Agency

“Agency” – A cartoon exploring how AI agents will need to be programmed with clear mandates to balance all the competing demands, not the least distinguishing right from wrong.

Mandates are how humans delegate power without losing control. In finance, a mandate is not “make me money.” It is target 6% yield, no leverage, stay liquid, avoid sanctioned venues, and stop if volatility spikes. It is permission plus boundaries. AI agents are turning those boundaries into executable policy. Today, users already give AI simple mandates, rebalance monthly, hold 20% cash, earn yield. The next phase is agent-to-agent finance … a strategy agent instructs an execution agent to route capital across wallets, exchanges, and DeFi protocols. Every downstream action must inherit the original mandate or risk breaching financial and regulatory constraints.

Recent regulatory shifts show why mandates cannot be static. In February 2026, a part of the US Department of the Treasury signaled that many stablecoin yield structures in the United States would be presumed non-compliant, reinforcing that stablecoins are payment rails, not savings accounts. For AI wealth agents, that means yesterday’s “hold and earn” logic can become today’s risk exposure. Mandates must update as rules change. Risk policy management is not optional, it is core infrastructure.

A wave of startups is emerging to solve pieces of this problem. Custody platforms (e.g. Fireblocks) are building programmable policy engines that define who can transact, under what limits, and with what approvals. Wallet infrastructure providers (e.g. Safe) are embedding rule-based transaction controls directly into smart accounts. Most focus on execution security or post-transaction oversight rather than mandate-native enforcement at the intent layer.

Thirdfy, a startup based in the EU, governs how AI agents move money. It encodes user mandates and policy constraints, then validates every agent action before execution, ensuring capital moves only within authorized parameters.

As Andrew Ng, AI pioneer and co-founder of Google Brain, has said, “AI will automate the tasks you don’t want to do, and augment the tasks you do, but we need governance that ensures it does the right thing.” When agents transact at machine speed, mandates, not intuition, must govern their actions.