[Interview] Alexis Sirkia, Chairman and Co-Founder of Yellow, “AI Agents Need Financial Rails Built for Machine Speed”

Artificial intelligence (AI) is advancing rapidly in the financial sector. AI systems are no longer just analyzing data and recommending actions; they are increasingly able to hold value, initiate payments and execute transactions under delegated authority. However, today's financial market infrastructure was largely built with human-paced activity in mind, not for autonomous agents that could continuously transact in fractions of a second.

Alexis Sirkia, Chairman and Co-Founder of Yellow, contends that intelligence is no longer the main constraint. Instead, the more pressing challenge is developing financial systems that can support machine-speed settlement while also ensuring secure identities, tightly scoped authorization and enforceable economic safeguards.

Sirkia expects payments and machine-to-machine micropayments to be the first areas to see practical adoption of autonomous financial agents, due to their narrow focus and low risk. Treasury management and liquidity provision will likely follow, but complex, open-ended autonomous trading will require further regulatory clarity and trust-building before it becomes mainstream.

Traditional financial infrastructure was designed for human-initiated and supervised transactions. To support AI agents that can transact continuously and at machine speed, significant technological changes are required. These include settlement systems optimized for rapid, low-value transactions with fees that scale accordingly, verifiable identities for agents to prove authorization without human intervention, and programmable settlement mechanisms that release funds only when predefined conditions are met.

Decentralized clearing and settlement systems could help manage counterparty risk and improve transaction speed by enabling collateral-backed escrow, off-chain clearing through signed state updates, and minimizing reliance on traditional intermediaries. Concerns surrounding accountability, cybersecurity, and market manipulation must also be addressed.

In the case of an error or harmful transaction by an autonomous agent, responsibility would typically lie with the entity that deployed and authorized the agent, similar to how corporate officers are held accountable for the actions of their companies. Safeguards should be built into the system to mitigate risks and ensure that autonomous agents operate securely and economically.