Intent and Uncertainty: What Happens When AI Goes Shopping?

How agentic commerce and the tokenization of intent will impact who you buy from, when you buy, and how much you pay

QUICK NOTE: Jamar’s Newsletter is now “The Long Build by Jamar Parris” with new posts 2x a month.

---

Imagine you tell your AI agent to automatically buy that TV you have wanted for months as soon as it falls below $1,500 and to pay with your debit card.

You have expressed your intent, but unlike a traditional checkout experience there is still a great deal about the purchase that remains undefined. You do not yet know which merchant will fulfill the order, what the final charge amount will be, or exactly when the purchase will take place.

Traditional payments have typically been categorized into two buckets: Card-Present, where you’re physically at a point of sale, and Card-Not-Present, where payment data is captured remotely on the web or in an app. To me, this progression to agentic commerce seems like a brand new category: what Google, PayPal, Coinbase and others are calling Human-Not-Present¹.

This payment type introduces three key uncertainties: which merchant you’ll transact with, when the transaction will execute, and what the exact amount will be. It assumes that you won’t need to be there to approve the final transaction at the moment it happens. We’re not there yet, though.

Even recurring subscription payments, which some might view as Human-Not-Present, feature more explicit user authorizations, providing a level of certainty absent from open-ended tokenized intent.

Consumers and B2B buyers are complex. There’s often a gap between what we intend, what we communicate, and what we actually do. For instance, you might tell your agent ‘buy the cheapest option,’ but actually prefer supporting a local business if the premium is minor. Or you state a price limit, but if your favorite color goes on sale just above that threshold, you’d want it anyway.

Purchases are about so much more than price. Logistical factors like cash flow timing, and personality factors like emotions and values can play a big role in us choosing what, when and how we buy, as well as the specific brands and merchants we want to support with our money.

This complexity is precisely why different merchants will react to agentic commerce in different ways. Companies like Walmart², selling staples with relatively predictable demand and intent, are embracing it and it’s not hard to see why. Imagine shift and gig workers being able to automatically buy groceries the second their paycheck is deposited.

Or the potential for food waste reduction, where they could incentivize purchases of perishables with discounts right before they spoil, a sustainability win mutually beneficial for both the merchant and consumer.

On the other hand, for discretionary spend, like travel, where demand-based pricing already means we often pay different prices, I could see agentic commerce leading to mitigations to protect their bottom line. They might limit agent purchases to only their most restrictive fares, preventing AI agents from constantly rebooking whenever fares drop, and us pocketing the difference.

The irony? These same companies will likely embrace agents for trip planning and recommendations, which drive revenue, while attempting to block the transaction optimizations that could benefit consumers.

In the US, I’ve long held that we focus more on earning rewards from spending than returns from saving. Merchants will exploit this, using loyalty programs to encode brand preference into your intent profile before the agent starts shopping. This enables them to avoid commodification and makes loyalty a key battleground in the fight to shape our digital intent.

These loyalty programs themselves are likely to become more tokenized, enabling agents to auto-redeem rewards and optimize card selection, providing both expanded utility for users and new ways for merchants to keep you locked into their ecosystem.

The travel sector already demonstrates this strategy. Major hotel chains won’t provide you with loyalty perks like free breakfast or that room upgrade, unless you book via their desired channels. You could see more industries following this playbook.

But this automation at scale also introduces new challenges. As reported by the WSJ³, Apple Card’s synchronized billing, where over 12 million cardholders receive their bills simultaneously each month, creates a tradeoff: users love the predictability of the end-of-month due date⁴, but it creates concentrated spikes in demand for Goldman’s customer service teams and systems, causing problems not encountered with staggered due dates common elsewhere.

Now imagine a future where hundreds of millions of AI agents are all trying to execute transactions simultaneously in a similar bursty manner, whether buying concert tickets the moment they drop, or processing payroll-linked purchases every other Friday.

Even longer term, I can’t help but wonder if this shift could accelerate the drive towards tokenization of real world assets. As users get accustomed to digital representations of themselves, their preferences and their intent, why wouldn’t that also expand to the actual goods, services and assets they’re actually purchasing? From something as simple as a record of the expiration date of the milk in the fridge to something as complex as the deed to the house they live in.

This progression is happening, even if still nascent. As I explored in my recent post on zero-knowledge proofs, we urgently need to pair this future with privacy-preserving technologies that let us share this type of information, without exposing our complete intent profile to merchants and platforms.

There’s also the question of who actually wins in all this, and who gets left behind?

---

1. Google

2. Walmart

3. WSJ

4. Apple