Every major evolution in how we pay, from cash to card and card to mobile, has one thing in common: it made spending easier. And each of those transitions sparked a measurable boom in consumer spending.
When Apple Pay launched, it didn’t reinvent payments. It shaved seconds off the checkout flow. But that small improvement had an outsized impact. According to industry studies, contactless payments increased average transaction volume by over 10% in some categories. A simpler path to checkout drove a fundamental behavioral shift.
But the next leap won’t be about shaving seconds. It’ll be about removing the act of checkout altogether.
We’re entering the era of agentic commerce, where autonomous AI agents don’t just recommend, they act. Where intent is delegated, not executed manually. And where the friction isn’t in payment, but in decision-making itself.
This isn’t just a UX improvement. It’s a paradigm shift with second and third-order effects across the entire consumer economy. Discovery, brand loyalty, platform dominance, fulfillment, payments, even what it means to “shop” are all up for grabs.
The goal of this memo is to unpack that shift:
Why it’s happening now
What the agentic shopping journey looks like
Who stands to win (or lose) as the stack reconfigures
And where the biggest opportunities lie for founders and builders
Agentic commerce isn’t just a new channel. It’s the logical endpoint of a decades-long trend: turning friction into fuel.
Defining Agentic Commerce
In traditional ecommerce, the consumer drives. You search, you scroll, you click. Even with recommendation engines and retargeting, the burden of action stays with the user. Commerce today is still pull-based where intent must be expressed manually.
Agentic commerce flips that model.
It’s not just about automation. It’s about delegation.
Agentic commerce refers to shopping experiences mediated by autonomous agents, AI systems that act on your behalf to discover, decide, and transact. Instead of clicking “Buy Now,” you say, “Get me a week’s worth of groceries,” and it happens. The agent infers your preferences, constraints, and intent. You approve, edit, or return. But you’re not executing. You’re delegating.
This isn’t hypothetical. The primitives already exist:
LLMs that can interpret broad intent and reason across options
Personalized embeddings that capture user taste and context
Embedded fintech that enables autonomous, rules-based payment
What’s new is their convergence, a shift from prompting tools (like ChatGPT) to ambient, task-executing agents that live alongside users.
“Buy me a black carry-on bag under $250 that fits Delta’s sizing.”
“Order the same kids’ snacks as last month, but replace anything they didn’t finish.”
“Find me a dress for a summer wedding that works with the shoes I already own.”
These aren’t Google searches. They’re commerce prompts. And they assume an entity on the other end that understands and acts.
That’s the key difference: traditional ecommerce starts at discovery and ends at checkout. Agentic commerce starts at intent and ends at fulfillment.
In this flow, the shopper doesn’t browse, they prompt. AI handles the upstream funnel: discovery, evaluation, even decision-making. The funnel collapses before a homepage is ever needed.
More Than Just a New Interface
It’s tempting to think of this as just a UX layer, like voice assistants with better LLMs. But that underestimates the shift. Agentic commerce doesn’t just compress steps. It changes who controls them.
The agent becomes the primary interface. The website, the ad, or the search result are no longer the starting point. They’re the downstream consequence of agentic filtering.
This has profound implications:
For brands: winning shelf space becomes winning agent preference
For platforms: the gateway moves from browser to context-native assistant
For infra: payments, fulfillment, data, and returns must adapt to new flows
The agent isn’t an app. It’s a new surface area for commerce.
The Second-Order Effects: Where the Boom Happens
If history holds, making commerce easier will unlock more commerce. The 10% lift from mobile payments wasn’t about technology; it was about reducing friction. Agentic commerce promises a bigger lift because it removes the need to engage at all.
But when that spending boom arrives, who benefits? And what infrastructure needs to evolve to capture it?
The New Gateways: Who Owns the Starting Point?
Agentic commerce will need a starting point, a place where intent is captured and translated into action. That starting point probably isn’t Google.
Search is too passive, too static. The user still does the work. Agentic commerce is about flipping the model: I express the outcome I want, and the agent figures out how to get there.
So where does this happen?
Generic LLMs like ChatGPT or Claude could act as catch-all gateways, generalist agents with plug-in access to ecommerce APIs (see Shopify’s recent announcement on what their “horse in the race" looks like).
Verticalized assistants could emerge (e.g., fashion agents, travel agents, parenting agents) with deeper domain understanding and memory.
New agent-native platforms could win by owning the agent layer itself, the place where context, intent, and execution converge.
But perhaps the most interesting model borrows from a human construct: the personal shopper.
The best human shoppers don’t wait for prompts; they proactively source items that match your needs and preferences. They curate, they suggest, they buy. That’s the “ClosetOS” model: always-on agents that know your wardrobe, anticipate needs, and optimize across fit, cost, and brand.
Discovery Gets Rewritten
If agents are doing the shopping, what happens to discovery?
Today, brands spend billions competing for attention through SEO, SEM, influencer spend, and paid social. But if the agent decides what to show the user (or whether to show anything at all), the discovery layer gets abstracted away.
This could create a centralizing force:
LLM-integrated giants like Amazon or Walmart could offer their backend product graph directly to agents via structured APIs. That would make them the default source of truth, faster, more comprehensive, and more reliable than parsing open web results.
Independent retailers who are the long tail of ecommerce could get shut out unless they plug into these closed systems.
Unless…
A counterforce emerges: an agent-optimized backend co-op, a DSP for brands that aggregates inventory, metadata, reviews, and fulfillment into a unified discovery API. The agent doesn’t care who hosts the data. It cares who responds fastest with the best fit.
The risk: if no one builds this, discoverability dies for everyone outside the platform giants.
Infra Starts to Bend
This isn’t just a frontend shift. Every layer of commerce infrastructure will need to adapt:
Payment orchestration becomes agent-native, invisible, dynamic, and rules-based.
Logistics platforms will need to expose metadata that agents care about: delivery windows, return ease, carbon impact.
Data and feedback loops get richer. Agents can collect and act on more feedback than humans ever would. You don’t ask 10 post-purchase questions. Your agent will answer them all automatically.
The winners won’t just be the platforms that power agent commerce. They’ll be the rails — payments, logistics, data — that make it feel magical.
Winners and Losers in an Agent-Mediated World
Every shift in interface reorders the stack. Google shifted power to search-optimized brands. Mobile shifted power to app ecosystems. Agentic commerce will shift power to those who can best serve machines, not just humans.
In this new world, some players are better positioned than others, and some moats start to crumble.
Platforms That Own the Product Graph Win
If agents are the new shoppers, the first question they’ll ask is: Where should I look?
The likely answer is: wherever the data is structured, accurate, and deep.
That gives Amazon and Walmart a major advantage. Their product catalogs are comprehensive, cleanly structured, and enriched with performance data, inventory status, shipping windows, and reviews.
Agents don’t want open-ended web crawling. They want a structured feed. And Amazon’s API gives them just that. If LLMs cut exclusive data deals with these platforms, or simply default to their feeds, the funnel collapses inward.
That’s great for platforms. Less so for everyone else.
Independent Brands Face an Existential Choice
If discovery is routed through agents, and agents rely on a few dominant sources, the open web becomes a dead zone. SEO, brand marketing, even paid social may stop driving meaningful traffic, because the agent doesn’t “see” those channels.
That means brands have two choices:
Plug into the closed systems, and accept the margin compression and platform dependence that comes with it.
Band together into agent-native coalitions, structured product feeds, verified metadata, pooled fulfillment, shared analytics.
We may see the rise of an agent-facing DSP for independent retail, an aggregation layer that acts as a counterweight to Amazon, just like The Trade Desk did for media.
But without it? The independents lose visibility. And eventually, viability.
Shopify’s Crossroads: RPA or Reinvention (or Both?)
Shopify enabled the long tail of ecommerce. But in an agentic world, the front end may not matter.
If agents aren’t browsing stores and are instead executing API calls, Shopify’s value shifts from storefronts to back-end infrastructure.
Originally, this created a fork:
Path 1: Shopify becomes an RPA layer, offering APIs for agents to read/write carts, access inventory, and trigger fulfillment.
Path 2: Shopify builds a headless agent interface, a unified discovery and purchasing layer optimized for machines.
The company that once said “we help you own your brand” might have had to pivot to “we help you serve the agent.”
Now, with Tobi Lutke’s recent announcement, Shopify appears to be pursuing both paths simultaneously. The company unveiled three new tools explicitly designed for an agentic commerce future:
Embedded commerce widget for LLMs, allowing shopping directly inside agent experiences
Headless product catalog optimized for structured data queries
Universal cart that can transact across multiple storefronts
This is more than a defensive move — it positions Shopify not just as the infrastructure for human shoppers, but as the API surface area for agents themselves.
Shopify is already drawing lines. As Victor Castro noted, it recently enforced new rules in robots.txt that restrict which agents can access checkout. Only sanctioned “humans” or partners using the official checkout kit can transact. This isn’t just about security. It’s a power move to preserve transaction fees and enforce walled gardens.
So, who wins?
Platforms with structured data and distribution (Amazon, Shopify, Walmart), API-native infrastructure providers, LLMs that own the user relationship, and new agent-native DSPs that aggregate the long tail are best positioned to thrive.
The agent reorders the stack, changing not just how we shop but who we shop from.
Infrastructure Shakeups: The New Ecom Rails
In a world of agentic commerce, the user interface disappears, but the backend gets more important than ever.
Agents don’t care about glossy product pages or conversion-optimized checkout flows. They care about structured data, clean APIs, clear rules, and real-time signals. Which means the stack that powers commerce must shift from being human-optimized to agent-native.
Let’s break it down.
Payments: From Frictionless to Invisible
For years, payment innovation has focused on removing friction, with Apple Pay, Shop Pay, and one-click checkout. The agentic era takes that to its logical endpoint: no checkout at all.
In agent commerce, payments become:
Invisible: Auth happens in the background, based on rules and past behavior.
Rules-based: “Never spend over $500 without confirmation”; “Always use Amex for travel”.
Atomic: Agents need real-time feedback, including success or failure, fraud, and delay, to adjust flows.
This benefits infrastructure players like Unit, Lithic, and Alloy, orchestration engines built for flexibility, composability, and API-first execution. The winners will be those who can support complex logic without introducing latency.
Fulfillment: Logistics Meets AI Metadata
When an agent buys something, it doesn’t just want to know if it can be delivered. It wants to know:
When it will arrive
How likely it is to be returned
How carbon-intensive it is
Whether it can be bundled with other purchases
Today, most of that metadata isn’t exposed. In agent commerce, it will be mandatory.
That creates opportunity for next-gen logistics players like Portless, PDQ, and others who can serve as intelligence layers between warehouses and agents. Not just shipping, but predictive delivery orchestration.
The future isn’t “2-day shipping.” It’s “2-day shipping with a 92% likelihood of fit and 0% return friction.”
Data Exhaust: Agents Are Data-Rich Intermediaries
Here’s the hidden unlock: agents generate exponentially more data than users ever would.
Humans don’t fill out surveys. Agents will.
Humans abandon checkout if you ask too many questions. Agents won’t.
Humans forget to give feedback. Agents will do it automatically.
With every transaction, an agent can log:
User preferences (implicit + explicit)
Purchase rationale
Outcome success/failure
Returns + reasons
Satisfaction across dimensions (price, fit, speed)
For agents, feedback might be less about “why did you buy this?” and more about “what additional metadata would have improved the decision?”
This creates a flywheel:
Agents collect more granular data
Infra providers use that data to improve performance
Agents learn which infra works best
The best infra gets more usage
The result is a feedback-rich ecosystem, where every layer from payments to logistics and discovery, gets smarter with each loop.
The Shopping Journey, Reimagined
Most discussions about agentic commerce stay abstract. But what does it actually feel like to use an agent to shop?
In traditional ecommerce, the user drives every step from search to checkout to feedback. With agentic commerce, that flow collapses. The user delegates. The agent handles the rest.
The table below outlines the new flow: what the agent does at each step, and how it transforms the consumer experience.
The Agentic Commerce Flow
In this new flow, “shopping” doesn’t look like an activity. It becomes a background process triggered by intent, executed by agents, refined by feedback.
What used to require attention now just… happens.
Voice, Latency, and the End of Manual Commerce
For years, voice commerce has been a punchline. People asked Alexa to reorder paper towels and then… stopped. The dream stalled.
Why? Because voice without intelligence is just another command line. It’s frustrating, brittle, and shallow.
Agentic commerce changes that. It gives voice a brain.
Voice Becomes Viable
With agents mediating the transaction, voice doesn’t need to drive you through a linear menu. It just needs to express intent.
“Get me new sneakers. Same style as before, half a size bigger, under $120.”
“Book me a room for Thursday night in SF. Same vibe as the last hotel you picked.”
This works because the agent:
Knows your context (purchase history, preferences, schedule)
Can interpret loosely structured language
Has permission to act, not just suggest
Voice now becomes the fastest input for complex commercial actions, especially in mobile and hands-free environments.
Latency is the New UX
In agentic commerce, there’s no interface to optimize. No landing page. No pixel-perfect funnel. The agent does the work.
Which means speed becomes the product.
The fastest inventory response wins the buy box
The fastest fulfillment estimate wins the shipment
The fastest personalized offer gets surfaced
Just as Google rewarded page load times, agents will reward data providers and infra layers with the lowest response latency. Milliseconds matter, not for clicks, but for actions.
This opens the door to a new competitive edge: real-time commerce APIs optimized for agent consumption.
What Disappears
As agents take over, much of what defined ecommerce UX simply… fades.
Gone are product pages, cart flows, filter sidebars, navigation menus, and exit-intent popups. Instead, you’ll see interfaces become conversational, choice becomes curated, and execution becomes invisible.
The best user experiences will be the ones the user barely notices.
Open Questions
Every shift in interface creates a new stack. Every new stack opens white space.
Agentic commerce is no different and we’re still early. The infrastructure is raw. The interfaces are clunky. The winners haven’t been crowned.
Here are some of the questions that matter most right now:
Who owns the gateway? Will it be generalist LLMs (ChatGPT, Claude, Perplexity), OS-level assistants (Apple or Google), or new platforms built around persistent context and intent?
The winner here controls the first moment of commercial intent and becomes the new Amazon, Google, or Shopify.
What’s the infra layer for agents? What is the “Stripe” for agentic commerce? What orchestration tools do agents need to operate securely at scale? Who builds the “Shopify” equivalent, not for stores, but for APIs, logic, and trust layers?
There’s a massive opportunity to define the primitives agents need to transact autonomously.
Can discovery be decentralized? How do independent brands avoid being shut out? Is there a viable co-op model that aggregates long-tail inventory and serves it to agents with performance guarantees? Is there a framework where agents can also draw on user-owned data like cross-platform purchase history built from email receipts to surface relevant products even outside walled gardens?
If not, discovery becomes a winner-take-all game and most brands won’t survive it.
What does brand even mean in an agentic world? How do you market to machines? Will we see a rise in “agent-optimized” product design, simple SKUs, clean data, fast fulfillment? Or will brands reassert themselves post-purchase, in packaging, retention, and returns?
When the agent picks what you buy, storytelling needs a new entry point.
What happens post-purchase? Who owns the relationship after the agent buys something? Can agents proactively manage returns, repairs, warranties, resale, and even donation? Could they collect IRL data post purchase to automate reordering (ex: user will most likely need another gallon of milk at the end of the week → place order and optimize delivery to meet the anticipated window).
Post-purchase is a frontier with massive inefficiencies and agents are positioned to fix it.
What does the agent dev stack look like? What tools help developers build and debug agents reliably? How do you simulate real-world user behavior across verticals? What are the equivalent of “browser dev tools” for agent flows?
This is where second-order leverage lives. The infrastructure that powers agent builders could define the ecosystem.
These are not just technical questions. They’re commercial ones. Strategic ones. Venture-scale ones.
The companies that answer them won’t just participate in agentic commerce, they’ll define it.
From Frictionless Checkout to Frictionless Demand
For decades, the evolution of commerce has followed a simple arc: reduce friction, increase spending.
Swipe replaced cash. Tap replaced swipe. Each step shrank the effort required to complete a purchase, and each one drove real, measurable gains in consumer spend.
Agentic commerce is the next leap, but it’s not just about payment. It’s about removing the need for active intent expression altogether.
It starts before checkout. Before search. Before the funnel even forms.
It begins with agents that understand you, anticipate your needs, and act on your behalf invisibly, intelligently, and at scale. In this model, demand doesn’t need to be expressed. It’s inferred, sourced, and executed ambiently.
That shift doesn’t just rewire the frontend. It reshapes the entire commercial stack:
Platforms become APIs
Brands become metadata
Interfaces dissolve
Infrastructure becomes intelligent
Speed and data quality become the new moats
We’re in the earliest innings. The tools are primitive. The systems are brittle. But the direction is clear.
The last two decades were about building for humans. The next decade will be about building for agents. And those who do it best won’t just make commerce easier.
They’ll unlock entirely new forms of demand.
If you’re building in this space or spend time thinking about building here, I’d love to chat. Feel free to ping me at jordan[at]groundup[dot]vc.
Special thank you to David for helping to bring this memo to life, as well as Cory and Dov for proofreading and sharing great thoughts.