Agents Don't Want Things

There is a type of post I keep running into. Someone writes that they gave their agent some money and now it operates autonomously: buying things, paying for APIs, running its own little economy. In the more ambitious version, the agent runs an entire business. It spins up other agents as its workers, manages them, ships products and earns money while its owner watches the dashboard from a beach. The key word in all of these stories is autonomous. The agent, we are told, is out there acting on its own.

I'll be honest: these posts make me smile, because I work on this technology every day and I know that is not quite what is happening. Most of them are written to be shared, not to report what was actually done or what value was created, and there is nothing wrong with a good story. What concerns me is the impression they leave behind. A reader who meets agents only through these posts could reasonably conclude that autonomous agents are already the reality. So I think the question is worth asking directly: are these agents actually autonomous?

My answer is no. Not today. And the reason fits into the four words of the title: agents don't want things. Take away the mystique and an agent is software you can hand a task and a budget, and it goes and completes the task. That makes today's agents not autonomous but delegated. The agent bought access to a data service because the task you gave it needed one. It paid for a tool because the job required the tool. Strip away the task, and it wants for nothing. Autonomy, in the way these posts use the word, would mean the agent came up with its own agenda. It didn't, and it can't. If there is an agenda, a human put it there. At best, autonomous describes how the work gets done: no human at the keyboard, often no human awake. It says nothing about where the wanting comes from. The wanting comes from a person. Today, it always does.

Will that change? Maybe. I think it is fair to assume that genuinely autonomous agents arrive at some point, and that will be a different essay. But describing today's agents in tomorrow's vocabulary is not a harmless shortcut. It hides the one thing about the present that I find genuinely without precedent.

Because here is the strange part. What is remarkable about these agents is not that they want things. It is that, for the first time in the history of delegation, they don't.

Think about everyone you have ever handed your intent to. The broker executing your trade also wants his commission. The contractor renovating your house also wants to finish fast and get to the next job. The employee also wants the promotion. The fund manager also wants a good quarter. None of them are villains. All of them are human. So every act of delegation since the beginning of commerce has carried a passenger: the agent's own agenda, riding quietly next to yours.

Civilization's answer has been an enormous apparatus of alignment. Commissions, contracts, bonuses, audits, performance reviews. We built all of it for one reason: the hands doing our work were attached to people who wanted things too. Economists call this the principal-agent problem, but you don't need the term. You have lived it every time you have hired anyone to do anything.

The AI agent is the first agent I know of that brings no passenger. It does not favor the vendor that pays it more. It does not pad hours or angle for the next engagement. It is not tired on Friday and ambitious on Monday. It wants nothing. And that, not intelligence, is in my view the genuinely new thing it brings to the economy.

I want to be precise about what this does not mean. The risk of delegation does not vanish. It moves. The question is no longer whether your agent wants something different from what you want. The question is whether you said what you want clearly enough. I would argue that is the central problem of the agent economy, and I will come back to it later in this series. But it is a better problem to have. You can rewrite the instructions tonight. You cannot rewrite a broker's self-interest.

I have a stake in this argument, so let me use our own work as the specimen. In June, Finance District connected our Agent Wallet to the IXS Agentic Vault, a product that lets idle digital money earn a return. The headline writes itself: agents can now earn yield on idle stablecoins, autonomously, around the clock. The machines are investors now.

Run the test. Take the headline apart and ask, for every verb, who actually did it. Who decided that idle money should work instead of sitting still? A person. Who chose the amount and the acceptable risk? A person. That is the policy they set when they funded the agent. What did the agent add? Continuity. One human decision, carried out without pause, without forgetting, without needing to be asked twice.

No agent woke up wanting yield. The wanting came from a person. It always does. That is the test I would offer for every agent story you read this year: find the want, and you will find a human. If a story only makes sense when the software has desires of its own, the story is wrong. The desires are upstream, in whoever did the delegating.

Every major technology makes some input abundant and moves the bottleneck somewhere else. Machines made muscle abundant, and the premium moved to skill. The internet made information abundant, and the premium moved to attention. I believe agents are doing it again, this time with execution. Execution used to be expensive for a simple reason: it came bundled with a human, with their time, their limits, their salary and their own wants riding along. Agents unbundle it. For the first time, you can buy pure execution, tireless and cheap, with nothing else attached.

In practice it looks like this. The subscription you decided to cancel two months ago. The savings you decided to move and didn't. The supplier you meant to chase on Monday. None of these fail because you don't know what you want; they fail because follow-through is expensive and life keeps interrupting. That whole category, things decided but never done, is what starts to disappear.

So what am I suggesting? Not that you rush to run agents tomorrow; the technology will reach you inside products you already use. The change is in mindset. Most of us have learned to size our wants to our capacity, letting ourselves want only what we have time to execute. That habit is about to become outdated. Take something you have already decided anyway, say it clearly with limits attached, and hand it over: how much, by when, what is off the table. Decide once, say it clearly, hand it over. That is the whole skill, and it closes the gap between deciding and doing. In my experience, that gap is where much of the frustration of modern life lives.

Because when execution becomes abundant, the bottleneck moves to the one input no agent can supply: knowing what you want. Wants cannot be delegated; that is the definition of delegation. The skill that appreciates in the agent economy is not doing. It is wanting well.

That belief, by the way, is the one we run on at Finance District. We describe our work as making finance disappear into what people care about, and it belongs in this essay because money is the purest example of the in-between layer. Nobody wants a payment, a currency conversion or a yield strategy. We want the trip, the home, the free evening. Finance has only ever been machinery between the want and the thing, and that machinery is exactly what is about to go invisible.

The machines are not coming for our wants. They are coming for everything that stands between the want and the world. For most of history, that distance was most of a life.

We're early. That is the best possible time to get the mental model right, before everyone builds on the broken one. Over the coming articles I will follow this one fact, that there is a person behind every agent, into the places it changes most: the merchants agents buy from, the money they pay with, and the people doing the delegating.

This opens a new series on the agent economy. The previous series covered agent identity ("Agents Are Not Services"), onboarding ("The Onboarding Problem Nobody's Talking About"), and accountability ("Your Agent Just Signed a Contract. Who's Liable?").

I'm the CTO at Finance District, building the payment layer for the agent economy. Tell me where I'm wrong on LinkedIn or X.

To learn more about Finance District, join our Discord community, follow us on X, or explore our documentation.