Issue #: 015
Send date: Nov 2, 2026
Subject line: Your next customer checks out inside a chat.
Preheader: Shopify just opened checkout to AI agents. Your product data is the storefront now.

The storefront moved. Most product pages didn't notice.

On October 28, Shopify shipped its Fall '26 Edition, and the headline is simple: agents can now buy. Browser agents can search a store, update the cart, and finish checkout once the buyer approves, on every Liquid and Hydrogen storefront. Shopify says every store is agent-ready by default. Shop Pay checkout now runs inside personal agents like Muse, Claude Duet and Instinct, and Shopify counts more than 400 million Shop Pay accounts. That's the payment rail an agent needs to finish the job.

Shoppers are already moving. Accenture's holiday survey of 4,048 US shoppers found two in five plan to use gen AI for holiday shopping this season. On average, they'd let an agent spend $102 without asking them first.

In #008, we said the money sat in the gap between an agent holding a payment card and product pages AI can't read. The checkout half just arrived. The product data half is still the merchant's problem.

So where's the money?

An agent doesn't see your photos or your brand story. It reads titles, variants, prices, and policies, then picks. Thin listings lose to clean ones, and nobody tells you that you lost. That makes catalog cleanup a real deliverable: for your own store if you sell, and for every merchant you know if you don't.

Black Friday is November 27. That's under four weeks.

● Beginner: see what the agent sees · ● Operator: fix it, then sell it

MOVE 01 · BEGINNER
Text. An agent shopping test is one plain-language shopping request that you run in three assistants to see who gets recommended.

Objective: you finish this move knowing whether an agent finds your product, and what it says about you when it does.

Write the request the way a real buyer would: "Find me a [product] under [price] that ships to [city] by [date]." Run it in ChatGPT, Claude, and Gemini. Write down who shows up, what price the agent quotes, and what it says about returns.

No store? Run it for a client or a friend's brand. That test is the opening slide of Move 06.

Tools: ChatGPT · Claude · Gemini

MOVE 02 · BEGINNER
Workflow. Fill with Gemini fills a column of Google Sheets cells with AI answers, and each cell now shows the web source it used.

Objective: you finish this move with a competitor price and policy table where every cell links to its source.

Put 10 competitor product names in column A. Add headers for sale price, return window, and shipping promise. Run the AI function down the columns. Since October 28, Gemini cites the exact source inside each cell and handles up to 3,000 cells per request.

Click five citations at random before you trust the table. That's the second-source rule from #004, built into the sheet. Available on Business Standard and Plus, Enterprise Standard and Plus, and Google AI Pro and Ultra.

Tools: Google Sheets · Gemini

MOVE 03 · BEGINNER
Text. An agent-readable listing is a product description written for a reader that can't see the photos.

Objective: you finish this move with one listing an agent can quote back without getting anything wrong.

Pick your best seller. Paste the listing into Claude and run this: "Act as a shopper's AI agent. List every question a buyer might ask about this product that this listing can't answer." Then answer each one in the listing itself: size, material, what's in the box, who it's for, every variant spelled out.

Check your compare-at price while you're there. Shopify now passes sale prices to agents automatically. An inflated "was" price gets repeated to the buyer word for word.

Tools: Claude · ChatGPT

MOVE 04 · OPERATOR
Workflow. Shopify's Catalog eligibility check shows whether a store qualifies for the catalog agents search, and what's blocking it if not.

Objective: you finish this move with the store eligible and the facts agents receive about it published by you.

Run the eligibility check under Sales channels > Agentic and clear what it flags. Then open the knowledge base on the same page. It holds what agents get told about your return window, shipping destinations, policies, and FAQs. Nothing Shopify drafts reaches an agent until you publish it.

If an agent misstates your returns policy, the customer blames you, not the agent.

Tools: Shopify

MOVE 05 · OPERATOR
Workflow. A catalog round trip exports your products to a spreadsheet, rewrites them with AI, and imports them back only after a human reads the changes.

Objective: you finish this move with your top 20 products rewritten to the Move 03 standard, with nothing going live unchecked.

Export the products CSV. Have Claude rewrite titles and descriptions against your Move 03 question list. Compare old and new side by side in Sheets, fix anything invented, then import.

Shopify's connectors let ChatGPT, Claude, Gemini, and Copilot edit products directly. Don't, for this job. A bulk edit to live listings three weeks before Black Friday is the checkpoint case from #005. Once the rules work, save them as a skill (#014), so the next catalog takes an hour.

Tools: Shopify · Claude · Google Sheets

MOVE 06 · OPERATOR
Text. An agent-readiness sprint is a fixed-scope project that gets a merchant's store found and quoted correctly by AI agents before peak season.

Objective: you finish this move with a one-page offer you can send to three merchants this week.

Scope it to five things: the Move 01 test before and after, the eligibility check, a published knowledge base, the top 20 listings rewritten, and the sourced competitor table from Move 02. Price by the project (Model 01 from #002). Add a monthly retest as the retainer, because agents change what they surface.

Promise readable, not ranked. Nobody controls which product an agent picks.

Tools: Notion · Google Sheets

Your first agent shopping test. About 20 minutes.

Pick one product. Yours, a client's, or something you'd buy anyway.

Write the request like a buyer. Include a budget, a deadline, and one detail that matters: "Find me a waterproof dog jacket for a medium dog, under $60, delivered to Miami by November 20."

Run it in three assistants. ChatGPT, Claude, and Gemini. Same words in each.

Log it in a sheet. Four columns: assistant, brands recommended, price quoted, and what it said about returns.

Read the returns column first. That's where agents most often guess.

Fix one thing. A missing size chart, a vague return policy, a variant with no name. Then rerun the same request next Monday and compare.

Q: “I want to learn automation in biotechnology and build AI devices in the field. Is it possible to make an AI-based bioreactor?”
Muhammad Sajjad

Muhammad, yes. It’s already real, and you can build a small one yourself.

An AI bioreactor is mostly software that reads sensors like temperature, pH, and oxygen, then decides when to feed the cells. Researchers have been testing machine learning on exactly that decision for years.

If we were you, we’d start with a Pioreactor. It’s an open-source bioreactor that runs on a Raspberry Pi, tracks growth in real time, and lets you plug in your own Python code. Log a few runs, then add one decision for the AI to make. You stay as the checkpoint.

BUY · Shopify's agent controls: the eligibility check and the knowledge base.

Where it's strong: both are free, and they take minutes to run. The knowledge base is the only place you control what an agent says about your policies. The order and the customer record still land in your admin, according to Shopify's VP of Product.

Where it breaks: the browsing and comparing now happen inside someone else's chat, and you don't see them. The adoption and account numbers are Shopify's own. Canvas, the new chat-based store designer, is in early access, so keep it off a live store until after the holidays. Several new fraud tools are labeled "rolling out soon," so don't build your Black Friday fraud setup around them.

The rest of the shelf: four things from the last week worth your attention.

WATCH · GPT-6.1 Terra. OpenAI launched it on October 26 as a fast model for computer use and visual reasoning. OpenAI reports it finishes computer-use tasks more than three times faster than Astra. It's in ChatGPT Work and Codex only, not regular Chat. OpenAI's own example is reading invoices and typing them into accounting software. Same rules as #007 and #011: one bounded task, a drift log, and a checkpoint before anything touches money.

WATCH · Gemini 4 Carbon. Google announced it on October 28, two weeks after Argon (#013). Only selected cyber-defense organizations can use it so far, and there's no date for everyone else. Nothing to do yet.

BUY, if your team lives in Microsoft 365 · Copilot skills. Users can now create skills in Copilot Chat in plain language and find skills colleagues have shared. That's #014's idea arriving where a lot of client teams already work. Rolling out worldwide in November.

BUY · Whole-deck edits in Google Slides. One prompt can now restyle or rewrite an entire presentation. Edits count as yours, so undo and version history work. English only at launch, rolling out since October 29.

Carried forward. The skills from #014 hold your listing rules for Move 05. The checkpoint from #005 still sits between AI and anything live. The second-source rule from #004 is now one click inside a Sheets cell.

A weekly dispatch for founders, freelancers, and operators who treat AI as a business tool, not a hobby. Every Monday, one question: what changed in AI this week that you should actually do something about?

Anthony | Founder of Uncover AI