

Edited by Sebastian Sinclair