For months, Andy Jassy has been telling anyone who'll listen that Amazon plans to partner with outside AI companies to plug parts of its shopping experience into their products.

Talk is cheap. Hiring isn't.

Last week Amazon posted a job listing for a Principal Technical Program Manager to run "a dynamic engineering team responsible for Amazon's strategic integration with third-party agentic platforms, building the next generation of on-site and off-site commerce experiences."

The role oversees 40 engineers. That's not a pilot. That's a department.

Industry sleuth Joe Kaziukėnas surfaced the listing, and Jason Del Rey at The Aisle connected it to a string of moves that paint a clearer picture of where Amazon is headed.

What Changed

Until recently, Amazon's AI shopping resume looked like this: launched Rufus, blocked AI crawlers from Amazon.com, and sued Perplexity for scraping.

Defensive. Closed off. Classic Amazon flywheel protection.

Then last week Amazon joined the tech council for the Universal Commerce Protocol, a new standard designed to let AI agents shop across retail and AI apps without custom integrations for every store.

Now the job listing.

Both point in the same direction: Amazon is preparing for a world where shopping starts inside AI products Amazon doesn't own.

Two Roads From Here

The job description hints at two possible plays, and Amazon may run both.

The distribution play. Amazon's catalog, Prime delivery, reviews, and checkout flowing into ChatGPT, Claude, Gemini, and whatever assistants come next. Customers ask an AI to buy something, and Amazon fulfills it behind the scenes. This one feels like a question of when and how, not if.

The merchant play. Amazon recruits Shopify stores, DTC brands, and other retailers into its Buy for Me agent, turning what currently looks like a scrappy experiment into a sanctioned channel for buying from outside Amazon's marketplace.

The listing doesn't say which comes first. The scope of the role suggests both are on the table.

Why Amazon Sellers Should Care

If Amazon succeeds at injecting its commerce layer into outside AI apps, the real estate that matters shifts. Your product page on Amazon still matters. But the surface where the buying decision happens may be a Claude conversation, a ChatGPT recommendation, or an agent acting on a customer's behalf inside an app that has nothing to do with Amazon.

That changes what "ranking" means. It changes what reviews do. It changes whether your A+ content even gets seen before the purchase.

It also raises a question Amazon hasn't answered yet: when an AI agent on someone else's platform buys your product through Amazon, who owns the customer data, who gets credit for the conversion, and who controls the experience around the sale?

Amazon clearly wants to be the one controlling where, how, and with whom its commerce capabilities show up beyond Amazon.com. The 40-engineer team is being built to make sure of it.

Watch this hire. Whoever takes that job is going to shape a meaningful chunk of how product discovery works for the next decade.

Original reporting by Jason Del Rey at The Aisle. Read his full piece here.

Amazon is signaling it could start mixing AI-generated commentary directly into its main search results, blurring the line between its Rufus chatbot and the traditional search bar that drives most of its retail sales.

According to Amanda Doerr, Amazon's VP of core shopping, the company is exploring hybrid modes, such as showing a conversational blurb above search results for certain queries without forcing shoppers into a separate chatbot interface.

Some shoppers have already spotted recent queries returning AI summaries plus product recommendations, with an option to keep chatting. Amazon says it tests with subsets of customers and iterates based on feedback.

The split Amazon is wrestling with comes down to query type. For straightforward purchases, the traditional grid wins. "If you're searching for milk, you want to see the milk you usually buy, how much it is today, and add it to your cart, not a paragraph about the pros and cons of 2% versus whole," Doerr said. Rufus is better suited for open-ended exploration, like researching hiking boots for the summer.

Why Amazon is moving carefully: the search bar powers most of its online retail revenue and underpins its $68.6 billion ad business, which is mostly sponsored search and product ads. Disrupting that flow has real risk. Google, by contrast, has already given AI prime real estate with AI Overviews and AI Mode.

A few other developments worth flagging:

Rufus revenue. Amazon says Rufus generated nearly $12 billion in incremental annualized sales in 2025 since its 2024 launch.

Grocery gets its own playbook. Retail chief Doug Herrington has made perishables and essentials a priority, and Doerr's team is building a grocery experience that looks different from the rest of the site.

Search for milk and Amazon wants to surface cereal and bananas to help build a basket, rather than 25 versions of milk. "Tap-tap-tap, add-to-cart" is the goal.

Autocomplete is doing heavy lifting. Nearly 60% of U.S. searches now originate from a shopper accepting an autocomplete suggestion. Amazon recently added thumbnail images alongside those suggestions.

Visual search is expanding. Lens Live, launched in September, lets shoppers scan their surroundings in the app without snapping a photo, with Rufus integrated for follow-up questions. Amazon wants visual search to become a shopping shortcut.

Outside agents still locked out. Amazon launched Buy for Me last year to let shoppers purchase from other retailers inside its app, though merchants complained about inaccurate product data. To fix the data problem on its own site, Amazon partnered with providers like Feedonomics in March so it can source details without scraping.

Meanwhile, Andy Jassy reiterated in January that outside shopping agents aren't welcome on Amazon, citing both poor customer experience and unresolved economics.

The takeaway for sellers: Amazon is hedging. It won't replace search with chat, but it will keep stitching AI into the existing flow where it earns its place. Listings still need to win the grid for transactional queries, but the conversational layer above search is coming, and it will reward products with the kind of clean, structured data that AI summaries can pull from cleanly.

Five months ago Walmart and OpenAI announced a partnership that was supposed to reshape e-commerce. Customers could shop Walmart inside ChatGPT using Instant Checkout. Doug McMillon and Sam Altman both hyped it.

Now Walmart has killed it.

The reason: Instant Checkout couldn't deliver. Conversion rates fell well below what Walmart sees on its own channels. The feature struggled with product accuracy and couldn't sync with Walmart's internal shopping tools.

The deeper problem is how Instant Checkout was built. It scraped retailer websites for product data. Scraping tells you what exists. It doesn't tell you what's actually in stock, what shipping looks like, or what's accurate right now.

Forrester's Emily Pfeiffer put it bluntly: crawling and scraping isn't enough to run real commerce. Gartner's Bob Hetu said OpenAI underestimated how hard transaction enablement actually is.

So Walmart pivoted. It's now embedding its own assistant, Sparky, directly inside ChatGPT and Gemini. ChatGPT Plus and Gemini Advanced users got access on March 20. Free-tier rollout is coming this spring. Anthropic is reportedly next, with Sparky potentially landing inside Claude.

The early numbers explain the move. Sparky users inside ChatGPT are converting at roughly 70% of the rate of direct Walmart.com shoppers. Sources say that's far above what Instant Checkout was producing.

Walmart's EVP of AI Acceleration Daniel Danker told investors Instant Checkout was "a very temporary moment in time" and that within a month customers wouldn't see it anymore.

Walmart isn't alone. Target, Instacart, Shopify, and Etsy are all building their own embedded experiences inside AI platforms instead of handing the keys to OpenAI.

What this means for sellers:

The agentic commerce land grab just shifted. The early bet was that ChatGPT becomes the storefront and retailers feed it data. The new bet is the opposite. Retailers control the agent. The AI platform becomes a distribution channel.

If you sell on Walmart Marketplace, your product data, inventory accuracy, and listing quality now matter inside Sparky the same way they matter inside Walmart search. Sparky is pulling from Walmart's verified catalog, not scraped snapshots. Clean data wins.

If you sell on Amazon, watch this closely. Rufus is Amazon's version of the same play. Same logic. Same incentive to keep the customer inside the walled garden. Expect Amazon to push harder on its own agent rather than open the front door to ChatGPT.

The bigger signal: scraping-based AI commerce just took a public loss. Verified, first-party data won the first real test. Sellers who treat their listings like a database, not a billboard, are going to keep getting picked.

The team at Incrementum Digital reminds you that Amazon added a side-by-side view to the Subscribe & Save Performance dashboard that exposes manual reorder revenue alongside subscription revenue. For CPG brands, this kills the excuse of not knowing how much money is sitting in unsubscribed loyal customers.

Why it matters: Subscriptions are recurring, predictable, and produce longer customer lifetimes than manual "buy again" reorders. Brands were previously blind to the manual-reorder side, so a high reorder rate with low SnS adoption looked fine. Now it doesn't.

Incrementum's example: a brand hitting 32% Reorder Rate vs. low-single-digit SnS the week of Feb 14. On a $50k/week business, that's ~$16k/week from loyal manual reorderers. Converting even a third to SnS adds roughly $5k/week in recurring revenue with zero new customer acquisition.

Steps to take this week:

The new data doesn't change retention math. It just makes ignoring it harder.

✌🏼 See you again Thursday …

The answer to today’s STUMP BEZOS is
11,000 sell over $1 million per year on Amazon