Shopify Dot Dev 2026: Two shifts in how we build, and how we work

Takeaways from two days in Toronto

Shopify's developer event ran over two days in Toronto in July, closing with a party downtown that brought together Shopify staff, dot dev partners, developers and app builders. The sessions covered a lot of ground, but the highlights were the time spent discussing product with Shopify's team and working through the same hard problems from different angles with the agencies and app builders in the room.

Plenty was announced. Two things stood out as likely to change both the way we serve merchants on the platform and the way we operate as an agency: the Universal Commerce Protocol and River. They sit in different areas but point to the same need - to get the knowledge and data buried in system internals or people's heads out into the open, in a structured form that agentic systems can use.

UCP and agent-driven channels

The Universal Commerce Protocol was co-developed with Google and published in January, with Etsy, Target, Walmart and Wayfair alongside Shopify's merchant base.

The UCP is simply a handbook of rules dictating how autonomous agents interact with a merchant store and what they can do on a customer's behalf. Agents publish profiles declaring what they can handle - "I can spend no more than $50", or "I only want items delivered for free" - and at each step of the buyer's journey the merchant's agent responds with a negotiated result - "Here are my products under $50", or "I can't deliver any items for free". Much like human-to-human interaction, they either reach an agreement or they don't. An agent also has defined points at which it must bring a human into the conversation, such as entering payment details or validating age. In this way agents respect human boundaries and humans remain the deciding factor when they want to be.

The consequence for our work is that we now have to build with a second kind of buyer in mind. An autonomous purchasing agent doesn't browse a category page, respond to a hero banner or read a size guide. It reads a declared profile, compares options across merchants, and completes or escalates. Everything we've historically relied on to influence a purchase (layout, merchandising sequence, promotional messaging, the checkout experience itself) is invisible to an agent. What it can see is structured product data, declared fulfilment options, available payment handlers, pricing and promotion logic, and whether the checkout can be completed without a human.

That reframes some familiar work. Product data quality stops being an SEO task and becomes channel eligibility. Fulfilment capability that currently lives in middleware and undocumented business rules needs to be expressible. Loyalty and B2B pricing need to be declared rather than applied at the last step. Where that data isn't available, an agent will route to a merchant where it is, because that's the only place it can follow its instructions.

Agentic storefronts are live, but agent activity is still low. For merchants, that makes this a conversation to have now, while deciding what role the channel will play for them.

River and how internal AI adoption is being handled

Shopify published the architecture behind River, its Slack-native agent, in late May. This agent has become widely used across the Shopify organisation. In one recent 30-day period, around 60,000 sessions ran across more than 5,000 Slack channels, and roughly one in eight merged pull requests across the company is now co-authored by it. Javier Moreno, who co-authored that post, ran a session covering the work in more detail: how the problem was framed at the outset, the issues that surfaced along the way and how they were resolved, and where the team is taking it next.

The underlying idea is that everyone at Shopify has access to a single expert in all things Shopify. Every interaction with it (asking a question, correcting an answer, clarifying an output) adds to what that expert knows, and those interactions come from across the organisation rather than the engineering team alone. The groundwork was laid in 2024, when Shopify consolidated its codebases into a single monorepo and rebuilt its environments on a reproducible substrate, so agents could navigate the full codebase and reach the conventions, decisions and changelogs written down alongside it. As a result, agents hold far more context about the codebase as a whole. River leverages this change and has also been deliberately constrained to public channels, with no direct messages. Every conversation with it in Slack is a public record, sessions are searchable, and those conversations feed back into the agent's skills and defaults.

What this means for our teams

Both shifts ask the same thing of us: write down what we currently hold in our heads or bury in code.

On the build side, that means treating product data, fulfilment rules and pricing logic as an interface rather than an internal function. This information has to be exposed and interpretable to agents at all times.

On the way we work, it means being deliberate about where our own AI use happens. An agent that only makes one person faster leaves us with exactly the gap we have today: the knowledge goes when they do. An agent working in a shared channel, in context that's written down, with decisions recorded next to their outcomes, improves the whole team.

The same conversation belongs with our merchants. Most of their teams are already using AI privately and getting the individual benefit without the organisational one. Helping them make that work visible is worth as much as anything we build for them.