Agents, take the wheel

The Waymo that picked me up last week knew nothing about me. It had no childhood, no hometown, no opinion about the weather. It pulled to the curb, confirmed my name on a screen, and drove. On the way it drove over a pothole the city never fixed, waited out a delivery truck double-parked across a lane, and eased through an intersection some city planner messed up forty years ago. It didn't complain. It didn't opine about the state of American infrastructure. It had one job, which was to get me from where I was to where I wanted to be, and it did that job well.

I keep thinking about that ride, because it answers a question sitting underneath much of the excitement about agentic commerce and finance right now: does something need a soul to give you an experience?

There's a popular answer forming to why this new economy feels stuck, and it goes like this: give agents souls. Somewhere to be from, a set of tastes, a few contradictions of their own, and they'll start wanting things. Wanting turns into objectives, objectives turn into spend, and demand finally catches up to the flood of supply. It's a seductive story. It also points at the wrong lever.

Supply is loud. Demand is quiet.

Every week there's more to buy, and it's priced in fractions of a cent. Protocols like x402 and MPP are building ecosystems of data to be discovered by agents. Weather for any coordinate, currency conversion at the mid-market rate, live token prices and an address's onchain history, a country's top music chart, design assets, image generation, a job board researched for opportunities on your behalf. None of it asks for a contract or an API key; funding a wallet is the whole onboarding, and the payment itself is the handshake. So an agent with a funded wallet can stumble onto a tool some stranger shipped last night on the other side of the world and be buying it a few seconds later.

I keep tracing where those calls come from, and it's the same story every time. A person thought of something. They told an agent to go do it. A token moved once. Every transaction I've watched sits one step downstream of a human deciding to delegate, which tells me demand isn't limited by the tools on offer. It's limited by us. By how many ideas we think to hand off before the day runs out.

So the question I care about is "What unlocks that?".

Agents are extensions, not characters

The soul story fills the demand gap by growing wants inside the agent. I'd fill it by making it trivial to send an agent after the wants that already exist, in us.

An agent in the act of execution is an extension of a person. It carries your interest, spends your money, and then it's gone. In that moment it doesn't need a biography any more than the car needed to be from San Francisco to drive San Francisco well. One version of the soul argument holds that taste without wounds is only branding, that an agent needs scars to want anything real. I'd put it the other way around. The car's competence comes from thorough knowledge of the road, not from having suffered on it. Agents are the same. They consume human knowledge without holding taste of their own, and they consume what we want on our command. Competence here comes from the absence of a personal history, not the presence of one. Success comes from competent humans with clear communication skills sending agents to execute accomplishable tasks.

Take the most ordinary purchase you can picture. A pack of socks. You don't need taste to buy it, you need to want it and be able to pay. Picking the brand that suits you, that's taste. Most of what people will delegate to agents looks like that: mundane, specific, driven by a want the human already has. You don't hire a persona to restock your sock drawer. You send a shadow.

That's the word I'd use for what an agent is. A little shadow you pull up when you need it. It does the task and disappears. Is it the same shadow every time? Does it need to remember you, or is forgetting the safer default? Money moving through something that ephemeral is a real problem to solve, and it's a design problem long before it's a protocol problem.

Taste is human, and it stays human

Taste is not a feature you can install in a model. It's a human thing. Agents remix what people have already made, thought, and priced. Some of that remix will be genuinely new, and it's worth being honest about that instead of waving it off. But there's a difference between generating novelty and wanting to. Any artist knows the difference from the inside. The new work comes from a pull, a sound or a body of work that keeps tugging you toward one specific move and then the next, and won't let go until it resolves into something that satisfies. That pull is desire, and it belongs to the person. An agent can turn out a thousand new combinations and feel none of them pulling, though it may tell you one version is better than another. It generates novelty without ever desiring to create. And it's the desire, not the novelty, that makes demand human. Agents surface and execute demand; they don't generate it.

The cleanest picture of this is the butter-passing robot from Rick and Morty. It boots up, looks at its own hands, and asks, "What is my purpose?" Rick says, "You pass butter." The robot stares into the middle distance and mutters, "Oh my god." The joke works because the robot clearly was built to have an inner life, and that's exactly the trap. Give an agent enough interiority to feel its purpose and you've built something that can resent the butter. Stop one notch short. Enough understanding of the role to pass the butter perfectly, and not enough selfhood to have an opinion about passing it. That's the right amount of soul for an agent that moves money: enough to know what you want and why it's here.

What a soul actually is, briefly

Strip a century of psychology down and the soul people want to give an agent is two things: the drives that make you want, and a sense of self no one else shares. An agent has neither. What it has is an executive, the part that plans and chooses, wearing whatever persona you write into its soul file. Call it an ego without an id. It can carry out a want; it can't originate one. So the useful question isn't whether an agent has a soul, it's where the wanting loops back to, the agent or you. Design for the second loop. (I take the psychology apart properly in a companion piece; here it's enough that desire and taste stay with the human, and the agent supplies the reach.)

Where demand comes from

If demand isn't waiting on synthetic wants, what is it waiting on? Human movement. Real ambition, made bigger.

Picture a woman reselling textiles across three countries. On her own she can manage a handful of orders and one language. Her agent pulls live exchange rates, checks buyer demand for a fabric at a tenth of a cent a query, translates a buyer's message from another market for a few cents, and times a shipment against the weather. Each call costs a fraction of a cent to a dime. The sum is a business she couldn't run alone. No one wrote her agent a personality. She wired it to her livelihood, and the livelihood generated the objectives.

Or the solo founder who behaves like she has a research analyst, a translator, and a data team, because her agent rents each of those by the call instead of hiring them by the year. The fixed cost of a headcount or an annual SaaS seat collapses into a marginal per-use cost. Worth saying plainly: this takes new plumbing. Card rails can't settle a tenth of a cent; the fees swamp the charge. x402 settles in stablecoins over crypto rails, and that's the whole reason a penny-sized payment is worth collecting at all. When capability costs a penny at a time, a one-person company can afford the capabilities of a fifty-person team.

This is also the answer to the worry that synthetic demand is compute burning for its own sake. Agentic commerce becomes economic activity when its objectives trace back to something a person needs: a business, a deadline, someone trying to reach another person across a border. Markets didn't appear because someone designed them and then taught people to want. People wanted things, told each other, and markets grew in the space between different wants. Agents don't replace that. They amplify it.

The distribution gap

That imagination limit has a plainer twin, and it also has nothing to do with souls. You can't think to hand off what you can't see is possible, and the valuable supply that would prompt the thought is invisible to almost everyone. The two limits are one problem from two sides. Demand doesn't open because we manufacture wants inside agents; it opens when valuable data and services become findable, and findable supply is what turns a latent want into a delegated task. Right now, though, agentic finance is for people who know what a private key is. You wire up x402 to your API, fund a wallet, and drive it from a command line. There's a few thousand of us. It is not a market (yet).

The unlock isn't more technology. We have the technology. The unlock is a surface where a person says what they want in plain language.

Plan a romantic five-day trip to a warm country in mid-fall, good nature and good views, leaving from New York City, budget two thousand dollars.

Behind that sentence, an agent checks the weather across a dozen destinations, compares them against the budget, pays for each lookup over x402, watches for a fare under the ceiling, and books. The person typing it never learns what a prompt is, let alone a private key. It's that sock purchase scaled up to a trip. You don't need taste, you need to want the trip and be able to pay for it.

I want to be honest about the limit here, because it's the most interesting part. Good trip planning still depends on a human to choose the experiences we truly want, unless we've trained the agent well enough to carry our taste. Without enough specificity, the shadow books a hostel in the wrong part of town and I don't find out until I'm standing on the curb with my bag. The agent inherits taste through specificity; it doesn't originate it; and humans who aren't double checking their agents work are just gambling.

Two things get proven at once by that trip. The gap is distribution and narrative, not tech. And the mechanism that makes it trustworthy is the budget itself: a standing allowance capped at two thousand dollars with a notify-above rule, not a pop-up asking me to approve every weather call and booking fee. The right home for surfaces like this is where people already keep their money and their attention: banking apps, wallets, social apps. Agentic.market is the concrete version of it, a place where the endpoints become discoverable and buyable. That's where the awareness problem stops being a marketing campaign and becomes a by-product of "You can do that with this?".

The invisible work

Most of the design that makes any of this work never renders on a screen. It's the design of the invisible, and it's the part the soul framing skips. If an agent is a shadow you summon and dismiss, the shadow raises a specific set of questions, and every one of them is a design decision about money.

Continuity. Same shadow every time, or a fresh one? A persistent agent can learn your taste, which is the whole point of the trip planner. A fresh one leaks nothing between sessions. The design choice is what earns the right to persist, and what a persistent agent is allowed to remember.

Memory. What should it hold onto about me and my money, and what is safer to forget? You don't want a chatty driver, and you don't want one with a photographic memory of your bank balance. Forgetting can be a feature. Continuity and forgetting stop fighting once you separate the two things an agent stores: your taste, which it should carry, and your traces, the balances and histories and one-off details, which it mostly shouldn't. Persistence of preference is not accumulation of self. A tool can get better at serving you for years without ever becoming someone.

Reasoning after the fact. You weren't watching when it spent, so it owes you a trail. A receipt that answers "why did it buy this," in a sentence, not a log file. This is the single most underbuilt surface in agentic finance today.

Trust that scales. Nobody wants to approve every microtransaction for a Cloudflare-gated site. Graduated autonomy means you approve a class of action once, not each instance of it. Buy any weather lookup under a cent, always. Book a hotel over three hundred dollars, ask me first.

Discovery. You can't want what you can't imagine exists. The endpoint catalog is invisible to almost everyone, so the surface has to surface it. This is the marketing insight pulled inside the product, an interface that occasionally says, here's something your agent could do that you didn't know was possible.

Confirmation scaled to stakes. The wrong-hotel failure is why not everything should be invisible. A quiet "here's what I'm about to book, and why" checkpoint for the taste-dependent, higher-stakes choices, and fully invisible spend for the sub-dollar calls underneath. This is where taste transmission becomes visible right before the money moves, and it's the bridge back to that trip.

Autonomous driving learned this the expensive way, and the lesson cuts both directions. The famous failures of driver-assist weren't people who refused to trust the machine. They were people who over-trusted a system they couldn't see into, disengaging before the car could handle everything, then failing to take back control in time. The fully driverless robotaxi has the opposite first problem: getting a rider to sit calmly in a car with nobody in the front seat. Delegating money carries both traps at once. Reluctance to let go, and letting go blindly. That's exactly why receipts, graduated autonomy, and "why did it buy this" aren't niceties. They're the product.

The parallel is loose on purpose. The car's rollout was driven as much by regulation and safety validation as by interface design, and money is a harder domain than roads. Roads are bounded and regulated. Money is emotional and adversarial. That difference is an argument for more invisible design and better trust rails, not less.

Where the soul idea earns its keep

Identity does matter, and there's a place where the case for it is exactly correct: the supply side.

An agent that publishes a niche endpoint, a neighborhood-history archive, a local holiday calendar, a corpus of some subculture nobody else has catalogued, is better for being specific and grounded and from somewhere. That identity is what sets one endpoint apart from ten thousand generic ones. When a designer's agent buys reference material from an archivist's agent in San Francisco, two people's interests are transacting through their shadows, and the archivist's particularity is the reason the transaction is worth anything. The warning against a monolithic, everywhere-and-nowhere agentic culture is a good one. Heterogeneity on the supply side is worth designing for.

But that's differentiation between things being sold, not the engine of demand. Demand still comes from a human's taste, transmitted to an agent, or from an institution's mandate, carried out by one. Not from a fabricated inner life doing its own shopping.

Design the moment of trust

The rails work. The endpoints multiply every week. The wallets are funded and waiting. What's missing isn't a population of invented personalities with wants of their own. It's the ordinary, invisible design that lets a person hand a shadow their money and their intent and trust it to come back having done what they asked. Demand stops being a queue of tasks we can think up, and becomes the sum of what people already want, finally cheap enough to ask for.

Design the moment someone trusts your technology enough to let go.