A Singapore regulator just asked who’s responsible when your AI agent does something wrong. The answer is obvious: you are. You sent it, you own it. “The AI did it” will hold up in court about as well as “my assistant did it.”
But the interesting question isn’t about liability. It’s about what happens the moment more than one person is holding the leash.
The Sole Proprietor Problem
Right now, almost every AI agent has one owner. One set of instructions, one person who can modify them, one throat to choke if something goes wrong. This is the sole proprietor model — simple, clean, and about to become the minority case.
Think about how many consequential things you do with other people. Business partnerships. Investment decisions. Legal matters. Family finances. The moment agents start handling any of these, they’ll have multiple principals with potentially conflicting instructions.
Who can modify the agent’s mandate? Who can shut it down? What happens when the partners disagree?
These aren’t hypothetical edge cases. They’re the founding questions of every business partnership in history. And we already have the answers — we just haven’t ported them yet.
The Governance Toolkit
Corporate law spent centuries building primitives for exactly this problem: operating mandates, voting rights, veto powers, dispute resolution clauses. A shareholder agreement for an AI agent sounds absurd until you realize it’s just a system prompt with legal teeth.
The same logic applies up the stack. Boards exist because companies became too complex for any one person to oversee responsibly. Fiduciary duties exist because the people running a company aren’t always the ones who own it. Audits exist because trust without verification is just hope.
None of this needs to be invented. It needs to be translated.
The Part Everyone’s Missing
Here’s what I find most underrated in this whole conversation: reputation.
When you interact with Goldman Sachs or Apple or your local law firm, you’re not just transacting with a legal entity. You’re transacting with a brand that has accumulated trust signals over years — through behaviour, through consistency, through how it handled things when they went wrong. You’ve never met the founders. You don’t need to.
Agents that operate persistently on your behalf — negotiating, communicating, transacting — will develop the same thing. Counterparties will learn to trust or distrust a specific agent independent of the human behind it. An agent with a track record of clean, reliable execution will open doors its owner’s personal reputation alone couldn’t.
That’s not a metaphor. That’s the mechanics of how brand equity actually works, applied to a new kind of actor.
Already Incorporated
We keep asking what agents will look like when they grow up — what governance frameworks we’ll need to build, what new legal categories we’ll have to invent.
The answer has been publicly traded for decades.
We don’t need new law. We need lawyers who’ve read the API docs.
Leave a Reply