Subscribe now

Bryan House is the CEO of Elastic Path, the composable commerce platform for complex B2B and enterprise commerce. He joined Elastic Path in 2020 to run product and became CEO in 2024.

ReFiBuy and Elastic Path announced a partnership in May, bringing ReFiBuy's agentic commerce optimization to Elastic Path customers. So this conversation is equal parts old friends catching up and partners comparing notes on where commerce is headed.

Listen/Watch Our Interview:

This week’s episode is a platform deep dive with one of the longest-tenured companies in e-commerce. Elastic Path has been building commerce software for 26 years and Bryan is now steering the company straight into the agentic era.

It’s part founder-journey (craft brewing to CEO is not a typical path), part product deep dive, and part industry-shift conversation about what AI does to the economics of running a commerce platform.

Highlights

• 𝗙𝗿𝗼𝗺 𝗯𝗿𝗲𝘄𝗲𝗿 𝘁𝗼 𝗖𝗘𝗢: Bryan spent eight years as a professional brewer, with an actual brewing degree, at a brewpub across from Fenway Park. "Everything I learned in my career that really moved the needle for me, I learned as a brewer."

• 𝟮𝟲 𝘆𝗲𝗮𝗿𝘀 𝗼𝗳 𝗰𝗼𝗺𝗺𝗲𝗿𝗰𝗲: Elastic Path's claim to fame was building the store for the Vancouver 2010 Winter Olympics, the birth of its self-managed commerce product.

• 𝗧𝗵𝗲 $𝟱𝟬𝟬𝗞 𝘂𝗽𝗴𝗿𝗮𝗱𝗲, 𝗸𝗶𝗹𝗹𝗲𝗱: Elastic Path built an AI updater so one developer can upgrade a self-managed platform in about two weeks, replacing what used to be a half-million-dollar annual project.

• 𝟭𝟬,𝟬𝟬𝟬 𝘃𝗮𝗿𝗶𝗮𝘁𝗶𝗼𝗻𝘀: where typical platforms cap out at 250 variations per item, Elastic Path supports 10,000, plus bundles of up to 1,500 items.

• 𝗦𝗶𝘅 𝘄𝗲𝗲𝗸𝘀 𝗮𝗻𝗱 𝟭𝟬𝟬 𝗴𝗿𝗮𝗻𝗱: Bryan's prediction… AI takes re-platforming from a 6-month-to-2-year, $500K–$3M digital transformation down to a setup fee. "That's the place we're gonna see the biggest impact of AI in commerce."

• 𝗧𝗵𝗲 𝟴𝟬-𝗽𝗮𝗴𝗲 𝗣𝗗𝗙: in a world of AI discovery, every product needs the equivalent of an 80-page PDF of data to train the LLM on when to recommend it. (Scot's line, which Bryan says he "steals with credit" as often as he can.)

• 𝗣𝗿𝗼𝗼𝗳 𝗶𝗻 𝗻𝘂𝗺𝗯𝗲𝗿𝘀: a UK custom greeting-card seller saw conversion jump 40% almost immediately after adopting Elastic Path's checkout

• 𝗡𝗼𝗯𝗼𝗱𝘆 𝘄𝗮𝗻𝘁𝘀 𝗰𝗵𝗮𝘁𝗯𝗼𝘁𝘀: customers are asking about UCP and ACP, discovery, and the long tail, not chatbots.

• 𝗪𝗵𝗲𝗿𝗲 𝘁𝗵𝗲 𝗿𝗼𝗮𝗱𝘀 𝗮𝗿𝗲 𝘄𝗲𝗹𝗹-𝗽𝗮𝘃𝗲𝗱: Bryan's framework for where AI lands first… well-understood patterns like checkout, front ends, and reorders, not judgment-heavy edge cases.

Bryan is a product guy after my own heart, and this one gets satisfyingly deep, from CPQ to protocols to the future of re-platforming. Enjoy the conversation.

🎧 Listen wherever you get your podcasts: Spotify, Apple, etc.

Or, we have a video version on YouTube (make sure to subscribe!) →

Want to Learn More?

Stay on top of what’s happening in agentic commerce, from the tools powering it to the founders building with it by following us across platforms:

• Podcast: https://retailgentic.transistor.fm

• YouTube: https://www.youtube.com/@Retailgentic

• Instagram: https://www.instagram.com/retailgentic/

• X: https://x.com/retailgentic

• TikTok: https://www.tiktok.com/@retailgentic

We're here to make sense of what’s next in retail + AI and bring you along for the ride. Subscribe and stay tuned!

This feels like the busiest week of the year, we’re in that frantic ‘post labor day’ heading into conference season with Holiday 2026 bearing down on us - Thanksgiving is 72 days way - basically we’re in that critical '2 months left to get all our ducks in a row’ period.

Going to RetailClub or GroceryTalk or ShopTalk?

At ReFiBuy we have folks at all three shows - Jason and I are at RetailClub where we’ll be doing live podcasting of both Retailgentic and the Jason and Scot show. If you’re not going, don’t worry, we’ll be streaming many of these pods live from the show floor (well sand, it’s on the beach!) so be sure to subscribe to our YouTube channel and turn on notifications.

Subscribe now

Heading into Holiday 26…

This is a great time to update some of our big picture Agentic Commerce Optimization thoughts so everyone can focus on getting the best bang for the buck in this short time. The Good news? → The easiest thing to change and also the highest impact is how you describe your products to agents.

In this part 1 of the 2 parter we focus on: "Personal Shopping Agents enter the Arena”

In part 2 coming after RetailClub on 9/29 we’ll cover the rest of the update and have a corresponding podcast/video presentation that puts both parts together and gives you all the data and slides/graphics so you can use them in your Q4/27 planning.

• New: Latest datapoints on Agentic Commerce Scale and Growth

• Updated: Anatomy of a Product Card

• Updated: 5 Levels of Autonomy

• Updated: ACO Everywhere strategy map

• Updated: ACO 11 Steps Sequential Strategy

Personal Shopping Agents Enter the Arena

In the last two weeks we have seen an entirely new category of Agentic Commerce come seemingly out of nowhere that we call Personal Agents - these are horizontal agents meaning they do many more activities, such as email, calendaring, event planning, travel booking, but they do all anchor on and tout their Agentic Commerce capabilities. These are different than the vertical shopping agents:

Top Five Brand and Retailer Questions about Personal Agents used for Agentic Commerce

These agents are gaining traction rapidly and generating a lot of question for retailers and brands. I’ve rolled them into these 5 questions:

Q1: Should I block these?

Q2: How do I track personal agent transactions?

Q3: How do I optimize for personal agents?

Q4: How do these personal agents work and where are they going?

Q5: How is this different than Answer Engine-driven Agentic Commerce?

In Part 1, of this series we’re going to dig into and answer these questions, then in part 2, we’ll zoom back out and answer question 5 by updating what’s going on in the rest of the world of Agentic Commerce which is not holding still at all either.

To answer these questions, we need to dig into background, talk a bit about Instinct and GrokBot as those are new here on Retailgentic (We have detailed Muse walk through with purchase flow here).

Personal Agent Backgrounder

To answer the big five questions, we need to look at:

• The underlying exponential model improvements - it’s the rising tide lifting all the capabilities we cover here.

• History of Personal Agents - The history here will help us see where this is going and how fast.

• What’s a ‘Harness’? - You’ve probably seen this floating around, Harnesses and Loops are all the rage in the AI builder community. We stay focused on 'commerce’’ here at Retailgentic, but to answer the five questions, it will help to have a working knowledge of what a Harness is, what it does and what it doesn’t do.

• Two Modes of Agentic Browser Use - The most important part of these harnesses that impacts us and leads to the answers for our five questions is Agentic Browser Use.

Underlying it all: Exponential LLM Improvements

At this point, you probably gloss over the charts showing exponential growth of LLM intelligence. Some look like this:

Or this→

For example, Claude Haiku came out 10/25 and scores an indexed 638 on this meta-benchmark aggregator and now Fable 5.1 is at 7,798. The models are improving at a compounding 10x rate and everytime we think there’s a slow down, a new innovation, busts through that. This time next year, it will be at 80,000. 🤯

But, just like humans, the model’s knowledge is based on what it learns, or is trained on and this creates what AI researchers call ‘jagged edges’. LLMs like Fable 4.1, Sol 5.6 can write code for hours if not days now that’s better than humans, but some simpler tasks, they completely fail at. If you want the best examples of these follow ‘Husk’ on YouTube or X.

Here’s a fun example: (remember this model can also code for days on end)→

How can something so smart also be so dumb? Jagged edges.

One of the edges the model companies have been working hard on is browser/computer use - training the models to understand human interfaces so they can better do work on our behalf. In our world we’re most interested in the Research→Find→Buy cycle, but computer use includes things like: “Create a ticket in my crm system”, “open up SAP and create a new widget”, “Load up my PIM and translate this SKU into 93 languages.”.

In fact, The Information recently reported:

The root cause of the emergence of these super smart personal agents is due to the new (last 6-12 months) focus from the frontier labs on models trained for computer use.

There are ~50 different benchmarks for browser/computer use and this one aggregates them:

To put this in perspective, Gemini Pro 3.1 Preview (the one on the far right that scored 457) came out in 2/26. Today the models are scoring 1720 on an indexed scale, that’s a 4-5x improvement in 10 months. Many of the benchmarks have scores that have a % of tasks completed. These models are scoring 93%, 96%+ at tasks that are much much more complex than buying something online.

Part 1 of the story is in the last 12 months, the underlying agents have gotten 10x smarter at general intelligence and 4X smarter at computer use. Researching→Finding→Buying a product on a consumer website is child’s play**

**with the right context, memory, tools, etc.

What’s a “Harness”?

That ** above is important. Think of the models as raw intelligence - computer brains, but they kind of all over the place and wild, they need direction. A harness, borrowed from a horse harness metaphor, gives the LLM direction for a specific use-case.

The picture above illustrates how the harness ‘surrounds’ the agent and wraps it with the pieces it needs to guide the raw intelligence, keep it focused, keep it grounded, help it remember where it went right and wrong (writing things downs in a scratchpad turned out to be a huge breakthrough - who knew!).

Harness are increasingly adding advanced capabilities:

• Agentic Loop - On a timed or event-based basis, the agent wakes up and takes pre-determined actions. For example, when I get a new calendar invite, do x, y, z or closer to home: if I get an email about a product I’m looking for that’s less than the 30-day average, go buy it.

• Multi-player capabilities - On the consumer side, this is usually inviting friends and/or family to book with you via a Resy/OpenTable or flight/hotel/airbnb

• Compounding self-improvement - The harnesses understand their goals (“buy the widget!”, “book the hotel” - the models execute and the harness learns what context to feed in and what overwhelms the engine.

• Browser use - This is where the magic happens in our World, stay tuned for more details.

If you’ve used chatgpt, claude - those are light harnesses. If you’ve used Claude cowork or ChatGPT Codex→Work, those are heavier more meaty b2b harnesses.

To understand the consumer agent harness progression, we need to go way way back to the earliest days of consumer - way back to January of 2025 (or Oct 25 depending on how we’re counting).

Retailgentic’s History of Personal Agents

The pre-cursors to these agents is OpenAI Operator, in fact ChatGPT’s co-founder, Greg Brockman has been saying since early 2025 that ‘computer use’ is a top priority for ChatGPT→

Jan and July you can see OpenAI working on computer use with the introduction of Operator which became Agent. Then they folded that into Apollo - their browser Atlas (here’s our first Atlas LIVE Retailgentic demo if you want to jump in the time machine for a spin.)

IMHO, what really broke this open, was the introduction of OpenClaw in 11/25 which by March 26 had a thriving community of people extending the harness to Agentic Commerce.

After OpenClaw got progress, some ‘clones’ came out that simplified the install, added some multi-player improvements and beefed up the security. Examples in this category are: Hermes, Perplexity Computer and Microsoft Scout is transparently based on OpenClaw. At Google I/O they announced and launched Gemini Spark.

I’ve tried all of these it’s very hard to get the right mix of proactivity/safety/pedantically asking permissions.

On one side you have OpenClaw that’s scary aggressive about access to your computer and all your files. On the other end of the spectrum you have Google Spark which asks for permissions so much and for every task and sub-task that it’s basically unusable (you’ll see in the demo below).

The Q3_26 ‘class’ of personal agents broke through this. They have a great balance of getting up fast, not bothering you for permissions every five seconds for permissions, but asking permissions at the right break points (e.g. hey you’re about to spend money, review this transaction for me).

Two Modes of Browser Use: Server and Local

Back in the Agentic Browser wars days, we spent some time walking through how the first ‘computer-use’ agents used servers to do their browsing. The challenge with this model is it’s easy to block - the destination (e.g. Amazon) or the infrastructure layer (e.g. cloudflare) can detect the bank of IPs via fingerprinting and say: “Oh this is a Meta muse bot, I will shut it down”. What seems to always happens is these systems start with server-based (or a virtual machine, or VM) and then add the ability to use the local machine. When you’re on the consumer’s local machine, you are effectively unblockable - the agent’s traffic looks exactly like a consumer. As a countermeasure, companies are developing technology that looks at how fast you’re using the browser, mouse movements and what-not to try and decide: “human or agent”. The problem with this is ‘false positives’ - I frequently, especially when using some specific sites and retailers get put through Captcha sequence (pain) or the slidey-puzzle thing, etc.

Of note - all of these ‘prove you are a human’ tests have fallen with the growth in computer use, so they are placebos at this point.

This all ties to the ‘track-ability’ of personal agents which also ties to ‘optimization’ so It’s worth going through this diagram in more detail:

1. The user is on their local computer - they have what we call a consumer IP

2. They use a personal agent which, even if it’s a desktop or mobile app, is running in a server/vm.

3. That agent needs to do something with a browser, so spins one up in it’s VM (usually chromium-based so it “looks like” chrome).

4. For Agentic Commerce the browser tool decides if it wants to crawl the merchant’s website, use UCP (or some other protocol like ACP, WebMCP, etc.) for discovery and transaction or some mix of those.

5. Alternatively, the agent may decide to connect back to the consumer’s computer and use a local browser, a browser extension, or IP routing.

In fact, literally as I was working on this post Elon conveniently posted this→

They very clearly say the intent is to get around server-side IP blocking - nobody is hiding the ball here.

Demos: Gemini Spark, GrokBot, Instinct and Muse

To demonstrate the agentic commerce capabilities of all these systems, I’ll briefly walk through Spark, GrokBot and Instinct to show you the experience.

The Veggiedent test

Long time readers have probably recognized that one of my favorite use cases is the dog treat →

On the surface this seems easy, but the lowly veggiedent has broken many an agent because, just like all of e-commerce, there is a lot of complexity hidden in this image:

• These are not widely available

• They have these variations:

  • FRESH and 2 other ‘styles’ of chew

  • Dog size - this is small 11-22lb

  • Pack size - this is the 30 chew

Agents struggle to get this right - spoiler alert, at least one of our agent demos will not survive the Veggiedent test!

Also, I happen to know that Chewy and Amazon battle it out for this SKU and they are two of the best blockers of agents.

Gemini Spark Walk Through With Transaction

After finding 10mins how to get to it (mobile only, personal account not business, pro or something, hidden in a menu, accept/yes/yes/yes/yes/yes/yes/yes/ok/upload pick/accept/ok/ok/ok/phew). I was able to get to shopping:

GrokBot Walk Through With Transaction

I’m in very deep in Grok Bot for work usage - i have a fleet of 15 bots doing things for me every day and collabing on many projects - it’s amazing for that use case. But let’s try it for ecommerce. I go in and setup a bot, I called mine cleverly: “Scot’s Shopping Bot”

I’m including the setup here because it’s pretty cool and gets a lot of the nagging questions out of the way in one big chunk. The most interesting one→

You can see I freed my shopping bot to buy anything under $50 without bugging me. Let’s hope that doesn’t backfire.

Only Chewy was able to block GrokBot - kudos to the team for figuring out how to get Amazon to work - not easy. (Note: I’m not using the local IP, not available to me at the time of this writing, that would get around the Chewy block though for sure).

From here let’s look at the Amazon buying flow:

To finish, I have to login to my amazon account - I can give this info to Grok to hand over to Amazon or take over. Both Muse and GrokBot have secure VMs that the LLM can ‘write to’, but not read from so their claims are that your credentials are secure. Personally I’m increasingly becoming a fan as a consumer of the Stripe Link System - it’s one click that has a complete review and you’re done - pretty amazing. Speaking of which, let’s look at Instinct now.

Instinct Walk Through With Transaction

First of all, Instinct is in private beta and you need an invite code, they are pretty hard to come by, but if you’ve made it this far and need one - DM me on LinkedIn - we have a pool of Instinct Invites here at ReFiBuy HQ for our BFFs.

Instinct runs right in your iMessages and it’s best characterized as “insanely proactive”. You have to be very clear with it or it will go and do things you may not have intended. But it always has the best intentions and that proactively has really been a game changer for me - it’s saved my 🥓 a couple of times and I've been using it 2wks.

Instinct famously is dealing with a crushing load of new users and can’t get servers fast enough, so there was a glitch with the image upload that was easy to fix→

Notice it got through to chewy, but not Amazon in this use case.

Then I clicked on the link to see the product before buying and…

Womp womp - Instinct has a bit of a ready/fire/aim thing going on and it got caught in the Veggiedent trap - (these are the extra small vs. small) - third variant got it.

Lesson - always check the transaction before approving. Speaking of which, I ordered a cooler using Instinct for tailgate season and, like Muse, it uses the Stripe Link connection for payment which I really am enjoying→

When I click on that link (I won’t show that as it has tons of PII), it lays out the whole transaction and once you authorize it and TFA, it magically works.

After you order something, Instinct checks your email and pings the order site proactively so you know where it is in it’s journey to you.

Some people have even wired Instinct up to a phone calling service and have it call merchants and do things IRL for it - wild stuff, on my weekend fun experiment list.

Muse Product Walk Through

Ok, I realized that I didn’t show the Veggiedent example in my Muse walkthrough so let’s do that real quick to be fair→

From there, Muse found a $20 off and processed my order with a stripe link!

Recap: Muse vs. Grok Bot vs. Spark vs. Instinct

Spark came in last for me: while it didn’t trip up on Veggiedent, it is so far behind the others, capability-wise it has a lot of room to make up. GrokBot is more a business agent for me, but it did well.

Muse is the real standout. I’m not sure why Chewy/Amazon aren’t blocking it yet, perhaps it’s early, or, if you’re META, you have one infrastructure for FB, WhatsApp, Instagram, Threads, etc. all that traffic, including Muse looks like ‘meta family traffic’ -perhaps it’s hard for the blockers to separate that out?

The real winner: Distribution

The hardest thing to conquer with a consumer business is getting consumer attention. In this field of 4 Gemini and Muse have a huge advantage - existing billions of consumers. All you have to do inside one on of the big existing platforms with massive distribution is put a little ‘try Muse now button’ and BOOM, off to the races. In fact, there’s early evidence they are doing exactly that→

Back to the Five Big Merchant Questions Personal Agents and Agentic Commerce

Q1: Should I block these?

A1: For me there are 3 angles on this one: retail/dtc brand/non-direct brand

Retail Answer: This is the toughest question you face right now and comes down to theology and where you are as a company. You see Amazon, Chewy and many of the top retailers in a category increasingly actively blocking. The argument here, is “I don’t want to be a logistics company and give away the front door.”

DTC Brand Answer: For DTC Brands, they are always looking for more distribution, so they tend to open the door wide. Because many of them both DTC and wholesale. they are already used to ‘giving away the front door’, so theologically they are ok - the argument here goes: “meet the customer where they are” and where they are is increasingly using Agentic Commerce across all these new surfaces.

Non-DTC Brand Answer: This is the easiest one - you should make sure your brand site is open so all agents can get all the product info they can consume and work with your retailers to optimize for their on-site agents and if they are embracing other outside agents (Walmart is prime example), collab with them on that.

At the end of the day, the theology question seems to come down to: How badly do you want new customers? Can you afford to lose the sale to a site that doesn’t block?

Q2: How do I track personal agent transactions?

A2: In their current configuration you can see with GrokBot, where i’m using my credentials, all you see that’s different is a new login from a new device. Eventually GrokBot’s IP range will become known, but again, they are already implementing local which gets rid of any hope of tracking that model.

I think the good news is once you start using these super efficient, fast and easy agents, you open up your mind to getting out of the middle of the checkout - the Stripe Link thing got me over that very quickly.

We’re looking into specifics, but there’s definitely a way to track that so at least you can bucketize “agents that paid with Stripe Link” and put a number on this category and see how big it gets, bounce it against your customer file and see the existings vs. new and what-not.

Q3: How do I optimize for personal agents?

These bots all see the meta data on your site. There’s speculation Muse is already or soon will start looking at UCP, they use it in other parts of the FB family. The WebMCP people see a world where everyone switches to that.

To be honest, the computer/browser use is so good, I’ve been using this very heavily for 30 days and it’s not getting stumped with anything. The discovery process does break down sometimes and you can see some of the agents are using intermediate product databases (Muse and Spark do this as best I can CSI) and in there they maybe using the UCP part of discovery.

We’ve said this here maybe 1000 times, but it works here too - the single thing you can do regardless of retailer/brand/etc. is work to make your catalog agent friendly - keep chipping away at those 5 levels of enrichment (details here).

The harnesses need context and content to chew on - feed them well!

Q4: How do these personal agents work and where are they going?

You can see in the timeline here that in just 12 months these agents have gotten very sophisticated. Using the Retail 5 levels of Agentic Commerce Autonomy as a guide→

These consumer agents are squarely straddling 3/4/5 with different elements. Where they need to go next:

• A ‘wallet’ for your favorite preferred types, loyalty programs, etc.

• Multi-cart/wish-list and deal watching (like Gemini multi-cart)

• Extracting your purchase history from email (they can do it, but requires prompting and they are forgetful compared to Answer Engines)

• Prediction of needs - Instinct is getting there, very cool and helpful.

Q5: How is this different than Answer Engine-driven Agentic Commerce?

That is the main topic of Post 2, we’ll be back after RetailClub!

Subscribe now

Jaime Sorgente is Head of Retail at Vercel, the developer-tools and infrastructure company behind Next.js (the open-source framework powering storefronts like walmart.com and target.com) and v0 (its vibe-coding tool). Jaime joined Vercel almost three years ago from a large hyperscaler, where an early-2023 habit of feeding client 10-Ks into ChatGPT convinced him that everything about go-to-market and software-making was about to change. A Stratechery podcast episode with Vercel CEO Guillermo Rauch sealed the deal, and he’s been building Vercel’s retail vertical ever since.

Listen/Watch Our Interview:

This week’s episode is a little different. We’ve has been noticing vibe-code platforms quietly showing up all over digital retail… at trade shows, in conversations with retailers and brands, and on the platforms’ own commerce pages. So we went to the source.

This is a platform deep dive with Vercel, the enterprise company behind v0 and Next.js, on what a “vibe coding” company is actually doing in our world of commerce, and it turns out the answer is: a lot more than you’d think.

Highlights

• 𝗧𝗵𝗲 𝟭𝟬𝟬-𝗺𝗶𝗹𝗹𝗶𝘀𝗲𝗰𝗼𝗻𝗱 𝗺𝗮𝘁𝗵: A Deloitte Digital + Google study cites clients found 100ms of load-time reduction drives a 7–8% lift in e-commerce conversion. “If the thing is faster, you’re gonna convert more. It’s just math.”

• 𝗛𝗲𝗹𝗹𝘆 𝗛𝗮𝗻𝘀𝗲𝗻’𝘀 𝗻𝘂𝗺𝗯𝗲𝗿𝘀: The Kontoor Brands-owned retailer cut load times from 2.5 to 1.5 seconds on Vercel, grew conversion 20% year over year, doubled its A/B test velocity, and blew away its Black Friday/Cyber Monday plan.

• 𝗧𝗵𝗲 𝗰𝗼𝗰𝗸𝘁𝗮𝗶𝗹-𝗽𝗮𝗿𝘁𝘆 𝗽𝗶𝘁𝗰𝗵: Vercel in three phrases… a developer tools company, an infrastructure provider, and an open-source software factory.

• 𝗬𝗼𝘂’𝘃𝗲 𝗮𝗹𝗿𝗲𝗮𝗱𝘆 𝘂𝘀𝗲𝗱 𝗩𝗲𝗿𝗰𝗲𝗹: walmart.com and target.com run on Next.js, and Under Armour, Crocs, Buck Mason, PetSmart, McAlister’s, DeWalt, Clorox, and even Meta.ai are Vercel-delivered applications.

• 𝗡𝗼𝘁 𝗮 𝗦𝗵𝗼𝗽𝗶𝗳𝘆 𝗸𝗶𝗹𝗹𝗲𝗿: Vercel doesn’t do product catalogs, inventory, or order management. It’s the storefront layer on top of your commerce platform. “We help you own your storefront and own your destiny.”

• 𝘃𝟬 𝗺𝗲𝗲𝘁𝘀 𝗦𝗵𝗼𝗽𝗶𝗳𝘆: v0 is now deeply integrated with Shopify, so you can customize your storefront in plain English

• 𝗧𝗶𝗺𝗲𝗹𝗶𝗻𝗲𝘀 𝗵𝗮𝗹𝘃𝗲𝗱: Custom storefront projects that were planned for a year are landing in six months, six-month plans in three.

• 𝗧𝗵𝗲 𝗮𝗴𝗲𝗻𝘁 𝗰𝘂𝘀𝘁𝗼𝗺𝗲𝗿: Next.js 16 was built to be readable and manipulated by Cursor, Claude, and Codex, “now we have this new customer, this agent customer, who wants to interact with your code base.”

• 𝗧𝗵𝗲 𝗹𝗶𝗴𝗵𝘁-𝗯𝘂𝗹𝗯 𝗺𝗼𝗺𝗲𝗻𝘁: Jaime’s conversion story: feeding 50-page client 10-Ks into ChatGPT in early 2023 and realizing everything about go-to-market, and the factories that make software, was about to change.

This one really lifted the veil on what the vibe-code platforms are actually doing in retail, and I think we’re just seeing the tip of the iceberg. Enjoy the conversation.

🎧 Listen wherever you get your podcasts: Spotify, Apple, etc.

Or, we have a video version on YouTube (make sure to subscribe!) →

Want to Learn More?

Stay on top of what’s happening in agentic commerce, from the tools powering it to the founders building with it by following us across platforms:

• Podcast: https://retailgentic.transistor.fm

• YouTube: https://www.youtube.com/@Retailgentic

• Instagram: https://www.instagram.com/retailgentic/

• X: https://x.com/retailgentic

• TikTok: https://www.tiktok.com/@retailgentic

We're here to make sense of what’s next in retail + AI and bring you along for the ride. Subscribe and stay tuned!

Moments ago, META announce Muse via a Zuck X-storm!

We’ve been waiting for this moment at Retailgentic for weeks and pounced right on it to give you this exclusive FIRST LOOK 👀!

Subscribe now

Here’s how it works:

1. Go to muse.ai

2. Login with our facebook/instacart credentials or create a new account

3. Name your muse, setup your Avatar and get it going!

In this FIRST LOOK, we’re going to show you all that and then go and buy something - let’s go.

FIRST LOOK Part 1: Setup: Create, Name and Avatar Your Muse!

When you first get going Muse walks you through what we’d call a warm-up or ice-breaker experience.

Once I named my Muse, “RetailMuse”, I figured it should have a more appropriate avatar, so clicked the ‘edit avatar’ link and this popped up:

I liked the vibe of bottom left and here we are, ready for action→

FIRST LOOK Part 2: Finding a Coffee Machine

Now that we’re done birthing RetailMuse, let’s jump into a shopping task and make it a tough one. This is a real-world situation, I need a new coffee machine (long story) and I know the one at work is too big, so I used it as my - ‘make it smaller than’,

Here’s the prompt→

RetailMuse immediately, liked the post giving me feedback it was thinking, then it popped up it’s HTML browser→

This is a browser-based agent, you can see it zip through the websites like crazy, looking at models, filtering by my price point and then quickly looking for size information - making sure my criteria are met.

I filmed 2 short videos, one with the browser in mini-mode and one in max-mode. You’ll note that it struggles with the Delonghi site, there are a lot of 404’s on tech specs and the site is the least browser friendly amongst the three brands I constrained it to:

Video 1: (small browser mode)

Video 2: large browser mode

FIRST LOOK Part 3: Time to BUY!

Well, look at that folks, yep, you guessed it….PRODUCT CARDS, they are everywhere. here we have the top five products that my RetailMuse has narrowed us down to based on my criteria.

I picked the top card to see what that does→

At this point I’m given a choice:

• Path A: Buy with RetailMuse (the name of my Muse agent) or…

• Path B: ‘Visit Website’ - here I take over the browser and complete the purchase

What’s behind Door 1 (path A)?

Choosing the “Buy with RetailMuse” had this interesting outcome→

Here we are days after Stripe predicted the death of checkouts and look, right there at the bottom, I am being strongly encouraged to use “Link by Stripe” to create a token for my credit card (a one use virutal credit card) that will be used by Muse for the guest checkout to complete my purchase in a secure way.

What’s Next?

We’ll continue working on this and finding the edges of it’s capabilities and thinking through the ramifications for Merchants. Until now we’ve been keeping a close on these consumer agents, OpenClaw, Hermes, GrokBot and Instinct. What they lacked was one thing, the hardest thing: distribution - getting consumers to use something new is very very hard. The easiest path is if you already have 1b users, offer them something new. None of those companies mentioned have that. But you know who does? Meta - they have 4 products with > 1b users, so this is now quite interesting. But that doesn’t guarantee consumers will adopt it. Stay tuned, we’ll be watching this closely and next week we’ll zoom out a bit with some best practices for merchants for these browser-based consumer options and thoughts on how to prioritize this opportunity vs. the Answer Engine side of things.

Subscribe now

Subscribe now

Rishabh Jain is the co-founder and CEO of FERMÀT, an AI-native personalization platform for brands and retailers selling direct. They’re betting that behavior, not identity, is the signal that actually drives conversion. He also hosts the 1 to 100 podcast on scaling companies from $1M to $100M in revenue.

Listen/Watch Our Interview:

This week’s episode is a deep dive into personalization, a huge category in e-commerce that’s getting rebuilt from the ground up for the AI era.

FERMÀT is an AI-native personalization platform that sits on top of your e-commerce platform and customizes the site experience based on a visitor’s behavior, not their identity. That distinction sounds subtle, but it turns out to be the whole ballgame and it connects directly to agentic commerce because agents show up with intent, not an identity. We also pick up Rishabh’s founder journey and some very candid company-building advice along the way.

Highlights

• 𝗧𝗵𝗲 𝟮𝟬% 𝗰𝗼𝗻𝘃𝗲𝗿𝘀𝗶𝗼𝗻 𝗹𝗶𝗳𝘁: FERMÀT's data shows personalizing on behavior alone, even for fully incognito visitors, converts roughly 20% better than personalizing on identity and past purchases.

• 𝗠𝗮𝗻𝘆 𝘁𝗼 𝗼𝗻𝗲, 𝗻𝗼𝘁 𝗼𝗻𝗲 𝘁𝗼 𝗼𝗻𝗲: the marketer's dream of one-to-one was "a lack of imagination", the same person is a different shopper on Instagram, Google, and ChatGPT.

• 𝗗𝗲𝘁𝗲𝗰𝘁, 𝗿𝗲𝗰𝗼𝗺𝗺𝗲𝗻𝗱, 𝗴𝗲𝗻𝗲𝗿𝗮𝘁𝗲: instead of humans watching a sample of session recordings and shipping site changes months later, agents trained on commerce watch every session, recommend changes, and generate the new experience.

• 𝗧𝗵𝗲 𝗕𝗹𝗼𝗼𝗺𝗿𝗲𝗮𝗰𝗵 𝗽𝗿𝗼𝘃𝗼𝗰𝗮𝘁𝗶𝗼𝗻: identity-based personalization was state-of-the-art 10 years ago. Rishabh's take on the incumbents: if nothing changes, "their 3 to 400 million of revenue will become 2 to 300."

• 𝗠𝗶𝗻𝗲 𝘁𝗵𝗲 𝗮𝗱, 𝗻𝗼𝘁 𝘁𝗵𝗲 𝗜𝗗: even when an Instagram visitor is anonymous, the ad they clicked carries demographics, performance history, and frequency data… context most brands never use.

• 𝗧𝗵𝗲 𝗿𝗲𝗮𝗹 𝗲𝘀𝘁𝗮𝘁𝗲 𝗮𝗴𝗲𝗻𝘁 𝘁𝗲𝘀𝘁: an agent is only as good as the feedback, context, and iteration speed you give it and every "actually, I also care about the collar" update is behavior data.

• 𝗦𝗮𝘆 𝘄𝗵𝗮𝘁'𝘀 𝘁𝗿𝘂𝗲: Platforms are using AI to verify your pages and truth gets rewarded, not dinged.

• 𝗦𝘂𝗽𝗽𝗼𝗿𝘁 𝗮𝗴𝗲𝗻𝘁 ≠ 𝘀𝗵𝗼𝗽𝗽𝗲𝗿 𝗮𝗴𝗲𝗻𝘁: support agents should stick to facts and policies; shopper agents should infer.

Rishabh is one of the clearest thinkers I’ve talked to on where personalization goes in an agent-first world, and his founder advice at the end is worth the listen on its own. Enjoy the conversation.

🎧 Listen wherever you get your podcasts: Spotify, Apple, etc.

Or, we have a video version on YouTube (make sure to subscribe!) →

Want to Learn More?

Stay on top of what’s happening in agentic commerce, from the tools powering it to the founders building with it by following us across platforms:

• Podcast: https://retailgentic.transistor.fm

• YouTube: https://www.youtube.com/@Retailgentic

• Instagram: https://www.instagram.com/retailgentic/

• X: https://x.com/retailgentic

• TikTok: https://www.tiktok.com/@retailgentic

We're here to make sense of what’s next in retail + AI and bring you along for the ride. Subscribe and stay tuned!

Today via a series of social media and blog posts, Anthropic released their 'Claude for Commerce’ offering.

Subscribe now

Background

Back in January, one of our 11 2026 Retailgentic predictions (number 7) was “Claude/Anthropic come out of nowhere”→

In June, we saw that Claude for product related conversations is surfacing product cards.

On the surface, this may seem like a minor deal, but at the intersection of computer science and e-commerce, this implies they are building a whole ontology - an interior product catalog, the equivalent of ‘hero skus’ (their own master catalog), understanding the canonicalization of offers to the hero SKUs, a taxonomy, datafeed and product-crawl capabilities, etc. Those are essential foundational building blocks for building anything in our world.

How Anthropic builds vertical solutions…

I’ve studied pretty closely how OpenAI and Anthropic are building vertical solutions:

• They have teams of engineers they deploy on-site for projects. OpenAI calls them ‘Forward Deployed Engineers’ and Anthropic calls theirs “Applied AI Engineers”

• They organize by vertical - both Anthropic and OpenAI have Commerce oriented groups

Once they do enough of these engagements, they create products, or re-usable agents, or whatever is needed around those verticals. They then use these to drive adoption of more token usage for their system and more Applied AI engagements, etc.

Anthropic has a pattern - they dip their toe into a vertical, then as it gets traction, they keep expanding and expanding, going wider and deeper. Here on the other verticals where they’ve planted flags that now have their whole own industry teams.

How Big is Anthropic’s Ambition?

How serious is Anthropic about this? Gavin Baker recently said on the All-In podcast: (about Dario Amodei, Anthropic CEO/founder):

A company spokesperson called this “completely false”. 🤔

What IS Claude for Commerce?

Claude for Commerce is two agents that Anthropic has built and has open sourced as ‘reference implementations’. In other words it’s a technical blueprint for how any retailer can leverage claude to solve the specific problems.

• Agent 1: Consumer Agent - This is what here at Retailgentic we call a retailer-hosed shopper agent. Claude is basically giving you the code to have a very powerful shopper agent ‘for free’ - basically you pay the cost (tokens) to customize it and operate it (you’d pay that anyway). Interestingly, their messaging is very consistent to mention that this works with “retail and marketplaces”.

• Agent 2: Merchant Agents - This ‘family’ of ‘store operations” agents that can track sales, stock, pricing, flag listings running low (marketplaces), markdown recos, and draft a restock plan or promotion on request.

Digging through this, the consumer agent feels like it is ready to work for SMB ($0-15m/yr GMV) to Enterprise ($15m+), but the Merchant Agent has much more of a SMB/marketplace seller feel for it.

Supporting this feeling, Harley Finkelstein, Shopify’s President, posted this→

Let’s dig into each of them in classic Retailgentic style!

Claude Consumer Agent Details

Let’s dig into the Consumer Agent…

For the consumer agent, the Claude reference demo for a retailer called ACME (I’m going to guess they sell large explosives to coyotes?) and show how the shopper agent works:

Here you can see this Acme is actually selling toys and gives and the shopper asks for some birthday gift ideas for a 9yr old, 2 products (product cards!) are recommended and you can add them to the card in-line and the agent does appear to be ‘underlying store context aware’ which is very Alexa for Shopping-esque.

Claude Merchant Agent Details

ACME is back on this side of the coin:

Here we have a LOT going on:

• Navigate - Here we see a menu with 5 options: Home, Catalog, Orders, Inventory, Assistant.

• Home Dashboard - On the home screen, we have a Claude-fueled dashboard surfacing 4 KPIs for this merchant

• Needs you today - This is an agent surfacing 11 priority items across ‘orders’/’low stock’/’slow’ categories of notifications. Think of these as ‘exceptions’ that a SMB operator would need to tackle ASAP

• From the assistant - Anything outside of the core notifications the assistant has surfaced- in this example we ee a surge in returns on a SKU, an abnormal sales bump and some capaign-related info.

• Recent orders - some highlighted orders that are coming through and their status (again very SMB).

• Merchant Assistant - here you can just chat with your assistant and have it do things under the hood for you.

From these you have seen enough of these to guess what Catalog/Orders/inventory look like.

8 Reference Examples

Finally, on top of giving you all the code to the Consumer and Merchant agents, Anthropic throws in not one example, but 8, yes 8 reference examples where they’ve taken these agents applied them to: retail + marketplacesX2 (our wheelhouse), TravelX2, TelecomX2, EntertainmentX2 (the X2 is * 2 meaning they demo the 2 agents for each of the 4 verticals and thus the 8 demos.

Putting it All Together, What Does this Mean?

Initial thoughts:

• There are probably 200 companies building retailer shopper agents from legacy companies adding them (all the personalization and on-site search companies have offerings) and this is going to really disrupt that industry. Gemini and Copilot are also trying to charge enterprise licensing fees here, this will be extremely disruptive to that.

• This explains why Shopify hasn’t been really pushing a shopper agent - they are a launch partner here.

• There are lots of Shopify apps that should be worried

• If I were operating a marketplace management platform, like Rithum, Sellercloud, Sellbrite, Teikametrics, this would be very concerning. Sure it’s small and a ‘toy’, but (go back to the background)

For More Information:

Want to learn more, we’ve put together the three jump off points to learn more from top-level down to the gory code-level details→

Claude Commerce Resource 1: Introducing Claude Commerce Agents Guide:

Introducing Claude Commerce Agents

10.9MB ∙ PDF file

Download

Download

Claude Commerce Resource 2: Blog Post: Anatomy of effective commerce agents here→

Claude Commerce Resource 3: GitHub Repository

Lives here and has this directory of files→

Happy exploring!

Subscribe now

August did not slow the Agentic Commerce news, so we could all take a rest. Instead It kicked down the door with inter, a wave of fresh consumer surveys, new platform moves and some very spicy predictions about what happens to checkout.

The theme this month that comes through: Discovery (what we call Research → Find) is mainstream, Buy is accelerating and the infrastructure race is officially on and moving across many different paths, protocols and systems. Let’s dig in.

Subscribe now

2 EDITORIAL NOTES:

1. Data statement - Observed data describes measured behavior (from panel data). Stated-preference data describes what surveyed consumers say they do or plan to do. Both are useful, but they answer different questions - we like to triangulate here at Retailgentic - look at all the data, pick out the trends that start to form. We’ll include in our synopsis of stories what type of data we’re talking about.

2. Vocabulary - We use Agentic Commerce with a ‘big tent’ definition to encapsulate all consumer digital retail influenced through the Research→Find-Buy. We use ‘Answer Engines’ to designate ChatGPT, Gemini, Claude, Grok, CoPilot, etc. We use Retailer Shopper Agents to designate chat bots that live on a retailer’s site like Alexa for Shopping (Amazon AFS), Walmart Sparky, Lowes MyLow, etc.

We have a plethora of news that we consider top-tier for this curation, but they fall into three themes so we’re going to cluster them together:

• Theme 1 - The Retail Shopper Agents Strike back - examples of datapoints indicating there is a lot of engagements and wins coming from retailer hosted shopper agents.

• Theme 2- A Cornucopia of Consumer behavior trends from lots of different surveys

• Theme 3 - Interesting Industry Moves that didn’t rate a big FLASH/BREAKING standalone Retailgentic alert, but still something we think you should have on your radar.

August Trend 1: Retailer Shopper Agents Strike Back

In these pages, we talk mostly about Answer Engines, and if you want to go DEEP on AFS (and soon) Sparky, etc. check out our sister publication Retailplaybook by Andrew Bell.

News-wise, there was a nexus of stories on this topic of shopper agents that I thought you’d find interesting and actionable.

Evercore’s AI survey: shoppers LOVE AFS and they spend more when using it.. Raises PT to $335, stock pops 4% on the call.

Friend-of-Retailgentic, Evercore analyst Mark Mahaney, dropped a major consumer survey focused on Agentic Commerce that’s a dense ~20 pages, chock full of insights.

The core findings are very bullish for Alexa for Shopping, formerly Rufus, which we’re going to call Amazon AFS or just AFS going forward.

• 66% of respondents are satisfied or very satisfied with how agentic AI affects online shopping (generally)

• Only 3% are dissatisfied or very dissatisfied

Then the survey turns to retailer hosted shopping agents and AFS specifically→

• 55% of respondents said they use ‘Retailer’s built in AI assistant’ (vs. 48% for ‘General AI ')

Then they turned to AFS specifically:

• 57% say Amazon’s AI shopping assistant helped them purchase a product they were not previous

• 36.4% say the assistant increased their Amazon spending, while 44.5% say spending stayed about the same

And in the first table you can see how this ranks against Sparky and other Shopper Agents (AFS stands alone in the upper right quadrant for this survey).

Keep that 57% number in your head for this article. AFS is not just compressing a journey that would have happened anyway, AFS is creating incremental product discovery and new demand.

Why is Amazon moving AFS into the center of the experience and it’s eating the ‘old search’? Follow the incentives, they must be seeing the same lift in their own transactional data.

→ Evercore: Amazon up 4%

Walmart’s Earnings Highlight: John Ferner 🫶 Sparky 😃 (and so do Customers!)

Amazon’s not alone in their love of the shopper agent, both John Furner, Walmart’s CEO and Target’s CEO, Michael Fiddelke had positive things to say:

Note, John did not talk about the causality, but we get that from Evercore where the consumers in the survey explicitly tell us they spent more and/or found a new product they hadn’t considered using AFS. That points to perhaps Sparky seeing the same here.

Michael also announced they have a new CAIO who is going to up their AI game. They need it, look at where Target lands on the

Eric Seufert, Mr. “Agentic Commerce is a Mirage”, is now Mr. “Ummm, yeah, I like 😍 LOVE 😍Agentic Commerce when it’s Shopper Agents.”

Eric Seufert publishes the very popular “Mobile Dev Memo”, who is closely followed by Wall St, Stratechery’s Ben Thompson and pretty much everyone (I get a lot of ‘what’s your take on Seufert’s agentic commerce mirage piece’ comments from that group) outside of retail. Wellllll…. since his March ‘26 intro of this concept, he’s softened it a bit with every update and today, well, he hates answer engines and Agentic Commerce, BUT, he love love loves shopper agents (Sparky and AFS specifically). See highlighted area (highlight mine).

→ Agentic Commerce is a mirage (part 3) on Mobile Dev Memo (Sub required)

But he’s still hedging: “may only be viable”. Come on Eric, you have one foot on the Agentic Commerce Maximalist bus, come on aboard. Maybe by ‘update 4’ he’ll be there. Eric’s going to come up later, remember this one too!

Next stops: Andrew Lipsman, Sarah Marzano and Jason Del Rey 🚀

Williams-Sonoma 🫒: Hey Walmart and Target, hold our 🍺 (or I guess 🍷+🧀)

If you’ve been reading Retailgentic for a while, it won’t surprise you to see mass-market retailers AFS and Sparky in these pages, but what surprised me was specialty retailer Williams-Sonoma throwing their hat in the Agentic Commerce ring from their earnings call where they gave an update on their

What caught my attention: Williams-Sonoma says revenue associated with its Olive AI shopping assistant grew 620% since the start of the year.

Olive has joined the chat (and they are a Sous Chef!)

And that’s not the only number. Engagement with Olive grew 700%, shoppers who use it convert at 3x the rate of other visitors and personalized visits now generate roughly 9x the revenue of an average visit.

• Olive engagement: +700%

• Revenue associated with Olive: +620%

• Conversion rate for Olive users: 3x other shoppers

• Personalized visit revenue: roughly 9x an average visit

• Pottery Barn’s Otto assistant (looks exactly like Olive - maybe they are cousins?) resolves more than 70% of engagements without an employee handoff

→ Digital Commerce 360: Williams-Sonoma revenue from AI assistant grows 620% in Q2

I really like that olive emoji, here it is again so this is a 2 🫒 news item! 🍸

Summary: Retailer Shopper Agents Strike Back

The big picture from these August news hits: Agentic Commerce across Research → Find → Buy is showing up not just in discovery, but it’s hitting mass-market AND specialty retailer P&Ls→

• Consumer use of shopper agents is growing rapidly

• Shopper agents are driving incremental sales

• Shopper agents save money too by getting shoppers answers to their questions faster so the human associates can focus on the higher-level questions.

Those trends that are becoming very clear now, are the highlight of August’s news for me.

If I were a retailer, I would implement a parallel path here for Holiday 26:

• Path 1 - Prepare your catalog for consumption both externally and internally.

• Path 2 - Get a high-quality shopper agent live ASAP.

Share

August Trend 2: A Veritable Cornucopia of Consumer Surveys

At this point, you’re probably thinking: “Wow, that’s pretty compelling data for retailer, bummer it’s only for shopper agents, and we’re not seeing ANYTHING like that for answer engines. Gosh, if only Retailgentic had some data around THAT…”

Well, funny you ask that I have good news dear reader….

BCG: AI Expands the Brand Consideration Set

STATED PREFERENCE PLUS OBSERVED BRAND ANALYSIS | 13,000+ CONSUMERS, 12 MARKETS, 1,000+ BRANDS

BCG (Boston Consulting Group) released a big global consumer report and the AI section is 👨‍🍳😘. The TL;DR: 31% of consumers now use AI somewhere in the purchase journey, 19% are AI loyalists and 70% of those loyalists ultimately buy the products AI recommends.

BUT….. the most important finding is about incremental demand.

BCG says: AI introduces shoppers to brands they otherwise would not have considered in 63% of AI-assisted journeys.

This is the opposite of the ‘Agentic Commerce creates a race to the bottom’ argument. Agentic Commerce expands the consideration set and increases perceived value because it explains why a product fits the shopper’s needs. This data dovetails very nicely with the Mahaney Evercore report.

Here’s what this means for retailers →

• Agentic Commerce is becoming one of the fastest-growing sources of trust

• The answer engine increasingly builds the shortlist before the shopper reaches a brand site

• Clear product data along with evidence and a verifiable value proposition matter more than brand advertising

• Brands need to measure whether Agentic Commerce answer engines recommends them, to whom and why and when in the funnel.

This 100% validates our theory that you need to be doing GEO for top-of-funnel/sentiment side of things, but also ACO to make sure in the 3-4 turn discussions around product features, your products are still in the game.

Need more data on that, well, let’s shoot from Boston to Seattle for that…

→ BCG: Five Consumer Behavior Shifts Reshaping Growth in 2026

Microsoft Says Product Data Is Your Most Valuable Marketing Asset

Microsoft’s advertising group (CoPilot/Bing) published a practical guide for AI shopping, and the advice is almost exactly the ACO playbook:

• ·Upgrade and enrich your product catalog

• Implement UCP

• Upload your complete feed to Copilot Merchant Center

• Use Microsoft Clarity to understand behavior (this is their Google Analytics plus more competitor)

Microsoft may be a Tier 2 search engine today, but it is a Tier 1 enterprise distribution machine. When Microsoft tells retailers that product data has become a marketing asset, the idea has officially moved from an early-adopter conversation into the enterprise mainstream.

→ Microsoft: Prepare for the Age of AI Shopping

Narvar Finds a Huge Holiday AI Readiness Gap

STATED PREFERENCE | N=1,348 US CONSUMERS AND N=100 RETAIL DECISION-MAKERS

Holiday 2026 is shaping up to be the first truly agentic holiday, and Narvar’s fresh survey shows consumers are moving faster than retailers.

• 65% of consumers plan to use AI for at least one part of holiday shopping

• Only 8% of retailers are very confident in their ability to use AI to improve the shopping experience

• 43% plan to use AI for gift discovery

• 33% plan to compare products and summarize reviews

• 29% plan to use AI for budgeting and spend planning

• 15% plan to use AI after checkout for tracking and return reminders

The 65% versus 8% gap is the big story. Consumers are bringing agentic commece into the journey whether retailers are ready or not.

→ Narvar: 2026 Holiday Shopping Report

Morgan Stanley Adds Another Big Number to the 2030 Consensus

Morgan Stanley now says agentic commerce could expand the total ecommerce market by roughly 6% and materially influence about 20% of global ecommerce volume.

Their base market goes from nearly $5 trillion today to about $7 trillion within five years, with annual growth accelerating from roughly 7% to 9%. In their updated models, Agentic Commerce isn’t just taking share from ecommerce, it’s growing the pie.

That 20% influence number keeps showing up. Morgan Stanley, Bain, PayPal and Juniper all use different paths, but they continue landing in the same 15% to 25% strike zone by 2030.

Consensus estimates are formed, now the debate is moving from ‘if’ to “who is going to capture all this?”

→ Morgan Stanley: Is Agentic Commerce AI’s Next Catalyst?

FinanceBuzz Says 80% of Americans Use AI to Help Shop

STATED PREFERENCE | N=2,000 US ADULTS VIA PROLIFIC

FinanceBuzz and Bank of America surveyed 2,000 Americans and found that AI has already become a mainstream shopping tool:

• 80% have used AI to help make an online shopping decision

• 67% use it to find the best product for a need

• 67% use it to compare shopping options

• 58% use it to summarize reviews or sentiment

• 55% use it to find a better price

Trust still has a gap. Only 7% trust AI more than customer reviews, 36% trust them equally and 57% still trust reviews more.

That is not a contradiction. Consumers can use AI heavily without handing it complete authority.

Koddi Finds Consumers Want a Co-Pilot, Not Full Autopilot (yeah, which is what we’ve been saying 🤦‍♂️)

STATED PREFERENCE | N=750 CONSUMERS AND N=150 COMMERCE-MEDIA PROFESSIONALS

Koddi’s 57-page State of Agentic Commerce Media report is one of the first rational looks at the intersection of agents and commerce media.

The consumer message is permissioned delegation: 72% want AI to be a co-pilot rather than full autopilot, and 75% of US consumers are comfortable with AI helping choose what to buy or book. Only 32% have ever let a service automatically complete a purchase beyond a simple subscription.

The industry side is already moving. Nearly two-thirds of surveyed advertisers expect to invest between $50,000 and $999,000 in agentic commerce products over the next year, while 92% plan to spend on agent-specific measurement and diagnostics.

Commerce media is moving from visibility to selection. The new questions are not just ‘did the shopper see the ad?’ but ‘was the brand eligible, trusted, recommended, shortlisted and chosen?’

→ Koddi: The State of Agentic Commerce Media

Croud Shows AI Building the Shortlist Before the Product Is Chosen

STATED PREFERENCE | N=2000 CONSUMERS

Croud’s Validation Economy report has a clean way to think about the new journey: AI shapes the shortlist before the product is chosen.

Seventy-three percent of AI users turn to LLMs before choosing a specific product. They are not just searching for a SKU, they are asking by price or deal, need or use case, brand, style and then specific product.

The autonomy data is also moving faster than many people expect:

• 69% are open to AI buying without approval in at least one category

• 50% are comfortable with AI purchasing across three or more categories without approval

• 75% would use instant checkout where AI acts with final approval

This is the bridge between today’s assistant and tomorrow’s agent. Consumers do not flip from zero autonomy to full autonomy all at once, they grant permission by category, task and risk level.

The brands that build for instant checkout today are building the trust and product-data foundation for autonomous purchasing tomorrow.

→ Croud Consumer Index: The Validation Economy

Summary: Cornucopia of Consumer Surveys

The highlights:

• Evidence is piling up that Agentic Commerce is incremental to the industry overall (Morgan Stanley and to retailers individually (BCG). Agentic Commerce isn’t a mirage, hallucination or myth, it’s real, the ‘race to zero’ is the fake news.

• Trust is growing with consumers and Agentic Commerce

• Consumers want a shopping assistant (a shopping GPS is my preferred metaphor), not a ‘hands off the wheel’ solution - this fine. The hallucination crowd will use this data to say that Agentic Commerce is over (because they are the only ones that define it this way).

• Structured product data is the foundation for everything (remember, not all enrichment is created equal - details here).

Subscribe now

August Trend 3: Interesting Industry Moves

OpenAI and Ads part 1

OpenAI announced that their ad platform has crossed $1B after launching only 200 days ago. To put that in perspective, it took Amazon about 5 years (2012-2017) to get there. That’s a good framing for this story, which answers the question…why?

Mr. “Hates Answer Engines and Agentic Commerce…But 🥰 Answer Engines and Ads…”

Eric Seufert (told you it would come up again!) has a piece out that has a great analysis of which platform is best for ads. It’s a matrix of “capacity to use off-platform behavioral data for targeting” by “strength of intent revealed through product usage.” The result - Chatbots come out on top. They combine the best aspects of commerce media (intent through use of product) and social networks (behavioral targeting) to create a super ad platform unlike anything we’ve ever seen.

→ Mobile Dev Memo: The Privileged Position of chatbot advertising

🌶️ Stripe Says Checkout Pages Are Going Away…🌶️

A Stripe interview went viral for a very spicy take: ‘Eventually checkout pages will go away.’

That does not mean checkout disappears. It means the visible page gives way to agent-controlled transaction flows where identity, intent, payment credentials, shipping information and approval travel through protocols and APIs.

The bigger picture? Platforms that own the checkout page today could lose a major control point, while the companies that own agent identity, payment orchestration and verifiable intent gain one.

We just wrapped our big Shopify 2 parter, this is the biggest question I get from Wall St: What happens to Shopify when checkout goes away? Only time will tell, Will think it’s coming pretty quickly.

→ Business Insider: Stripe says checkout pages face an agentic future

Meta’s Hatch Could Put a Consumer Agent Inside Instagram and WhatsApp

Meta is reportedly preparing a consumer agent called Hatch that could compare products, book restaurants, follow deliveries and coordinate tasks across apps. A separate agentic shopping experience for Instagram may bring conversational discovery and buying directly into Meta’s commerce and creator ecosystem.

When I put the pieces together, Hatch looks like a consumer-friendly OpenClaw mixed with TikTok’s Tako and Meta’s massive distribution. The big question is whether the commerce layer will run on product feeds, browser actions, UCP or a combination.

This one is still speculative, so keep it in the watch bucket. This seems close as several pieces of Hatch are starting to appear. I expect it in early September.

→ YourStory: Meta is building Hatch, an AI agent that can act for you

Google is Morphing Core Search Into AI Mode

Google keeps moving AI Mode closer to the center of Search. Each interface change looks incremental on its own, but the direction is unmistakable: AI-generated answers are becoming the primary discovery layer, not a side experiment.

For retailers and brands, this means the classic blue-link playbook keeps losing surface area. If the answer engine builds the consideration set before the consumer clicks, your product data, brand evidence and agent readability determine whether you make the list at all.

We’re seeing a pattern here. Google is not replacing Search overnight, it is steadily teaching billions of people to shop through an AI-mediated interface.

→ Barry Schwartz on Google’s latest AI Mode interface test

That’s a Wrap

These three trends point to a continued rise in Agentic Commerce from all angles: Answer Engines are growing nicely and consumers are getting deeper in their adoption. The same thing is happening with Retailer shopper agents. Finally, industry players continue to expand existing offerings and Holiday 26 could give us META and Claude in the market in a big way, and maybe even Grok.

We’re taking the week of Labor day off for a last break before it gets crazy. We’ll be back in your inbox with a pod on 9/10 and a fresh content 9/15 (unless there’s something FLASH worthy, then all bets are off!).

Everyone have a great Labor Day!

Leave a comment

Subscribe now

Caila Schwartz is Head of Agentic Commerce Shopper Insights at Salesforce, a brand-new title that tells you where the industry is heading. She’s been at Salesforce for nearly 12 years, arriving through the Demandware acquisition in 2016, and her job is to turn Salesforce’s massive data footprint, platform data from 1.5 billion global shoppers plus regular surveys of consumers and commerce professionals, into a read on the state of the consumer and the state of the industry.

Listen/Watch Our Interview:

This week’s episode is a data deep dive… two of them, actually. Salesforce just dropped a back-to-back data release: the latest edition of its State of Commerce report and its Holiday 2026 Predictions. Caila Schwartz, Salesforce’s newly minted Head of Agentic Commerce Shopper Insights, walks us through both. Think of it as a rearview mirror on the first half of 2026 and headlights into the holiday season, with agentic commerce showing up in nearly every number.

Highlights

• 𝗧𝗵𝗲 𝟮𝟬𝟬% 𝗷𝘂𝗺𝗽: 15% of shoppers now start their buying journey with an LLM or agent, up from 5% last spring.

• 𝗛𝗮𝗹𝗳 𝗼𝗳 𝘀𝗵𝗼𝗽𝗽𝗲𝗿𝘀: About 50% have used a general-purpose LLM, social media agent, or branded agent somewhere in their buying journey in the past 12 months, a 67% increase over last year.

• 𝗖𝗼𝗻𝘃𝗲𝗿𝘀𝗶𝗼𝗻 𝗸𝗶𝗻𝗴: Agentic search traffic converts at roughly 2.5–3.5%, about 2–3x overall conversion and far above social’s sub-1%.

• 𝗧𝗵𝗲 𝘁𝗿𝘂𝘀𝘁 𝗻𝘂𝗺𝗯𝗲𝗿𝘀: 74% of shoppers trust agent recommendations, and over 80% feel more confident in purchases made after interacting with a brand-owned agent like Rufus or Sparky.

• 𝗧𝗵𝗲 𝗿𝗲𝗳𝗲𝗿𝗿𝗮𝗹 𝗯𝗹𝗶𝗻𝗱 𝘀𝗽𝗼𝘁: Discovery is happening inside conversations that never send a click back to the brand site.

• 𝗧𝗵𝗲 𝗖𝗮𝗶𝗹𝗮 𝗙𝗶𝘃𝗲: Salesforce’s Holiday 2026 predictions: non-human traffic, branded agents everywhere, social commerce acceleration, the store’s new hybrid role, and the free-shipping squeeze.

• 𝟮𝟬% 𝗻𝗼𝗻-𝗵𝘂𝗺𝗮𝗻 𝘁𝗿𝗮𝗳𝗳𝗶𝗰: This holiday, Salesforce predicts about one in five e-commerce visits will come from crawlers and agents, not people.

• 𝗕𝗿𝗮𝗻𝗱𝗲𝗱 𝗮𝗴𝗲𝗻𝘁 𝗿𝗮𝗰𝗲: About a third of B2C orgs have agents deployed today, over half plan to deploy within six months, and by Cyber Week one in three e-commerce sites will have a branded agent on site.

• 𝗦𝗼𝗰𝗶𝗮𝗹 𝗴𝗿𝗼𝘄𝘀 𝟵𝗫: Social commerce is predicted to grow nine times faster than traditional e-commerce this year, and it’s high-income shoppers, even more than Gen Z, saying they’ll buy in-app this holiday.

• 𝗛𝘆𝗯𝗿𝗶𝗱 𝗵𝗼𝗹𝗶𝗱𝗮𝘆: 37% of holiday dollars will be “hybrid sales” that rely on both digital and physical to close the deal, 80% of shoppers use their phone while standing in your store.

• 𝗗𝗮𝘁𝗮 𝗶𝘀 𝘁𝗵𝗲 𝗯𝗲𝗱𝗿𝗼𝗰𝗸: Four in five commerce pros say their tech stack grew over the last two years, making data unification the #1 barrier to good agents.

Caila brings the receipts, real numbers on a shift most of us are still trying to eyeball. We’ll get her back in the fall when the actual holiday forecast drops. Enjoy the conversation.

👉🏻 Check out the State of Ecommerce

👉🏻 Look at the Holiday 2026 Predictions

🎧 Listen wherever you get your podcasts: Spotify, Apple, etc.

Or, we have a video version on YouTube (make sure to subscribe!) →

Want to Learn More?

Stay on top of what’s happening in agentic commerce, from the tools powering it to the founders building with it by following us across platforms:

• Podcast: https://retailgentic.transistor.fm

• YouTube: https://www.youtube.com/@Retailgentic

• Instagram: https://www.instagram.com/retailgentic/

• X: https://x.com/retailgentic

• TikTok: https://www.tiktok.com/@retailgentic

We're here to make sense of what’s next in retail + AI and bring you along for the ride. Subscribe and stay tuned!

Welcome to the last piece of a two part series where we’re going deep into what Shopify is doing in Agentic Commerce.

• 8/12: Part 1 - In Part 1, we started by looking at Shopify’s Q2 results through the lens of Agentic Commerce - it’s helpful to understand what makes Shopify ‘tick’ what drives their economic engine (checkout/ShopPay) and specifically what they said about Agentic Commerce. Available Here.

• 8/25: Part 2 - In Part 2, we’re diving deep into Shopify’s Agentic Catalog offering. We’ve talked to hundreds of larger Shopify store operators and there is a LOT of confusion and misunderstanding around what Shopify Catalog does and doesn’t do and how it does it.

Part 1 Recap

In part 1, we looked at Shopify’s strong Q2 through an Agentic Commerce lens and the agentic commerce data was impressive:

• AI-driven traffic to Shopify stores grew 3x y/y.

• Orders originating from AI searches grew 3x y/y.

• New-buyer orders from AI channels came in at nearly 2x the rate of other channels.

• Half of AI-referred sessions landed directly on a product detail page, 2.5x the rate of traditional search.

• 75% of AI-attributed orders came from outside Shopify’s top 100 categories.

• Shopify also said AI searches powered by Catalog converted at 2x the rate of searches using scraped product data.

That last datapoint is the springboard into today’s deep dive.

We’ve talked to hundreds of larger Shopify store operators and there is a LOT of confusion around Catalog. Is it a product feed? Is it the data already sitting in Shopify Admin? Is it another sales channel? Is it the API developers use? Does every AI engine see it? Can a merchant control it? What happens to all those meta datafields teams have spent years building?

Those are all good questions because Shopify uses the word “catalog” in several different ways.

Before we get super-tactical, let’s look at the strategic picture.

Subscribe now

The Shopify Big Picture

Shopify doesn’t disclose their merchant numbers, regional details for merchants or regional GMV, but you can get very close on both based on public disclosures and Wall St. estimates.

For merchant count, there are 4-5 companies that crawl the internet, scan the html for signals of the underlying technology being used and report details on what they find. According to these services, there are about 3m active Shopify stores globally. Of those 3m stores, 1.1m are in available to US consumers and of THOSE ~15k are the largest stores that use the more enterprise-ready Shopify Plus platform.

On the GMV side, Shopify does disclose GMV quarterly, but not by region, but they give regional growth and some other datapoints annually that make it relatively easy to build a model that’s a good guesstimate. By triangulating on these models our Retailgentic forecast for GMV for 2026 is right at $500B globally. If we take out POS GMV sales and some other pieces and also look at the US, we get a US GMV prediction of $265B for 2026.

Using these numbers - 1.1m US merchants and $265B in 2026 GMV we can then look at the distribution curve.

Merchant Distribution ➡️ Ecommerce Power Law

There are many different distribution curves. A popular distribution curve is a bell curve you’ll potentially remember from college curves. Shopify is an extreme head/tail story. The top 15k (or 1.4%) of stores drive 80% of the GMV ($212B).

That yields:

• Top 15k stores average GMV/yr: $14.1m GMV.

• Bottom 1,085,000 average $50k/yr GMV.

There are many gradients and tiers that would make this more granular, but this binary comparison gives you a feel for how stark the head and tail of the Shopify merchant base are.

The long tail doesn’t stop there, it’s in the category mix too.

Shopify Category Mix - Another Long Tail

If you look at those 1.1m US stores and put them in general categories you get the pie chart above. The top 3 store categories are:

1. Apparel (28%)

2. Home and Garden (11%)

3. Beauty and Fitness (10%)

From there you’ll see it gets long-tail quickly with 19% of the stores in a category with less than 4% share (marked ‘Long Tail Categories” on the pie chart).

Implications of the Shopify Platform Long Tail

Because they have so many customers with diverse needs, Shopify has to act like a ‘platform’ - finding the center of the bullseye for the majority of their merchant features. On top of that they provide APIs, headless capabilities and the app store which give the merchants that fall outside of that center of the bullseye to build their own solutions.

Said another way, it does not make economic sense to build features just for the 10k largest stores. They do some of that in Shopify Plus, but because the subscription part of their business is now dwarfed by the checkout (tied to penetration of GMV by ShopPay which also favors smaller merchants), there’s increasing incentive misalignment for enterprise-specific features.

As we dig deeper into the Shopify Catalog, features, you’ll see that some of them are not able to be controlled 100% by merchants. While this frustrates many of the largest Shopify users, economically it makes a lot of sense for Shopify to build them this way.

Before we get into the how Agentic Catalog works, we want to introduce a tool that has helped us immensely to understand and visualize how all this works….

Introducing: Shop Cat Explorer

To help understand the Shopify Catalog API and everything made available to agents, because the API is openly available and well documented, I was able to build an ‘Explorer’ App that lets anyone explore the Shopify Catalog on your own. It’s the Shopify Catalog Explorer (SCE), which is a mouthful, so I shortened it to Shop Cat Explorer. The way it works is first go to: https://www.shopcatexplorer.ai and you can start exploring on your own. Here’s a quick getting started guide.

Getting Started with Shop Cat Explorer

SCE has a four-pane user interface:

In the Starting Pane, you choose your geography, then if you want to look at the Shopify Catalog API results from a store perspective, a brand view or a product view.

Shop Cat Explorer Scenario One: Store View

Here we look at the store stevemadden.com which flips the other panes to:

• Browse Pane: Shows the store’s categories and price range filter

• Product Pane: Here all the products are listed - the top product is ranked by Shopify as the most relevant, and therefore the one the API returns first to the agent and so on. The number of products reported by the agent is at the top of the pane. As you scroll, it could pull more as the API ‘pages’ products.

• Product Details Pane : In the product details pane, for the selected product from the Product Pane, you’ll see all the SKU-level detail the API exposes to the agent including:

  • Product Description

  • Conversational attributes - These are the “Inferred product fields” (metadata.attributes, metadata.tech_specs, metadata.top_features, metadata.unique_selling_points)

  • Reviews (summary - like 4.3/5) - We know this is in the catalog API, but rarely surface it - still exploring why not consistent in shop cat explorer..

  • Reviews - review-level details - we’ve seen these showing up in UCP sometimes, not sure how that’s happening if not supported here, consider this an area we’re still exploring or a WIP.

  • Variants returned

  • Agent Visibility Check- Here we check 6 basic attributes and if all 6 available, we mark it ‘not visible’. The 6 attributes are: image, description, price, SKU, Rating, compare-at-price. Note: This is our Shop Cat Explorer algorithm/signal not Shopify’s.

  • Catalog field/value data - Data for all catalog attributes provided by the API

Shop Cat Explorer Scenario Two: Brand View

In this example, we look for the beauty brand: ELF, this results in:

• Browse Pane - Shows there are 51 stores carrying the brand, and lists the top 12

• Product Pane - Shows the ELF products returned from those 51 stores in order from top relevance to least (this is the order sent to the agent calling the catalog API).

• Product Details Pane - same information as the store view - the 7 sub-categories of product-level details for the selected product.

Shop Cat Explorer Scenario Three: Product View

In this example we look for the hottest fashion accessory this Summer: “Flip Flop Heels” and it turns out there are 346 core SKUs across the Shopify Catalog.

• Browse Pane - Shows there are 136 stores carrying flip flop heels, and lists the top 12 stores.

• Product Pane - Shows the flip flop heels from most relevant to least.

• Product Details Pane - same information as the store view - the 7 sub-categories of product-level details for the selected product.

Shopify Catalog - The Basics

There are four parts to the Shopify Agentic Commerce Stack. They are:

Think of Shopify Catalog as the core product database. The Catalog API is the access layer for that product data, Agentic Storefronts give merchants the controls and an user interface into what’s going on. For Agentic transactions, UCP is used for the transaction rails.

Shop Cat Explorer is your detailed view into the Catalog API. You are seeing exactly what shopify is surfacing to the agents.

Inside of the Shopify catalog, Shopify has spent nearly two years building a search index that now covers more than 1 billion products. Under the hood, Catalog does several things → (these are Shopify’s terms not mine):

• Standardization: Catalog maps products into Shopify’s Standard Product Taxonomy.

• Enrichment: Catalog infers attributes such as color, material, pattern, size type, age group, key features, technical specifications and value propositions.

• Clustering: Catalog groups related listings and variants, including products sold across different stores. (Note: this is what I prefer to call canonicalization).

• Freshness: Catalog keeps the product record connected to current price, inventory, variants and other commerce data.

• Retrieval: It lets agents search for a product from natural language, look up a known product or retrieve current variant details.

Shopify’s engineering team has given a revealing look at the clustering solution. Catalog groups versions of the same real-world item under a Universal Product ID (UPID). Think of UPIDs as the equivalent of an ASIN in the Amazon world. Note: Sometimes Shopify uses UPI - Universal Product Identifier and other times UPID, we’re 99.9% sure they are interchangeable, but it’s quite inconsistent.

For example, you can see these in the Shop Cat Explorer- here I looked for a popular beauty product: “Olaplex 3” (use the Products slider):

In Product Details Scroll down to variations. Here you’ll see in the variations something different than you’ve probably seen before→

Notice how each variation has a different seller. This is Shopify’s canonicalization clustering at work. This is where the API stops, I am a bit concerned that it’s clustering things with wildly different prices (feels like a classic mis-canonicalization- the 3.4 and the 8.5oz being mixed accidentally), but without more information it’s hard to tell.

Merchant Controls

Our readers are merchants, and here’s what you want to know, what can you control→

• Eligible products are generally included automatically; inclusion does not guarantee placement or ranking (the agent is the front-end, it decides what to do with the catalog data passed to it).

• The Agentic dashboard lets merchants preview Catalog search, examine listing quality, manage channels, and review performance.

• Catalog Mapping lets merchants select title, description, category, and grouping sources when custom metafields or conventions are used.

• Some API fields are inferred by Shopify and can vary in accuracy; downstream AI channels control their final wording and ranking.

• Merchants can disable Catalog access for third-party agentic channels, but cannot opt out of Shopify Catalog itself while retaining otherwise eligible public products.

• Promoted placements can blend paid variants into Global Catalog results for approved developers.

Shopify Catalog and ACO

If we look at our 11 step Retailgentic ACO strategy steps:

The Shopify platform gives you steps 1-5 ‘for free’. You have to provide the basic attributes and you can use merchant controls to optimize your basic attributes.

For steps 6 and 7, Shopify has made what I’d call at a ‘platform choice’ - they create extra product content for you, without your review and these cannot be turned off or changed. While this is probably the best decision for the 1.1m sellers, for the largest sellers, those top 15,000, if you want to control your own conversational attributes, review them for errors, this enrichment is not under your control.

Note there is no monitoring of how agents display your cards, product-card level canonicalization, external-to-shopify context capture.

Shopify Enrichment Level

If you recall the 5 levels of product enrichment article where we introduced the five levels of product enrichment. Using the Shopify Catalog, you can get to Level 1 and the Shopify enrichment (they call it inferred enrichment) gets you to Level 2, but as said above, it is not merchant controllable at this time.

11 Shopify Catalog FAQs with Answers

We combed through questions we’ve been asked and the most common questions in larger Shopify store online communities and

Q1. Is Shopify Catalog the same as my Markets or B2B catalog?

No. Shopify Catalog is the global product-discovery layer used by Shop and AI shopping experiences. A Markets or B2B catalog controls the assortment and prices visible to particular countries, markets, companies, or company locations.

If you are working in Sales channels → Agentic, you are dealing with global Shopify Catalog distribution. If you are configuring country-specific assortments, wholesale prices, or company-location access, you are working with Markets/B2B catalogs.

Q2. Are my products included automatically?

Generally yes, when the store and products are eligible. The store must be on Starter or higher, not password-protected, compliant with Shopify policies, and in acceptable account standing. Products need a title, at least one image, a price above zero, an identifiable product URL, and publication to Online Store, Hydrogen, or Headless.

Unlisted, search-hidden, sensitive, or ineligible products are excluded. Account standing considers operating history, genuine sales, fulfillment, chargebacks, invoice payment, and unresolved compliance issues. Non-Shopify brands using the Agentic plan must also provide an external product URL.

B2B-only products, products requiring customer authentication, and password-protected products are generally excluded because current Agentic Storefronts support direct-to-consumer sales.

Q3. Where can my products appear, and does checkout happen inside the AI?

Depending on eligibility and channel availability, products can appear in Shop, ChatGPT, Microsoft Copilot, Google AI Mode, Gemini, Meta experiences, and smaller Catalog-powered shopping agents.

ChatGPT currently operates as a discovery and referral channel: customers complete their purchase through the merchant’s Shopify checkout in an in-app browser or a new tab. Microsoft Copilot, Google AI Mode/Gemini, and Meta can support Shopify-powered direct checkout where the channel and merchant are eligible and direct checkout is active.

Google and Meta product discovery are supplied through their respective Shopify sales-channel integrations rather than the same Catalog-access toggle used for ChatGPT and Copilot.

Q4. What data is shared, and does an AI receive my customer database?

Catalog-access channels receive product information such as titles, descriptions, images, prices, options, variants, and availability. They do not receive the merchant’s private admin information, general customer database, full order history, or orders from unrelated channels.

When a customer places an order through a direct-checkout channel, that channel receives the contact and delivery information required to process the order. Shopify’s supplemental terms authorize Shopify to process, index, combine, display, reproduce, and create derivative representations of merchant materials through participating channels.

API developers are separately prohibited from caching Catalog search results or downloading and reusing merchant images. That is intended to keep representations aligned with the merchant’s current price, availability, and presentation preferences.

Q5. Does “compression” alter my images, descriptions, or storefront?

Compression is a misinterpretation of some of the engineering team’s articles that talk about “compressing discovery” and “journey compression” (for the shopper). In this piece on LinkedIn called
Shopify Catalog functionality”, the Shopify engineer uses “complexity compression” - meaning that Shopify saves you all the complexity of DIY’ing all these attributes.

That being said, there are two aspects of the functionality that are out of the Merchant’s control that can introduce errors, not what I’d call compression though - they don’t ‘shrink the size of the catalog’ like compressing a file.

• Inferred categories (sometimes called enrichment)

• Clustering - If Shopify erroneously clusters your SKUs, that can introduce ‘up-stream’ problems for you.

I would not call these two areas ‘compression’, I’d call them AI generated information and mappings that have no human-in-the-loop and can introduce errors.

Q6. What if Shopify groups or describes my product incorrectly?

Open Sales channels → Agentic → Sources and review Shopify Catalog Mapping. All stores can map the Catalog title, description, and category to preferred product fields, metafields, or metaobjects.

Stores with unusual product structures can configure custom variant grouping using product attributes, metafields, title delimiters, and other conventions. This is useful when every color is maintained as a separate Shopify product but should appear as one configurable product in AI search. Shopify Combined Listings is used by default where it is already configured.

Merchants cannot guarantee every inferred field or downstream AI sentence. Shopify recommends that for persistent false grouping, capture the exact search query, affected product URLs, and the expected grouping, then provide those examples to Shopify Support. Alternatively, you can take over your channel feeds from Shopify and do ACO yourselves.

We recommend using our free Shop Cat Explorer to get a view of all these features so you can see if you find any errors in the Shopify inferred attributes or clustering for your products.

Q7. How do I improve my products’ AI ranking?

Use the search-preview feature in Sales channels → Agentic. It shows raw Shopify Catalog results and whether the store’s products reach the global top results for a query.

Shopify’s listing-quality indicators evaluate description completeness, image coverage, verified reviews, variant and option completeness, in-stock availability, readable option names, and store policies. Ranking also uses less controllable signals such as popularity, engagement, and brand recognition.

Individual AI platforms can re-rank Shopify’s output, so the preview is directional rather than a guarantee. Inclusion in Catalog is only eligibility; competitive visibility requires relevant, complete, trustworthy product information. Promoted placements may also be blended into Global Catalog results for approved developers.

Q8. Can I opt out?

Merchants cannot opt out of Shopify Catalog itself while retaining otherwise eligible public products, but they can control which third-party agentic channels receive Catalog access.

In Sales channels → Agentic, turn off Allow Shopify to manage for me. You can then disable global Catalog access, future automatic enrollment, or individual supported channels. Removal may take up to seven days. Disabling third-party access does not remove products from Shop (The Shop app) and does not prevent ordinary web crawlers from discovering a public product page.

To hide a specific product from Catalog, Shop, and search engines, use Unlisted status or the appropriate search-hidden setting. Changing robots.txt affects web crawlers, but it does not stop Catalog syndication to channels that remain activated.

Q9. How current are price and inventory data?

Shopify describes Catalog data as continuously updated and real-time, but the public materials do not provide a contractual freshness SLA measured in seconds or minutes.

Search defaults to available, sale-ready products. lookup_catalog retrieves current records for known products or varient identifiers, and get_product returns current variants, availability, seller details, and checkout links. Developers are prohibited from caching results or copying product images to their own servers.

The architecture should be materially fresher than periodic scraping or CSV feeds, but real-time should not be interpreted as a guarantee of zero propagation delay.

Q10. How do I know whether Catalog is producing sales?

The Agentic dashboard reports sales, orders, online-store sessions, online-store conversion, queries where products appeared, and queries where products failed to rank competitively.

Orders from supported AI channels appear in Shopify Admin with channel or referrer attribution. Shopify says merchants retain ownership of the customer relationship and post-purchase experience. Headless-store Agentic analytics are not yet available.

Merchants should separate AI answer-engine traffic from conventional organic search, compare conversion and average order value, and evaluate both total volume and per-session economics. Early AI traffic may be small while still carrying higher purchase intent.

Q11. Does Shopify Support the Q+A Functionality of ACP/UCP?

We’ve scoured the documentation and looked heavily through the raw data from Shop Cat Explorer and do not see evidence that Shopify has worked on this functionality at this point. They have something that looks kind of like UCP/Google conversational attributes, but have a different name (inferred product fields) and different individual fields than those recommended by Google.

We’ve found optimizing ACO via Q+A to be extremely powerful with measurable impact and something merchants should prioritize as they get more Agentic Commerce savvy.

Merchants will note that Shopify does have a merchant-managed answer functionality via the Knowledge Base App. These are store-level questions, we’re talking about product-level questions here. E.g. “What’s your return policy” is a store-level question vs. “Is this product safe for overnight application on my skin” is a sku-level FAQ.

Where to Learn More…

That’s a LOT on Shopify Catalog, we hope you’ve enjoyed this overview and connecting the tactical to the strategic. We also hope you find Shop Cat Explorer as helpful as we did to understand how all this works and ‘see’ it live and in action against your store, brands and products.

Also, to their credit, Shopify has published a TON of very detailed content around the Catalog, Catalog API, UCP, Agentic Commerce and all their capabilities. Here are the top items we used for this piece in our research:

Related Announcements and news

• Agentic commerce for every developer: The Spring ’26 Edition - Public Catalog API access, multimodal search, richer product data, Shop sign-in, and developer monetization.

• AI-native, developer-ready: Unpacking Winter ’26 Edition - The broader developer rollout of Shopify Catalog and the agentic commerce stack.

• The agentic commerce platform: Shopify connects any merchant to every AI conversation - UCP, the Agentic plan, and Catalog availability for non-Shopify brands.

• Millions of merchants can sell in AI chats - The March 2026 merchant rollout across ChatGPT and other AI channels.

• Selling everything, everywhere, all at once: The Spring ’26 Edition - Merchant-facing Catalog, Agentic dashboard, analytics, and distribution announcements.

• Five apps that show what Catalog API and UCP make possible - Examples of visual, contextual, conversational, and cross-merchant shopping experiences.

Engineering and analysis

• Clustering billions of products for agentic commerce with Catalog API - Detailed architecture for product identity, clustering, precision and recall, LLM processing, and structured outputs.

• AI-referred shoppers convert better and spend more - Shopify’s Q1 2026 data on AI traffic, conversion, average order value, and journey compression.

• Millions of merchants speak UCP - A technical explanation of UCP and Catalog as complexity compression from Shopify product leader Ilya Grigorik.

Catalog API documentation

• About Catalog interfaces - Global versus Storefront Catalog, tools, usage restrictions, saved catalogs, and promoted placements.

• Global Catalog MCP reference - Global endpoint, search, lookup, get-product operations, inferred fields, and request shapes.

• Storefront Catalog MCP reference - Product discovery scoped to a single Shopify storefront.

• Global Catalog extension and filters - Shopify-specific filters, inferred metadata, availability, shipping, rating, and price tiers.

• Agent profiles and UCP negotiation - Capability discovery and UCP profile requirements.

• Authentication and rate limiting - Anonymous, signed, and token traffic tiers and their capabilities.

• Discover products with Shopify Catalog - End-to-end developer example for Global Catalog search.

• Storefront Catalog MCP now implements UCP - The April 2026 endpoint, tool-name, and schema migration.

Merchant documentation

• Shopify Catalog overview - Definition, syndication behavior, optimization guidance, and ranking limitations.

• Requirements for inclusion in Shopify Catalog - Store, product, publication, policy, and account-standing eligibility requirements.

• Shopify Catalog and product discovery for Agentic Storefronts - Product mapping, crawler behavior, hiding products, discovery files, and B2B exclusions.

• Mapping product-data sources for Shopify Catalog - Field mapping and custom variant-grouping controls.

• Shopify Agentic Storefronts overview - Channels, checkout behavior, attribution, and customer relationship ownership.

• Managing Agentic Storefronts - Channel controls, opt-outs, performance reports, search preview, and listing-quality signals.

• Data sharing and privacy for Agentic Storefronts - Product and order information shared with AI channels and protected merchant data.

• Agentic-plan setup for non-Shopify storefronts - Requirements and setup steps for brands using Shopify’s Agentic plan.

• Optimizing your store for AI - Product-page, structured-data, FAQ, and Knowledge Base guidance.

Related Shopify catalogs

• Catalogs for Shopify Markets - Developer overview of market and company-location catalogs.

• Markets catalog management - Merchant guidance for market-specific products and pricing.

• Admin GraphQL Catalog reference - The separate Admin API object used for Markets, B2B, publications, and pricing.

Share

Subscribe now

Daniel Vitiello is CEO and co-founder of Cooklist, the Austin-based company behind a home inventory and meal planning app with over a million users and, increasingly, the AI shopping assistant running on grocery retailers’ own websites.

Daniel leads all of Cooklist’s technical teams and wrote the first couple of years of the platform himself, including the retailer loyalty-card integrations.

Cooklist went through the Techstars Retail Accelerator in partnership with Target in 2018, raised a priced round led by Mercury Fund in 2021, and has been profitable since with five straight years of triple-digit revenue growth.

Listen/Watch Our Interview:

This one is a build deep dive. Daniel has spent eight years on the same problem, and the answer has changed shape three times along the way: a consumer app in 2018, a shoppable recipe platform in 2023, and now a conversational shopping assistant running inside grocers’ own sites.

We go deep on the plumbing here: product catalogs, ontologies, eval systems, model selection, retail media, and what an agent does when a shopper types “taco night.” If you like knowing how these things actually work under the hood, this is your episode.

Highlights

• 𝗧𝗵𝗲 𝗻𝘂𝗺𝗯𝗲𝗿𝘀: Live in over 700 stores, reaching over 10 million shoppers, with the number of locations set to double before the end of the year. The consumer app has over a million users.

• 𝟰𝟭 𝗱𝗮𝘆𝘀 𝗰𝗼𝗻𝘁𝗿𝗮𝗰𝘁 𝘁𝗼 𝗹𝗶𝘃𝗲: A one-line JavaScript install that reuses the retailer’s existing catalog, add-to-cart, analytics, and retail media. National scale in six weeks.

• 𝗦𝗲𝗮𝗿𝗰𝗵 𝗮𝗻𝗱 𝘀𝗲𝗹𝗲𝗰𝘁 → 𝗿𝗲𝘃𝗶𝗲𝘄 𝗮𝗻𝗱 𝗮𝗰𝗰𝗲𝗽𝘁: Daniel’s framing for the whole shift. The old era: you search one thing, you select it. The new one: the agent assembles a solution and you approve it.

• 𝗧𝗵𝗲 𝗲𝘃𝗮𝗹 𝘀𝘆𝘀𝘁𝗲𝗺: Thousands of test scenarios with grading criteria aligned to each retailer’s merchandising principles, run tens or hundreds of times to get a distribution of responses.

• 𝗧𝗵𝗲 𝗳𝗶𝘀𝗵 𝗿𝘂𝗹𝗲: Retailers write procedures the agent pulls when a scenario matches. One of Daniel’s favorites: if you’re cooking fish today or tomorrow, it recommends fresh. If it’s later in the week, frozen.

• 𝗧𝗿𝗮𝗶𝗻𝗶𝗻𝗴 𝗮 𝗻𝗲𝘄 𝗱𝗶𝗴𝗶𝘁𝗮𝗹 𝗲𝗺𝗽𝗹𝗼𝘆𝗲𝗲: Onboarding is essentially an interview with the business. What are your values? How do you respond? What are your merchandising principles? That becomes the agent’s identity and personality.

• 𝟭,𝟬𝟬𝟬 𝗙𝗼𝘂𝗿𝘁𝗵 𝗼𝗳 𝗝𝘂𝗹𝘆 𝗯𝘂𝗻𝗱𝗹𝗲𝘀: Instead of one holiday carousel every shopper sees, merchant teams define ~20 axes — price, dietary needs, brand preference — and an agentic process generates permutations personalized against transaction history.

• 𝗔𝗜 𝗖𝗵𝗲𝗳𝘀 𝗼𝗻 𝗧𝗶𝗸𝗧𝗼𝗸: Long before ChatGPT, Cooklist ran a TikTok series generating recipes with GPT-2 and early GPT-3. The models got good at basic recipes so fast that the fun moved to challenging them to be creative.

Daniel is one of those founders who has clearly built the thing himself, and it shows in the answers. We learned about a thousand things from this one. Enjoy the conversation.

🎧 Listen wherever you get your podcasts: Spotify, Apple, etc.

Or, we have a video version on YouTube (make sure to subscribe!) →

Want to Learn More?

Stay on top of what’s happening in agentic commerce, from the tools powering it to the founders building with it by following us across platforms:

• Podcast: https://retailgentic.transistor.fm

• YouTube: https://www.youtube.com/@Retailgentic

• Instagram: https://www.instagram.com/retailgentic/

• X: https://x.com/retailgentic

• TikTok: https://www.tiktok.com/@retailgentic

We're here to make sense of what’s next in retail + AI and bring you along for the ride. Subscribe and stay tuned!

Subscribe now

The quick history of ChatGPT and checkouts has been:

• Phase I: ChatGPT announced in 2025 they were doing their own checkout, ACP, etc.

• Phase II: ChatGPT abandoned/paused that in Spring 2026 in favor of ChatGPT apps integration - this was not, ummm, how best to say it, “an ideal customer experience”. In engineering parlance it was Janky.

• Phase III: Begins today! ChatGPT apps have been renamed to Plug-ins for a while and now there’s a new, much more integrated experience. We’ve been able to finish the first checkout from start to finish. Let’s jump in!

Live Video Demo (3mins)

Detailed Walk Through

It all starts, as always in Agentic Commerce with an offer card. Here I’m looking for a monitor→

When I click on the Samsung 34” I have this offer card→

You’ll note that BestBuy stands out with the “Open” button.

Initiating checkout via open button

When you click on that button, it pretty quickly pops up a ‘in-flow’ checkout experience→

From there you enter your shipping→

And then payment

And you are done.

Conclusion

This is a very clean and basic checkout. Yes it’s one item, no there’s no loyalty, but this is a foundational very good experience that is at/beyond the ACP experience. From here it’s not a huge lift to add multi-cart and loyalty connects, etc. We’ll be watching to see if they go there next - stay tuned!

Thanks!

Thanks to our friend Joe, the Agentic Commerce Sleuth for the tip on this one and pointing us the right way!

In early 2026, I had what I’d call ‘a wacky founder’ idea. The kernel of the idea goes back to my ChannelAdvisor days, specifically 2004, when Internet Retailer Magazine (!!) launched the first IR300 – a ranked listing of the top 300 retailers selling online with this new-ish thing called: “e-commerce” (obviously a fad!!).

Both retailers and vendors were excited. Retailers loved seeing the benchmarking data and what vendors everyone was using. As vendors it was super helpful showing the world that we were so dominant in the ‘marketplace management category’. From there the list expanded in depth (ultimately stopping at the IR2000) and width (different countries, regions, etc.). IR was acquired, the remaining piece separated from the tradeshow, was re-branded Digital Commerce 360 but the IR1000 list is still an annual publication and on its 22nd year of publication (available here).

Back to the wacky founder idea… That 2004-inspired idea was: “What if we took the IR1000 and we added multi-dimensional benchmarking data for Agentic Commerce?”. The DC360 folks loved the idea and we collaborated for ~6mos and in July, 8 weeks after the 2026 IR1000 dropped in May 2026, we released the Q1 AI1000.

But…my pitch had one important caveat that came along, a ‘must have’ if you will: Because the underlying AI models and Agentic Commerce is evolving so quickly, we had to update the AI1000 quarterly. In 2026, annual updates feels like the difference between X and Y (5yrs). This was a big investment on our side (people, tokens, code, infrastructure) , we had to not only build the infrastructure for the Q1 AI1000, but going forward, our aspiration was that days after the end of a Q, we’d be in a position to drop the next quarterly update.

The Results?

The response to the Q1 AI1000 exceeded our wildest expectations. We’ve had a ton of positive feedback, and a lot of retailers asking all the important questions. Most importantly we’ve been flooded with awesome feature requests:

• Can I update this in more real-time – we just implemented UCP and want it to show up!

• When Q2 comes out, can I be alerted? Can I login and look at the six-factors and trends at that level for our company?

• We have four sub-brands, can we track those separately?

• Can I create a list of companies I want to track on a watch list?

Unfortunately, Q2 took a bit longer than we wanted. There’s two main reasons.

1. We had a large influx of customers at the end of Q2 (which we’re hugely thankful for 🙏) and had to prioritize customer on-boarding vs. AI1000 work.

2. Being Agentic, once we got onto the Q2 AI1000, the team wanted to make it worth the wait and pushed themselves to add all (and more) of the features above (minus the ‘watch list’ we’re looking at that for Q3).

I’m particularly excited by the self-service portal so let’s start there.

Introducing the AI1000 Self-Service Retailer Portal

The way this works is you have to be a team member of the company. For example, let’s say you are in marketing at Chewy. You go to refibuy.ai/ai1000/chewy (where chewy can be any of the AI1000 company names) and you enter your email. We enter a magic login link to that email to verify you are with the company. Click the link and you are off to the races. To give you an idea of what it looks like here’s the top part of Chewy’s data:

Note formatting modified for this newsletter format, some details redacted

Not only do you get complete insight into the four factors that combine to form the score that determines your AI1000 rank, but with the Q1->Q2 slider (which in Q3 will have a third datapoint and so-on), you can track all your detailed progress Q/Q and ultimately Y/Y, etc.

Also - you’ll see the 90-day momentum, that’s coming in Q3 when we have a large enough corpus of trend data.

You asked, we delivered! I hope everyone enjoys seeing their rankings, if you aren’t on the AI1000, let us know, we can run a similar report for you, free of charge.

Now for some of my personal highlights from the AI1000 Q2 data drop, many more can be found that the team put together at our comprehensive landing page here and today’s press release here.

Four AI1000 Q2 Highlights

The first highlight I wanted to share illustrates why quarterly updates are critical in the AI Era: 972 of the 1,000 retailers changed positions from Q1-Q2!

This graphic illustrates the ‘rank move cohorts’ from downward moves to upward:

In the 119 ‘top climbers’ club, here’s the 8 companies that moved the most:

Congrats to these AI1000 top movers, your hard work is paying off!

Category Trends

Peeling the onion on all this individual retailer movement, here’s an illustration of how each category moved:

Congrats to all our friends in the automotive category -you guys are really pushing hard and it’s really paying off.

That’s the Q1->Q2 story, here’s where the categories rank stack as of Q2:

Answer Engine Moves

You’ve probably all seen the Similarweb chart that shows the dramatic loss of share by ChatGPT of referral traffic (but overall traffic for all players is still up substantially). In that data you see Gemini surging, but also many have noticed that Perplexity is fading and Claude is on the rise – yes Claude, with it’s enterprise audience, consumers are flocking to Claude:

Claude hasn’t surpassed Perplexity in our dataset, but one more Q like this and that will happen.

UCP Adoption

One of the big factors we measure is what retailers have a UCP endpoint. Sadly, only 225 of the AI1000 (22.5%) are there. The bulk of these picked up the UCP endpoint ‘for free’.

Want to move your score for Q3 – Implementing UCP is your easiest and fastest unlock.

The Clock Is Ticking Toward Q4

We’re down to 14 weeks until Thanksgiving, there are four things you can do in those 3 months to be ready for what we believe is going to be a huge Agentic Holiday. We’re not alone, Salesforce just dropped their Holiday Predictions (pod coming soon!) and their second prediction is that AI agents will influence 22% of sales this holiday or $73B in Agentic Commerce fueled opportunity for you to grab.

1. Always Updated Product Data - Enrich your product data for agents and keep it continuously ready for agents to answer questions agents will be asking.

2. UCP - Implementing agentic commerce protocols give you an unfair advantage

3. Optimize for a multi answer-engine market – The big two are ChatGPT and Gemini, and you can get them with datafeeds and ACP/UCP endpoints. From there, the other engines are crawler based, so take that awesome product data you’re exposing and put it in your PDP-level meta data.

4. Retail is a Zero Sum Game - If your score (available in the portal) is below your category average, you’re not lagging your competitive set, time to really focus on getting caught up, because at this pace of change, if you fall too far behind, it will get exponentially harder to catchup.

Summary

The full Q2 findings, category analysis, and the new Retailer Portal are live at: https://www.refibuy.ai/ai1000 .

That’s just the tip of the iceberg, keep an eye on our ReFiBuy social channels, we’ll be sharing even more insights and analysis over the coming weeks. And of course, in less than 90 days we’ll be back with the Q3 data, the last AI1000 ‘print’ before Holiday 26. Then in early 2027, we’ll close out the year with a recap of how everyone did with Agentic Holiday 26 with the Q4 data.

I want to wrap by thanking Digital Commerce 360 to listening to my wacky idea (and then saying: “Let’s Go!!”), the AI1000 retailers who gave us so much feedback from the Q1 data, our media partners and friends who help spread the word, our amazing customers for their support and the team at ReFiBuy for working tirelessly to get this out and deliver an amazing set of new features on top of the data – enjoy everyone!

Subscribe now

Subscribe now

Ben Rodier is the CEO and co-founder of FrontlineIQ, an AI sales coach built for in-person sales teams. He's a three-time founder with nearly two decades in retail, and his path is a straight line to this moment.

The store world is buzzing about how FrontlineIQ is blowing up and accelerating store sales for its customers, and we wanted to hear the pitch straight from Ben.

Listen/Watch Our Interview:

This week Retailgentic goes full Tron… in reverse. Instead of getting sucked into the digital world, Scot steps out of it and into the analog one, bringing the AI angle with him.

Ben Rodier is the first AI guest on the show who isn’t from e-commerce. He’s building physical AI for the sales floor: an AI sales coach for the people who sell to you face-to-face in furniture stores, apparel shops, bank branches, and beyond. It’s part founder-journey story (three companies, two exits, twenty years in retail) and part product deep dive into how you actually coach a human salesperson with AI without being creepy about it.

Highlights

• 𝗧𝗵𝗲 𝗥𝘂𝗹𝗲 𝗼𝗳 𝟯𝟬: There’s an inflection point (around 30 AI coaching sessions) where practice turns measurable. Reps who cross it sell roughly $2,000 more per week, about $100,000 more per year, than those who don’t.

• 𝗧𝗵𝗲 𝗰𝗿𝗼𝗰𝗼𝗱𝗶𝗹𝗲 𝗰𝗵𝗮𝗿𝘁: Plot the reps who practiced 30+ times against those who practiced 10, and the two lines pull apart into a “crocodile-shaped separation”… the clearest picture of coaching ROI Ben’s ever had.

• 𝗧𝗿𝘂𝘀𝘁-𝗳𝗶𝗿𝘀𝘁 𝘀𝗲𝗾𝘂𝗲𝗻𝗰𝗲: Data (find the goal) → AI role play (practice it) → privacy-compliant listening (coach the real conversation). Each step earns the right to the next.

• 𝗥𝗼𝗹𝗲 𝗽𝗹𝗮𝘆 𝗶𝗻 𝘃𝗼𝗶𝗰𝗲 𝗺𝗼𝗱𝗲: The coach is a ChatGPT-style voice agent trained on your company’s own playbook. It throws curveballs, like someone walking into the room mid-pitch. “No one likes to do role play with their manager. It’s super awkward.”

• 𝗔 𝗽𝗿𝗼𝗳𝗶𝘁 𝗰𝗲𝗻𝘁𝗲𝗿, 𝗻𝗼𝘁 𝗮 𝗰𝗼𝘀𝘁 𝗰𝗲𝗻𝘁𝗲𝗿: Unlike most training software, if you practice you sell more so the tool pays for itself.

• 𝗕𝗼𝗿𝗻 𝗼𝗻 𝘁𝗵𝗲 𝗳𝗹𝗼𝗼𝗿: Ben invented retail-media ads on Walmart’s website in 2009 before the term existed. Google execs opened a meeting with: “Who are you, and how did you figure out how to get these properties as advertising publishers?”

• 𝗧𝗵𝗲 𝗺𝗲𝘀𝘀𝗮𝗴𝗲 𝘃𝗲𝗿𝗱𝗶𝗰𝘁: McDonald’s tried a humanless AI drive-through and it “failed miserably”. They pivoted right back to a human. Ben’s bet: five years out, most buying and selling is still face-to-face.

It’s not often we leave the digital surface behind on this show, but Ben’s take on physical AI, and what happens when the sales floor and the digital world finally connect, is one of the more grounded, human conversations we’ve had. Enjoy the conversation.

🎧 Listen wherever you get your podcasts: Spotify, Apple, etc.

Or, we have a video version on YouTube (make sure to subscribe!) →

Want to Learn More?

Stay on top of what’s happening in agentic commerce, from the tools powering it to the founders building with it by following us across platforms:

• Podcast: https://retailgentic.transistor.fm

• YouTube: https://www.youtube.com/@Retailgentic

• Instagram: https://www.instagram.com/retailgentic/

• X: https://x.com/retailgentic

• TikTok: https://www.tiktok.com/@retailgentic

We're here to make sense of what’s next in retail + AI and bring you along for the ride. Subscribe and stay tuned!

This is a two part series where we’re going to go deep into what Shopify is doing in Agentic Commerce.

• 8/12: Part 1 - In Part 1, we’re going to start by looking at Shopify’s Q2 results through the lens of Agentic Commerce - it’s a good way to get a handle on what’s going on generally and specifically with Agentic Commerce.

• 8/25: Part 2 - In Part 2, we’re going to dive deep into Shopify’s Agentic Catalog offering. We’ve talked to hundreds of larger Shopify store operators and there is a LOT of confusion and misunderstanding around what Shopify Catalog.

And without further delay, here’s Part 1:

Shopify’s Q2 Agentic Commerce Highlights

Shopify is really leaning in on Agentic Commerce and mentioned these highlights throughout the release, call and Harley’s comments with press later:

• AI-driven traffic to Shopify stores grew 3x y/y.

• Orders originating from AI searches also grew 3x y/y.

• New-buyer orders from AI channels came in at nearly 2x the rate of other channels.

• Half of AI-referred sessions landed directly on a product detail page, 2.5x the rate of traditional search.

• 75% of AI-attributed orders came from outside Shopify’s top 100 categories.

That last bullet is interesting, Harley expanded:

I’ve talked a lot in these pages (like here) about how Agentic Commerce has now made shopper intent so large that it is swamping product-level information. That’s advantaging Shopify’s merchants right now. They are both more long-tail, but also at the head and torso of the product curve they are more nimble than bigger companies and able to adjust their product catalogs much faster.

Analysis: Traditional ad-based search reinforces winners and given the focus on 2-3 keywords it puts smaller merchants at a disadvantage (e.g. competing for ‘running shoes’) but Answer engines have more prompts to work with (24+ * 4-6 turns of the convo) AND shopper intent which flips the discovery advantage to Shopify’s eclectic long-tail catalog.

The volume is still small relative to Shopify’s $115.6B of GMV - the company said that repeatedly - but this is the clearest set of company disclosures yet showing the Agentic Commerce flywheel starting to rotate.

Financial Highlights from the Q

Wall St. loved the Q - it was a classic beat and raise (beat current quarter’s expectations, raised next Q’s. Here’s the financial highlights from the Q

Note: Going forward in here for simplicity, I’m going to report these in non-constant currency because their currency swings are less than 10%

Deeper Dive into GMV and Revenue Growth:

One theory about public Agentic Commerce amongst SaaS companies is that you should be seeing a material acceleration from AI. The logic goes:

• if you’re putting AI into your products and customer usage/engagement (merchant and consumer side in the case of Shopify) is increasing.

• Revenue should be going up to reflect that investment.

Let’s look at that.

Here we first see Shopify’s y/y Revenue growth over the last 8 quarters (2yrs):

You can see (red line - right axis) that y/y growth for the last year has been in the 30-35% with Q3_24 and Q1_25 the only two quarters in the lower 25-30% zone.

Let’s look at GMV as well:

Here we see a similar story - with even a bit of a more pronounced slow-down (Q2 26 GMV grew 31.6% vs. Q1’s 34.8%.

Conclusion: There’s no real acceleration of Revenue or GMV due to Agentic Commerce. To their own points, it’s early, and I predict this Holiday (Q4 26) we’ll see the first material acceleration due to Agentic Commerce, so we have to wait until Late Jan/early Feb to check back in on this one.

To understand what’s really driving Shopify’s ‘beat and raise’, we need to dig into what makes Shopify tick outside of Agentic Commerce.

Shopify’s ‘Real’ Economic Engine: Checkout/Payments (Merchant Services)

A lot of people in my founder circles mistakenly think of Shopify as a software company. Sure they sell software, but that’s one of their two revenue streams (subscription solutions), the lion’s share is “Merchant Solutions” as seen by this chart from their investor presentation:

What’s in Merchant Services, according to wall St analysts the biggest pieces are:

• Shopify Payments and currency conversion : $2.2b or 82%of Merchant Services

• Partners (app store fees, etc.) , Shipping (rev share from usps/ups/fedex/dhl/etc.) , Hardware (POS terminals), Ads (not clear if this is a skim or mark-up) and ‘Other Financial Products’

So the “Lion’s share of the lion’s share” is ‘Shopify Payments’.

It’s growing insanely fast because if you look at the GMV number - the transactional volume flowing through Shopify, a larger and larger % of that (60%→ 68%) is flowing through while growing. Take something compounding ~30% and multiple 8% on top of that - you get two compounding numbers multiplied together which is always a winning formula:

The result: a doubling in Shopify Payments processed from Q1_24 → Q2_26 and the driver of Shopify’s 2026 performance.

Conclusion

This quick ‘highlights’ tour of Shopify’s Q2 results as a springboard into what makes Shopify’s economic engine hum tees us up for Part 2: Deep Dive into Shopify’s Agentic Catalog, coming soon August 25th!

Three newsworthy scoops as we exit Summer and head into peak back-to-school and then a fast down-hill ride to Holiday 26. (15 weeks until Thanksgiving).

FLASH 1: ChatGPT Product Card Spike

Over the weekend our ACO monitoring system started going bonkers on ChatGPT data:

• Product Card visibility up 20% across many clients (% of prompts that elicit a product card)

• Card count per prompt that elicits a product card spiked from a range of 1-2 to 6-8

This has persisted through today.

Subscribe now

Examples of New Experience

Here we’re looking for ‘face serum’:

Here I am a logged in user and ChatGPT knows I am in Raleigh, NC. Notice that the top of the map is dominated by a very detailed store finder. Notice the local map now is front and center, for this shopper experience. As I flip through different product cards, the beauty retailers that have the product turn on and off in the map, so it’s more of an interactive view than you would get from the google ‘show me in stock near me’ feature of Google Shopping.

There are also a ton of product cards under neath here - not in a carousel, each with a lot more exposition about the product.

When I click on a product card the offer card has changed:

The biggest change is the hero image has been enlarged compared to what we’ve seen previously and I can flip between images:

Finally, if I drill down on a store it pulls up a ‘store card’→

Here you get store-level reviews aggregated from Yelp! and what looks a lot like a “Google places” type treatment.

We’re seeing this consistently across free and logged in accounts - what are you seeing?

Leave a comment

FLASH 3: Three New Shopping Related Feature Flags Detected

We have another set of agents that monitor the chatgpt front-end code for any interesting visible changes. As mentioned previously, ChatGPT has a framework:

1. Push new UI front-end with feature flags turned off - presumably for internal testing or tight A/B controlled audience testing (or both)

2. Over time, on the back-end they can turn these feature flags on and off when they want to roll out new features with a big bang - meaning you don’t have to wait for the new code to roll out via downloads for the app-based users (web can typically roll-out instantly).

Editorial note: Because we can’t turn these on and we’re dealing with unannounced features, this is pure speculation - educated speculation, but still 100% speculative, whereas the product-card visibility FLASH changes are live, observable and measurable.

New Feature Flag 1: shopping_sidebar_enabled

ChatGPT is using a company called statsig which is a platform for running A/B tests and gating off these types of new features. Speculating here, but we can see two statsig experiment IDs associated with this one. This pattern usually means - this feature is coming, let’s A/B it as we slowly roll it out and see which one has the highest KPIs (traffic/clicks/sales/cart adds/likes - could be anything).

New Feature Flag 2: ‘AllProducts’ Setting

The other setting is in the code in the settings/admin area. The sky’s the limit on this one. It could be a ‘god mode’ feature where super ChatGPT admins can see all the products in the system, it could be a user setting where you can see a wishlist type functionality, it could be a ‘product history’ view. This one’s harder to predict.

New Feature Flag 3: 'navigate_checkout’ and ‘prepare_checkout’

These are part of the checkout sequence. The current checkout situation at ChatGPT is basically what we call cart injection (where available down-stream). Most likely this does not indicate that they are revisiting a full-blown checkout, but is some process improvements to the injection. That being said, we haven’t seen this much action since March 26 when checkout was abandoned.

Subscribe now

META’s Crawler goes Crazy - What are they building?

Popular ‘build in the open’ account on X, levelsio on August 6 posted that he was seeing across his various sites. He then followed up with this post:

Levelsio claims that META told him they are building their own search index. EU GEO company Peec chimed in:

Cloudflare Radar tracks bot traffic across their massive network and they conveniently break it down by 'category of target’ and they have “Shopping & General Merchandise” in the 7 day window, it’s clear that META is crawling the retail category especially heavily.

Zooming out and looking at 4 weeks, you can see another interesting trend:

Look at that Claudebot activity. Why do you think Claude AND META are both showing a dramatic increase in traffic right now?

I have some guesses on what’s going on, I suspect in the next 2-3 months there’s going to be some pretty exciting announcements along the lines of my original 2026 predictions.

On the META side, digiday reports that prior to 1H 2026, the categories were balanced, but now S+GM is significantly the majority of the crawls, showing this was not ‘business as usual’.

Summary

To Summarize it’s mid August and:

• ChatGPT is adding local inventory to their product and offer cards and showing way more cards than previously

• They are in test with a shopping_sidebar feature on the consumer side.

• There’s a merchant or internal feature called “AllProducts”

• Two new checkout variables or tests were introduced

• META and Anthropic are crawling shopping sites at a frenetic pace

Stay tuned, we’re tracking this closely…

Subscribe now

Subscribe now

Editorial note: As a reminder, we prefer news items that surface new data and insights. We look for both observed data (usually in the form of permissioned panel data) and stated preference data (usually in the form of survey data). We believe no one set of data is perfect, but by looking at a landscape of datapoints, together we can spot important directional trends and shifts in consumer behavior before they happen. We’ll always stated which of the above the article incorporates at the top of the coverage.

Mastercard spent its entire earnings call on agentic commerce

Mastercard Q2 2026 earnings call · July 30

I listen to a lot of earnings calls and have to admit that prior to this quarter, I had never listened to a Mastercard earnings call. The day after the call, my Agentic Commerce alert system went through the roof about this call so I went back and listened and was blown away about not only the density (about 80% of the call by my guess) but the enthusiasm from the tone and body language of the call around Agentic Commerce. I’ve done these calls before, these are the most forensicly reviewied and lawyered content you’ll ever hear. I could fill 5 pages with the quotes from this one, here’s the top of the top: (These are all Mastercard’s CEO Michael Miebach - bolding is mine).

1. "Agentic Commerce is the next evolution in payments where the importance of security, transparency and control only increase. Agentic Commerce creates a significant opportunity for Mastercard. It leads to incremental transactions and even more opportunity for our services."

2. "It's unique because we have additional capabilities that we put into the agent pay protocol from us, and one of them is verifiable intent, which allows you to basically challenge a transaction, say, I never wanted to buy this and then the chargeback process can kick back in. This was innovated together with Google."

3. "The rise of Agentic Commerce also brings about an entirely new class of payment use cases, machine-to-machine payments. This is an expansion of our addressable market and one that we are at the forefront."

Mastercard has built the chargeback path for agent purchases. You don’t build the exception path for something you think is three years out.

Their strategic framing is a virtuous cycle: more switch transactions, more data, more value-added services to attach - all extended to agent-initiated payments. And they’re explicit that they think cards prevail in an agentic world (I tend to agree, consumer behavior is increasingly stubborn to change.

→ Mastercard Q2 2026 earnings call transcript

→ Agent Pay for Machines launch release

Keep an eye on Mastercard, they are up to big things in this space and they went from more of a defensive stance with Wall St. (“We’re going to be ok guys!”) to pure offense: (“This blows up our TAM, we’re in front, we’re adding new capabilities, we’ve partnered with Google, LETS GO GUYS!”) Note: These are my interpretations not their quotes.

Visa Declares Agentic Commerce a Growth Driver

h/t to Marcel Van Oost for putting out these sweet earnings charts.

Visa also decided Q2 2026 was the time to take a similar stance.

Two networks, one quarter, on the record. Whatever you thought about the timeline before, the rails have stopped debating it.

The contrast that makes it land came three days earlier → PayPal’s new CEO published five strategic priorities and agentic commerce wasn’t one of them.

We’re seeing a big divergence from the card companies and PayPal, you can easily guess which path I think is the right one.

→ Visa declares agentic commerce a growth driver

Similarweb Clarifies What Happens After AI Recommends You…

Similarweb — “The Downstream Impact of AI Visibility” · release July 12 with accompanying webinar July 16

This is observed data from Similarweb’s massive panel.

The key findings:

• AI-recommended brands got 2.5x more visits within seven days.

• AI-influenced visitors averaged 12.0 pageviews and 11.8 minutes on site, against 6.5 pages and 5.6 minutes for everyone else.

• 55.9% of that downstream traffic arrived as branded search. Only 8.8% arrived as a direct AI referral link.

That last one is the story. Roughly nine out of ten visits an AI recommendation generates land in your analytics wearing someone else’s clothes - branded search gets credited to SEO, direct gets credited to nothing. Anddd.. we’re back to the age-old attribution problem.

This all but verifies the Dark Agentic Commerce Traffic we’ve been talking about. Similarweb frames it as users coming back later. I’m sure in the panel data it looks this way, but it is not.

The companion webinar re-iterated another number we’ve seen over time: zero-click is now 65–68% of Google searches and it’s heading to 100% rapidly.

→ Similarweb — The Downstream Impact of AI Visibility

McKinsey and ICSC: US Agentic Retail Forecast: $1 Trillion

McKinsey & ICSC - “Shopping in the Age of AI: Redefining Stores for a New Era” · webinar July 13 and Shopping in the age of AI paper

McKinsey uploaded a YouTube and it’s a good watch. this data is based on stated preference from 3,004 US consumers plus C-suite interviews and their global expert network, across grocery, fashion, health and beauty, home, electronics, and food and beverage.

The headline number is up to $1 trillion in US B2C retail revenue through agentic commerce by 2030. But the stat that actually changes how you think about urgency is this one → the top decile of retailers is projected to capture more than 85% of sector economic profit.

McKinsey believes: “agentic commerce concentrates.”

The Mckinsey store paper adds:

Their convenience-versus-discovery framework is a genuinely good model - physical retail bifurcating rather than declining, with a single store unable to optimize for both missions at once. And they name the other macro forces stacked behind it: ~$100T of intergenerational wealth moving to Gen Z and millennials, and ~40% of US workers still hybrid or remote, which re-anchors shopping patterns near home instead of the office.

This bit hits home: “agents compare your store to competitors before the consumer ever leaves home, and data accuracy is table stakes”. Not only does your online catalog need to be upgraded, but you need to make sure agents looking at local inventory have access to as real-time store-level inventory as possible.

→ McKinsey — Shopping in the Age of AI

LDWW Survey: 70% of Americans are already shopping with AI

LDWW survey of 2,000 US adults - stated preference

Here’s what this survey reveals:

• 70% already use AI to help them shop, and plan to use it more

• 71% expect to increase their use in the years ahead

• 67% say AI influenced a recent purchase decision

• 30% have already bought products or services through AI, not just researched

• 19% say AI tools are the most trustworthy source they use to shop

Nearly one in five Americans trusts a language model more than any other shopping input they have.

→ LDWW — 7 in 10 Americans Now Use AI to Help Them Shop

Similarweb’s 2026 Gen AI Landscape

Similarweb — “Gen AI Landscape” (observed panel data)

Similarweb used to share the highlights of their panel data around GenAI monthly then stopped about 6 months ago. This is the first GenAI mega update since, and the best single map of where AI search traffic is actually going.

The two charts are the tip of the iceberg, here’s the Alexa shopping data for example:

→ Similarweb — 2026 Generative AI Landscape

Gener8’s AI Landscape

Speaking of panel data, Gener8 sits on a big panel dataset and the July analysis has some genuinely interesting trends from the data:

(Note: I’m not sure what they mean by AI Mode being the default, that is not the case at least in the US), but they are making it increasingly easy to hit the ai mode button.

Gener8 Ai Landscape July 2026

7.92MB ∙ PDF file

Download

Download

That’s the End to July 2026

As we head to the end of Summer and beyond not only is the information’s going to be coming out fast and furious, but the big answers engines have been quietly building and I predict we’re going to have a bunch of news after Labor Day so buckle up, it’s going to get busy.

Subscribe now

Subscribe now

Colin Sebastian is Baird’s Managing Director of Internet Research, one of the leading sell-side analysts covering e-commerce, social media, and online advertising (and video games!). His coverage universe runs from the mega caps (Amazon, Alphabet, Meta) through mid caps (Airbnb, Shopify, eBay, Instacart) down to smaller names like Pattern.

He started in investment banking, “saw the light” and moved to equity research, and has now covered internet at Baird for over 15 years.

Listen/Watch Our Interview:

This week Scot goes on a markets deep dive with one of Wall Street's leading internet analysts. It’s a tour through the mega-cap playbook (Meta, Amazon, Google, eBay, Shopify) and how professional investors are actually pricing the agentic-commerce shift. Colin covers over $12 trillion of market cap across five names alone, so this is the view from the models, the channel checks, and the trading desk.

This is a wide-ranging, numbers-forward conversation about where the money moves as agents become the buyers and which companies Wall Street thinks come out ahead.

Highlights

• 𝗧𝗵𝗲 𝗦𝗣𝗘𝗘𝗗 𝗳𝗿𝗮𝗺𝗲𝘄𝗼𝗿𝗸: Baird’s formula for internet winners: Scale, Product, Engineering, Efficiency, and Data. Nail all five, and the competitive advantage compounds.

• “𝗟𝗲𝘁 𝘁𝗵𝗲 𝗮𝗴𝗲𝗻𝘁 𝗱𝗼 𝘁𝗵𝗲 𝘄𝗮𝗹𝗸𝗶𝗻𝗴”: Colin explains why skeptics are dramatically underestimating the size of the opportunity.

• 𝗠𝗲𝘁𝗮 𝗶𝘀 𝘁𝗵𝗶𝘀 𝘆𝗲𝗮𝗿’𝘀 𝘁𝗼𝗽 𝗽𝗶𝗰𝗸: Colin’s highest-conviction AI bet is Meta driven by WhatsApp business agents, Meta AI, and unmatched consumer reach. (Last year’s pick: Google.)

• 𝗠𝗼𝗱𝗲𝗹𝘀 𝘄𝗶𝗹𝗹 𝗯𝗲𝗰𝗼𝗺𝗲 𝗰𝗼𝗺𝗺𝗼𝗱𝗶𝘁𝘆: The real value won’t live in foundation models, it accrues to the platforms, applications, and distribution layers built on top.

• 𝗔𝗺𝗮𝘇𝗼𝗻’𝘀 𝗼𝗻𝗲-𝘄𝗮𝘆 𝗱𝗼𝗼𝗿: By opening its catalog to AI agents, Amazon may have crossed a line it can’t reverse and that could strengthen its ecosystem even further.

• “𝗖𝗹𝗼𝘀𝗶𝗻𝗴 𝘆𝗼𝘂𝗿 𝗲𝘆𝗲𝘀 𝘁𝗼 𝘁𝗵𝗲 𝗯𝗶𝗹𝗹𝗯𝗼𝗮𝗿𝗱𝘀”: If AI agents skip search results and category pages, they also skip the ads, putting pressure on retail media’s biggest revenue stream.

• 𝗪𝗵𝗼 𝗼𝘄𝗻𝘀 𝘁𝗵𝗲 𝗮𝘂𝘁𝗵𝗲𝗻𝘁𝗶𝗰𝗮𝘁𝗲𝗱 𝗰𝗼𝗻𝘀𝘂𝗺𝗲𝗿 𝘄𝗶𝗻𝘀: The long-term winners will control identity, payments, checkout, and fulfillment, not just discovery.

• 𝗪𝗵𝘆 𝗪𝗮𝗹𝗹 𝗦𝘁𝗿𝗲𝗲𝘁 𝗶𝘀 𝗻𝗲𝗿𝘃𝗼𝘂𝘀 𝗮𝗯𝗼𝘂𝘁 𝗦𝗵𝗼𝗽𝗶𝗳𝘆: AI could bypass storefronts and even Shop Pay. That uncertainty alone is enough for investors to discount the stock.

• “𝗪𝗲’𝗹𝗹 𝘀𝗲𝗲 𝗶𝘁 𝘁𝗵𝗶𝘀 𝗵𝗼𝗹𝗶𝗱𝗮𝘆”: Colin expects agentic commerce adoption to meaningfully accelerate after Q4, even a one- or two-point lift on today’s numbers would be significant.

Colin is one of the sharpest lenses out there on how the market is actually pricing all of this. Enjoy the conversation!

🎧 Listen wherever you get your podcasts: Spotify, Apple, etc.

Or, we have a video version on YouTube (make sure to subscribe!) →

Want to Learn More?

Stay on top of what’s happening in agentic commerce, from the tools powering it to the founders building with it by following us across platforms:

• Podcast: https://retailgentic.transistor.fm

• YouTube: https://www.youtube.com/@Retailgentic

• Instagram: https://www.instagram.com/retailgentic/

• X: https://x.com/retailgentic

• TikTok: https://www.tiktok.com/@retailgentic

We're here to make sense of what’s next in retail + AI and bring you along for the ride. Subscribe and stay tuned!

Most of what we cover in these digital pages, is the Micro picture - meaning very specific things in the our World of Agentic Commerce which is a huge opportunity in it’s own right. That being said, in this post, we’re going to cover both the Micro and the Macro for the first time really, so I wanted to first explain the ‘why’.

Everything we talk about here at Retailgentic around Agentic Commerce sits on top of an even bigger wave of innovation and capital, that I’ll call the AI Buildout which is supported by the Wall St. AI Trade. That’s the Macro. Let’s start at the Micro because that’s what you come here for and if you’re interested in the Macro, keep reading 😄

In this post we’re going to focus on two specific parts of Amazon’s Q2, if you want to do a complete deep dive, check out:

Subscribe now

Jason and Scot Amazon Q2 Deep Dive→

Watch here→

Listen here→

(back to the post)

The Micro: Andy Jassy’s Comments on Rufus Alexa Shopping

On the Rufus side, there were four datapoints that were interesting. In typical Amazon style, they are ‘just enough’ to get you excited, but leave you with more questions than they answer.

The big four Alexa quotes were:

• More than 350 million customers used Alexa Shopping during the past 12 months.

• In Q2, active users nearly doubled, while interactions increased more than 5× year over year.

• U.S. customers who use Alexa for Shopping spend more than 40% more per order.

• Shoppers clicking a sponsored prompt converted to a sale 48% more often and spent 21% more on average than shoppers who did not.

This is amazing and it’s clear Alexa Shopping is doing really well. I’ve learned over 30yrs to listen to Amazon’s words, but to put 80% of the focus on their actions. The fact that in the last year we have prompt pills all over the discovery and PDP experiences and most recently we have

We’re going to be on-site at Amazon in mid August and here’s the questions we’re going to be looking to figure out answers for:

• 350m customers - Is this global/US? What % of customers does this represent?

• Q2 2x active users and 5x y/y interactions - Was this 2x from Q1 or from year ago? Interactions - is that prompts, does it include clicks, etc?

• Alexa Shopping AOV up 40% - Is there selection bias in here (you take amazon’s best clients, put them in a cohort, that’s an aov that’s 40% higher) - or there’s a more scientific ‘hold back group’ A/B test going on here?

• Sponsored prompts - wow!, but again, selection bias?

As I said, I can tell there’s really something here, because of the actions, but I’d love to have the dots connected a lot more here.

Stay tuned, we’re on it. Now let’s zip back up to 30,000 ft for the Macro…

The Macro: Jassy Single-Handedly Shows Wall St the Clear Path to Huge ROIC on AI Spend

Jassy’s Legacy at Amazon

Foundational to this part of the story, is understanding just how much credibility and Jassy has with Wall St. Jassy isn’t an Amazon newbie, he started in 1997. Amazon urban legend has it that in 2003 he came to Bezos (or they co-developed the idea) for what would become AWS. This is now a huge driver of Amazon’s business. In 2016 he was officially tagged as CEO of AWS and then in July or 2021, Jassy took over the ‘main’ CEO gig.

In that period of time, Amazon spent more and more on Data centers and servers and Amazon showed them how the company has become world class at ROIC - return on invested capital. If you build a $2B data center, how much revenue does that produce over it’s life? What’s the break-even period? Jassy has 23yrs of proving out, via AWS that he is a great steward of investor’s capital and will invest it with an eye towards getting a payback as fast as possible.

Share Retailgentic: Agentic Commerce meets Retail and Brands

Setup going into the call…

The big AI companies, Meta and Hyperscalers are spending so much on AI datacenter and GPUs that it has created a huge wall of worry on Wall St. that can be summarized as: “What if all this capital being deployed is a bubble? What if it' never generates an ROIC? The market has been searching for a clear sign that there is ROIC. Heading into Q2:

• Google Announced -increased their capex and had a great, Q, but didn’t answer the questions the market has - neutral.

• META Announced - Zuck basically said - I want to spend a lot more and to make it work, maybe we’ll do, I don’t know like enterprise apps and stuff! The market absolutely hated this - and 🤮 all over the place.

• Microsoft Announced - this was better, Azure accelerated and Co-pilot seats increased an astounding 50% Q/Q to 30m. The market liked this, but still, it’s hard to tie this to AI spend - nobody connected the dots.

That’s how the table was set going into Amazon’s call, which I’m calling:

Jassy’s Shifts Into AI MAXXING Mode!

Amazon’s AWS results crushed expectations from a y/y growth, but on top of the revenue expansions/beat, margins EXPANDED. As if that wasn’t enough…

Jason and I have 37 of these Amazon quarterly recaps in the rear view window, we have only missed 2, yes 2 in the last 10 yrs. I listen to every conf call, study the transcripts, read at least 5, sometimes 10 wall st reports, read the annual letters and so on. In the Bezos and Jassy eras, I have never heard a call that had so much unbridled enthusiasm as this call. Written words take away a lot of the audio clues/’body language’, but with that back drop. Here are the biggest quotes from the call:

• Booming - “Q2 was another very strong quarter for Amazon. I’ll start with AWS, which is booming right now, and I’ll share the numbers, what we think is going on and why we’re enthusiastic about the ROIC equation, even with heavy CapEx the next few years.”

• ROIC? - “At this level of spend and higher, we have clear line of sight to strong financial returns.”

• ROIC Dots…Connected! && Breakeven…revealed! - “For servers and networking equipment, on average, it takes a little less than 3 years to break even on that investment.” He added that the equipment has a useful life of at least five to six years, while most AI capacity is being contracted for at least five-year terms—creating significant free cash flow during the two to three years following breakeven.

• AI vs. CPU correlation - “We see the margins and returns in AI tracking what we saw with core at the same point of evolution, actually a little ahead.”

• Icing on the cake - “The resulting revenue, free cash flow and return on invested capital is very compelling.”

This chain of five statements was EXACTLY what Wall St. was looking for PLUS it was delivered by Jassy who has a world of credibility PLUS he delivered it with enthusiasm PLUS he delivered it extremely clearly and coherently.

The Icing on the Icing on the Icing on the cake….

As if that wasn’t enough, Jassy put the icing on the icing and said:

Conclusion

To Summarize: Booming, ROIC, Breakeven, Accelerating revenue, growing margins and AWS could be a $1T business. Phew, that was a lot.

Today the market took a breather and there was other dramatic Google news (a topic for another day), but prior to that, Amazon was up 12% which at their ~$3T market cap is $360B (or more than 5 Target’s in market cap growth to put it in comparison).

On top of that Google, and Microsoft, the other Hyperscalers were up too, because Jassy helped show the ROIC on ALL AI investment, not just Amazon’s. I haven’t done the math, but that could be highest $/hr ever recorded in value created from that conference call.

Subscribe now

Subscribe now

We’re excited to have Shopify on the pod! Mani Fazeli is VP of Product at Shopify, leads agentic commerce, and owns a huge slice of the core commerce loop.

He’s spent a little over six years shaping the core commerce loop with merchandising, checkout, customer accounts, B2B (he brought B2B on Shopify to life), plus work on natively supported stablecoins, and for the last ~15 months has led agentic commerce.

Listen/Watch Our Interview:

This is a platform-and-stats deep dive: Scot gets Mani to pin down what agentic commerce actually is (with a four-stage framework that’s the cleanest map we’ve heard), then pulls out a run of genuinely wild numbers on what Shopify is seeing across millions of merchants, before going deep on the catalog, the protocol soup, and everything that shipped in the Spring ‘26 Edition.

Highlights

• Crawl, walk, run, fly: The framework of the episode. Crawl is AI-driven discovery (the biggest, most impactful stage today); fly is fully discretionary, budget-based buying. Mani’s verdict on where the industry actually is: “somewhere between walk and run today.”

• 8X traffic, 13X orders: Shopify’s Q1 numbers… AI-driven traffic to merchants up 8X year over year, and orders through those channels up 13X.

• AI buyers just hit harder: New buyers who discover a merchant through AI order at roughly twice the rate of organic search, are about 50% more likely to convert, and spend about 14% more per order because they show up deeper in the consideration funnel.

• 70% of AI orders are long tail: About 70–71% of AI-attributed orders last year fell outside the top 100 product categories. Models “don’t need to learn how to reach for the small merchant, they just need to learn the intent of the buyer.”

• 2X on fresh data: The Shopify catalog, over a billion products across millions of merchants, converts about 2X better than stale or scraped data, thanks to real-time pricing and availability straight from the merchant.

• A “7 out of 10 at everything”: On how AI is reshaping product, design, and engineering roles.

• The non-merchant-of-record era: Don’t become the marketplace reselling goods, just be the interface to the merchant. The merchant keeps the customer relationship and runs their own checkout.

• UCP was a “labor of love”: Shopify was at the genesis of the Universal Commerce Protocol.

• Sidekick, Pulse & Autopilot: Sidekick usage is up about 4X year over year; Pulse turns Shopify’s data into proactive recommendations right on your admin homepage; and the guiding principle stays “always put the control in the merchant’s hand.”

• B2B may reach “fly” first: By 2030, Mani splits the world into DTC and wholesale/B2B and bets B2B, where both sides transact professionally and obsess over efficiency, may hit full automation before consumers do.

Mani’s four stages are the cleanest map of agentic commerce we’ve heard, and the numbers behind them are frankly staggering. And as he reminds us, we’re still in the early innings. Enjoy the conversation!

🎧 Listen wherever you get your podcasts: Spotify, Apple, etc.

Or, we have a video version on YouTube (make sure to subscribe!) →

👉 And check out the Shopify newsroom blog here

Want to Learn More?

Stay on top of what’s happening in agentic commerce, from the tools powering it to the founders building with it by following us across platforms:

• Podcast: https://retailgentic.transistor.fm

• YouTube: https://www.youtube.com/@Retailgentic

• Instagram: https://www.instagram.com/retailgentic/

• X: https://x.com/retailgentic

• TikTok: https://www.tiktok.com/@retailgentic

We're here to make sense of what’s next in retail + AI and bring you along for the ride. Subscribe and stay tuned!

Here at Retailgentic, we just completed our 4-part ACO Everywhere series (starts here if you missed it) and fundamental to that series was the concept of context capture. Once we wrapped it up, I hopped on a plane and went out to NRF Nexus to speak and hang out with all my favorite brand and retailer people and talk about all things Agentic at the Think Tanks and in the halls.

After the first several conversations, I realized we have a big problem in the industry that spans across brands, retailers and all the vendors…

Subscribe now

The Problem

Ask ten vendors, ten brands and ten retailers what “product enrichment” means.

You’ll hear:

• Attribute completion

• PIM automation

• Image tagging

• Catalog distribution/endpoint publication/MCP-ification

• AI-generated descriptions

• Taxonomy mapping

• That thing my PIM does?

They’re all called Enrichment, yet they are all solving different problems and actually taking very different actions around your product catalog.

What I discovered is that like “Agentic Commerce” itself, we have overloaded one phrase (enrichment!) to describe five completely different capabilities.

The Enrichment Gap: Introducing the 5 Levels of Product Enrichment

Instead of thinking about enrichment as “better data,” think about it as increasing levels of shopper understanding coming from the marriage of shopper intent and product data.

Level 1 — Structured Basics

This is what most vendors do and where merchants, especially brands, are investing today:

Examples:

• Dimensions

• Weight

• Color

• Material

• Brand

• UPC

• Images

These basic attributes are obviously necessary table stales, but add zero differentiation.

Level 2 — Conversational Attributes

In Level 2 - it’s starting to get interesting because in giving robotic: “material=cotton” …now we’re answering questions.

We’re answering:

• Is this breathable?

• Will this shrink?

• Does it wrinkle?

• Is it comfy?

• Can I wear it in the summer?

• Is it good for travel?

The shift from Level 1 to Level 1 is we’re now enriching conversations, not just product data.

It’s important to note that this is just not me ‘guessing’ or making a prediction - Google is focused on getting this data from merchants for getting brands and retailers to this level and it’s part of UCP.

Level 3 — Shopper Context

Because the Answer Engines have tons of shopper intent, on top of basic product data and conversational attributes, they are looking for product-level shopper context from authoritative sources (like the brand and their retail partners).

Shopper context like:

• Use cases

• Occasions

• Compatibility

• Seasonality

• Product lifecycle

• Pros

• Cons

• Selling points

• Related products

At Level 3, we’re past attributes and into buyer context.

Level 4 — Market Context

Products don’t live in isolated, they exist in markets, markets that change a lot and that rate of change is accelerating.

Market Context includes:

• Promotions

• Popularity

• Inventory

• Trends

• Reviews

• Price changes

AI agents care enormously about these and it’s a top input into who wins the offer card.

Level 5 — Agentic Context (new)

At the final, most advanced level of enrichment, it isn’t about products, it’s about decisions.

Examples:

• Which product best fits this shopper

• Which product best fits this prompt

• Why should the AI recommend Product A over Product B?

• What objections must be overcome?

• What evidence supports those claims?

• Which customer segments respond best?

Now you’re optimizing for reasoning instead of 2002 basic database filtered navigation.

Why This Matters

In these digital pages we’ve covered these first principles of agentic commerce:

• Consumer behavior has changed more in the last 3 years than 20 years of keyword jail, causing us to over-optimize for keywords

• Agentic commerce optimizes to matching the large amount of shopper intent with the product context.

• Multi-surface context capture is your competitive advantage and your moat.

• All of this has to be done in a closed loop, or you aren’t optimizing, you’re guessing.

The New Definition of Enrichment: Agentic Enrichment

Product enrichment is no longer the process of plugging in missing attributes.

In the Agentic Commerce Era, it is now the process of continuously adding decision-ready context that enables AI to confidently recommend, compare, explain, and sell the product.

How to tell if you aren’t enriching for Agentic Commerce…

• Each product doesn’t have a detailed and growing FAQ (substantially longer than what’s on the PDP)

• Your vendor/team doesn’t know that UCP/ACP also have a product catalog spec in addition to the payments piece

• You’re using an off-the-shelf AI model to recommend and fill in data (it will not make good recommendations and it will fill your catalog with reddit-fueled AI slop)

• They aren’t talking to you about how to get access to more context

• You aren’t enriching in your company’s ‘voice of brand’

• You’re sending the same data without channel customization to Google, ChatGPT, CoPilot and Meta.

Conclusion

The companies that win Agentic Commerce won’t necessarily have the biggest catalogs or the happiest PIM that has every mandatory field filled out to perfection. They will have the richest context that is continuously optimizing and responding not only to underlying product changes, but shopper behavior and market dynamics fueled by context capture across as many surfaces as possible.

THAT’s the enrichment gap.

Subscribe now