Tobi never stops pushing on AI, inside Shopify and across the platform itself. Three updates from the past month caught our attention at Ambaum. Each one says something specific about how commerce is changing, and what Shopify Plus merchants should be doing about it. Here is the rundown, plus our read on what to do next.
River, Shopify's in-house AI teammate
Tobi built River, an AI agent that lives inside Shopify's company Slack. You talk to River exactly the way you would talk to any teammate. Mention River in a channel and go from there.
River reads code. It runs tests. It writes code. It opens pull requests. It queries the data warehouse. It looks at production traces. The list keeps growing. In short, River does a lot of what a senior engineer does, only faster and at scale.
The constraint that makes it interesting
What separates River from a hundred other internal AI tools is one hard rule: it only works in the open. Every action, every prompt, every commit happens in a public Slack channel where anyone in the company can see it. No DMs. No private workspaces. No invisible workflows.
That rule is the real product. It turns an AI tool into a transparency engine. Anyone curious about how a piece of code got written can scroll back and watch the conversation that produced it. Anyone wondering about a data query can see exactly what was asked. Mistakes are visible. So are wins. Knowledge compounds in public instead of getting trapped in private threads.
Why merchants should care
Most Shopify Plus merchants are not going to build their own River. The lesson is not the tool. The lesson is the design choice. When you bake AI into your team's workflows, the rules you set around it matter more than the model you pick. "Only in the open" is a constraint with a clear payoff: shared learning. Pick your own constraints just as deliberately.
/llms.txt and /agents.md, now native to every storefront
Shopify now auto-generates two storefront routes on every store: /llms.txt and /agents.md. Visit either path on any Shopify store today and you will find structured content explaining what the store sells and what agents are allowed to do there.
What llms.txt actually does
llms.txt is a proposed standard for telling large language models how to read your site. Think of it as a sitemap built for AI. It points at the content you most want indexed, summarizes your brand in plain language, and structures the information so an LLM does not have to crawl a hundred pages to figure out what you sell.
agents.md is the action layer. Where llms.txt describes your store, agents.md describes what agents can do with it. Browse collections. Compare products. Initiate a checkout. Track an order. As AI shopping assistants get more capable, this file is where you spell out the contract.
How to customize yours
Want to override the default? Add a templates/llms.txt.liquid file to your theme. Same pattern as overriding robots.txt. From there you can shape what AI agents see: which collections matter most, what your value proposition is, which sizing or compatibility details belong up front, and which CTAs to highlight.
A good custom llms.txt for a Shopify Plus merchant should answer a few questions in plain language. Who is this brand for? What do you sell that no one else does? Which collections drive the business? What information do you wish a shopper had before they asked support? Treat this file like the page you would write if a buyer had thirty seconds to understand your store.
This is small in scope and big in implication. AI agents are first-class citizens on the storefront now. The merchants who write a thoughtful custom llms.txt today are setting the rules for how their store gets represented in every AI-driven discovery moment that follows.
AI rewards specialists, not the top of the bestseller list
Here is the number that should reshape how merchants think about assortment, product copy, and brand positioning: 71% of AI-attributed purchases in 2025 came from categories outside the top 100.
Read that again. The long tail is where AI commerce lives.
Why this happens
Search engines and AI agents rank results in fundamentally different ways. A search engine rewards popularity. It looks at clicks, backlinks, reviews, brand authority, and a hundred other signals that tend to compound for whoever is already winning. The output is a ranked list of the most popular options.
An AI agent does not produce a ranked list. It produces a recommendation. It reads the user's specific question, weighs the context, and tries to find the single best match. That changes everything downstream.
The kite example
When a buyer asks an AI assistant for "the best kite for flying when there is no wind," the agent is not going to return the bestselling kite. It is going to find the specialty brand that makes a kite engineered for low-wind conditions. The merchant who specialized wins. The generalist who happens to carry kites does not get surfaced at all.
The same logic plays out across every category. The buyer asking for a desk lamp that does not flicker on camera. The buyer looking for shoes that fit a wide foot in a narrow toe box. The buyer hunting for a coffee maker that hits a specific brew temperature. Specialists win these conversations every time.
What this means for Shopify Plus merchants
A few practical takeaways.
- Audit your assortment for specialization. The products where you are clearly the best answer to a specific question are now your most valuable SKUs. Promote them. Document them. Write about them.
- Rewrite product copy for clarity, not for keywords. AI agents parse plain language. A spec sheet that explains use cases, edge cases, and what the product is not for is more valuable than a keyword-stuffed description.
- Build category pages around questions, not categories. "Low-wind kites" beats "All kites" when an AI is matching intent.
- Invest in unique data. The reviews, photos, sizing guides, and FAQs your store owns are what an AI uses to recommend you over a generic competitor.
- Lean into the niche. AI commerce penalizes the middle. Brands without a clear point of view get skipped.
Shopify's full breakdown of the data is here.
The Tobi quote we keep coming back to
"The company moves at the speed of its slowest secret."
That one stuck with us. The lesson cuts every direction. For Shopify. For Ambaum. For every merchant we work with. If a critical piece of context lives in one person's head, that bottleneck sets the pace for everything that depends on it.
How this plays out in practice
Think about the secrets quietly slowing your team down right now. The reason a product was reformulated three years ago. The customer service workaround that only one rep knows. The campaign that flopped and never got documented. The Klaviyo flow that someone built and nobody maintains. The vendor contact whose phone number lives in one operator's personal phone. Each of those is a slowest-secret in the making.
River's "only in the open" rule is one answer to this problem. Better internal docs are another. Public planning, shared dashboards, recorded decisions, and async-first communication all chip away at the same thing. The goal is the same: keep no secrets that only matter because someone forgot to write them down.
What we are doing with all of this at Ambaum
Three signals from one month, and they point in the same direction. Shopify is building for an AI-native commerce stack. Merchants who treat that as a "later" problem are going to wake up to discovery channels they do not show up in and competitors who do.
Here is where our team is focused right now.
- Auditing every Shopify Plus merchant we work with for AI readiness, starting with custom llms.txt and agents.md.
- Rewriting collection and PDP copy to answer specific buyer questions instead of just ranking for keywords.
- Documenting brand specialization clearly so AI agents have something concrete to recommend.
- Borrowing the "only in the open" pattern internally so our team's work compounds across clients.
If you are a Shopify Plus merchant and you want help thinking through what these shifts mean for your store, reach out. We are watching this space closely, and we would rather get our clients ahead of the curve than help them catch up later.
