May 04, 2026

Agents Start Buying, and 99% of Catalogs Aren’t Ready

April 28 - May 4, 2026 - The buyer side became autonomous, and the seller side learned how unprepared it is.

Last week we covered Anthropic’s Project Deal and the UCP Tech Council doubling in size. This week the story moved from governance and experiments to the operational reality of agentic commerce. Amazon launched Scheduled Actions, the first feature that lets a major retail agent place orders on a calendar or a price trigger without the shopper saying anything. Mirakl, the marketplace platform behind Macy’s, Best Buy, and Carrefour, published the first hard data on how unprepared product catalogs actually are: less than 1% of product pages meet the bar for reliable LLM recommendation. Publishers from the New York Times to Bloomberg filed an amicus brief siding with Amazon against Perplexity. And BigCommerce used its annual conference to ship the most aggressive agentic feature set the platform has put out so far.

Here’s what happened.

Amazon’s Rufus crosses the line from assistant to agent

On April 28, Amazon launched Scheduled Actions for Rufus, its conversational shopping assistant. For the first time, Rufus can place orders without a shopper prompt. Customers set up automations such as restocking pet food, paper towels, and detergent on a monthly cadence, getting alerts when a favorite author releases a new book, or surfacing gift ideas ahead of family birthdays. Rufus does the product research and either notifies the customer or adds items directly to the cart. Scheduled Actions integrates with Rufus’s Shop Direct and Buy For Me features, which means Rufus can autonomously place orders with third-party merchants outside Amazon.com when the catalog match or price is better.

Three days later, on May 1, Amazon expanded Rufus’s price history feature to 365 days in the U.S., UK, and India. The 30-day and 90-day views remain available in the U.S., UK, Canada, and India. Amazon disclosed that more than 50 million customers have used the price history feature since its 2024 launch, with the average customer checking three times per month. On Amazon’s Q1 2026 earnings call on April 29, CEO Andy Jassy said monthly active users on Rufus grew over 115% year over year, with engagement up nearly 400%. He said Amazon wants Rufus to become “the best shopping assistant anywhere.”

Why it matters for merchants: Until last week, every major retail AI agent (Walmart’s Sparky, Target’s Bullseye, Sephora’s ChatGPT app) was still gated by a human prompt. Scheduled Actions removes that gate. The agent now decides when to buy, and Buy For Me means the agent can decide where to buy from. For third-party merchants on Amazon, this is the moment Amazon’s algorithm starts evaluating you against external listings on a recurring basis without the customer being involved. For merchants off Amazon, this is the first credible “outbound” agent: Rufus is going to start arriving on your product pages with a 365-day price history of the same item on Amazon and a budget already authorized. If your pricing logic, your product data, and your delivery promise are not legible to a machine reading them in five seconds, you lose that comparison every time.

Mirakl: less than 1% of product pages are LLM-ready

On April 28, Mirakl announced Agentic Activation, the first enterprise infrastructure aimed at making catalogs discoverable and transactable by AI agents at scale. The release was paired with the most concrete merchant-side data we have seen this year. Using its GEO Readiness Analyzer, Mirakl scanned 427 product pages across 35 countries. Less than 1% scored above 80 out of 100, the threshold above which products are reliably surfaced by AI agents. The average score was 48 out of 100, below the 61-point competitiveness threshold. 86% of pages had poorly optimized images. 43% lacked customer reviews, ratings, or Q&A. Only 9% provided machine-readable data structures.

The product itself has two pieces. Agentic Product Enrichment uses commerce-tuned generative models to rewrite product content into a form that LLMs can parse and compare. Mirakl claims a 98% success rate across 47 million transformations. Agentic Channels handles the connector layer to LLM platforms, starting with Microsoft Copilot, including inventory sync, pricing, fulfillment, and the transaction itself. Adam Bohbot of Ana Luisa, an early customer, said the brand had to build a new attribute set “for the first time, shoppable directly through an AI channel like Copilot.”

Why it matters for merchants: Mirakl just put a number on what we have been arguing in this newsletter for six months. Discoverability, not checkout, is the binding constraint of agentic commerce. If your catalog scores 48 out of 100, you do not have a checkout problem. You have a “the agent never recommends you” problem. The 9% machine-readable data figure is the most striking. JSON-LD, structured product attributes, ingredient lists, dimensions, compatibility tags - this is plumbing, not branding. It is also the only thing the agent reads. The merchants winning in 2026 will be the ones whose catalogs can be parsed by a model in under two seconds. The 99% who can’t are not invisible because the technology is hard. They are invisible because no one has built the catalog for the new reader.

Publishers back Amazon against Perplexity’s Comet

On April 29, Digital Content Next, the trade body whose 60+ publisher members include the Associated Press, Bloomberg, BBC Studios, Conde Nast, Dow Jones, Fox News, the Financial Times, The Guardian, Hearst, NBCUniversal, News Corp, the New York Times, NPR, Paramount, Vox Media, Warner Bros. Discovery, and the Washington Post, filed an amicus brief in the Ninth Circuit siding with Amazon in its appeal fight against Perplexity. The brief argues that Comet’s user-agent spoofing and unauthorized access to Amazon’s password-protected systems threaten publisher revenue through three mechanisms: corrupted advertising metrics, eroded subscription models, and uncompensated training data extraction. DCN cites a $63 billion annual loss to invalid traffic and an 86% year-over-year increase in bot traffic in the second half of 2024.

The legal lineup now looks like this.

Side	Filers
Backing Amazon	Digital Content Next (60+ publisher members), SIIA
Backing Perplexity	EFF, ACLU

This is a coalition we have not seen in earlier rounds of the fight (see our coverage of the ACLU and EFF amicus filings). Publishers are adjacent to retail in the agentic commerce question because the same legal mechanism (the Computer Fraud and Abuse Act, applied to user-agent spoofing) governs whether AI agents can access password-protected merchant accounts at all. If the Ninth Circuit affirms the injunction, every retailer with a logged-in experience gains a defensible standard for blocking unauthorized agent access. If it overturns, the floor moves the other way.

Why it matters for merchants: This case is no longer about Amazon and Perplexity. It is about whether retailers get to decide which agents touch their authenticated surfaces. If you are a merchant on a platform that ships agent-blocking controls (Cloudflare Web Bot Auth, Shopify’s signed-agent infrastructure, Akamai’s bot management), the Ninth Circuit’s ruling determines how aggressively you can use them. Watch this case the way you would watch a payments-rule change: it sets the legal floor for everything else.

BigCommerce ships agentic infrastructure at Commerce Live

At Commerce Live 2026, held April 28-30, BigCommerce’s parent company Commerce unveiled an agentic commerce push covering distribution, checkout, and merchant tooling. Merchants now get out-of-the-box distribution to ChatGPT, Gemini, Microsoft Copilot, Perplexity, PayPal, and Stripe. The expanded PayPal partnership ships PayPal Store Sync, which links BigCommerce catalogs, inventory, and order management to AI surfaces. The platform also added BigCommerce Companion, an AI assistant inside the merchant admin, and AI-driven catalog enrichment tied directly to the discovery surfaces.

The pattern is becoming standard. Shopify shipped Agentic Storefronts in March. Adobe Commerce defaulted to MCP at Adobe Summit in April. Shopware shipped a native sales channel in version 6.7.9.0 (covered in our European platforms post). BigCommerce now joins the list. The interesting subplot is the channel mix. BigCommerce explicitly named Perplexity, Microsoft Copilot, PayPal, and Stripe alongside the chat surfaces. That is the first major platform vendor to treat payment-platform AI surfaces (PayPal AI, Stripe agentic) as discovery channels in their own right.

Platform	Native agentic channel shipped	Default protocol
Shopify	Agentic Storefronts	ACP + UCP
Adobe Commerce	MCP integration	MCP
Shopware	Agentic Commerce sales channel	ACP
BigCommerce	BigCommerce Companion + Store Sync	ACP + UCP
WooCommerce	Not yet	n/a
PrestaShop	Not yet	n/a

Why it matters for merchants: If you are on Shopify, Adobe Commerce, Shopware, or BigCommerce, the platform is doing the heavy lifting on agent integration. You inherit it. If you are on WooCommerce or PrestaShop, the work is still on you, either through a plugin (the WooCommerce ecosystem has several emerging) or a custom integration. The gap between “platform-native” and “build it yourself” is the kind of question that becomes a re-platforming decision once a competitor on a more agent-ready stack starts capturing your category.

Asia-Pacific is moving faster than Europe

On April 30, Deloitte Asia Pacific published a report on agentic shopping in APAC with two numbers worth pinning. 76% of consumer-facing businesses in the region expect to adopt agentic AI within two years, up from 29% today. The adoption curve is steeper than anything we have seen reported for Europe. The trust gap is real, though, and a separate Visa survey of 14,764 APAC consumers released earlier this year quantifies it. 74% already use AI tools to discover, track, or learn about products. 32% will not share personal or payment information with AI. 45% want stronger payment security guarantees before they buy through an agent.

The European angle is regulatory timing. The EU AI Act’s high-risk obligations land on August 2, 2026. The DSA’s first scheduled review is May 3, 2026. While APAC merchants are racing to plug into Visa Intelligent Commerce, Mastercard Agent Pay, and the various regional partnerships, European merchants are still waiting for liability clarity on the question of who pays when an AI agent makes an unauthorized purchase. The practical effect: European merchants who delay shipping agentic surfaces because the regulation is unsettled will find themselves three product cycles behind their APAC counterparts by Q3 2026.

Why it matters for merchants: If your catalog ships internationally, expect the Asia-Pacific signal to land first. AI agent traffic from APAC chat surfaces is going to grow faster than from Europe in the next 12 months, and the conversion patterns will look different (lower trust, smaller carts, more comparison). Build for it.

Where the AI commerce stack stands this week

Layer	Update this week
Buyer-side autonomy	Rufus Scheduled Actions: agent buys without a prompt
Catalog readiness	Mirakl: <1% of product pages are LLM-ready, average score 48/100
Legal floor	DCN publisher coalition backs Amazon in Comet appeal
Platform infrastructure	BigCommerce ships agentic distribution + PayPal Store Sync
Regional momentum	Deloitte APAC: 76% of businesses adopting agentic AI within 2 years

What merchants should do this week

1. Score your own catalog before an agent does. Mirakl’s 48/100 average is a warning, not a benchmark. Run a representative sample of your product pages through any LLM-readability check (structured data, image optimization, review density, attribute coverage). If you score below 60, treat it as a P0. The catalog rewrite is the discoverability moat in 2026.

2. Treat Rufus as an outbound competitor, not just an Amazon feature. With Buy For Me and Scheduled Actions, Rufus is now placing orders with third-party merchants outside Amazon. If your products show up on Amazon at all, audit the price gap and product data parity between your DTC site and your Amazon listing. The agent will buy from whichever surface offers the cleaner deal, and the customer will not see the comparison.

3. Watch the Ninth Circuit ruling, and document your agent access policy now. Whether or not Comet wins, you want a defensible record of which agents you allow on authenticated surfaces, on what terms. If you are on Shopify, Cloudflare, or Akamai, your platform already has a story here. Get it in writing before a customer dispute or a CFAA claim makes you write it under pressure.

4. If you sell into APAC, start tracking AI-channel referral traffic separately. The Deloitte numbers say agentic adoption in the region is accelerating. The sooner your analytics surface AI-referred traffic as its own channel (separate from organic search), the sooner you can see what the agent is showing the buyer.