Shopify's new AI design workspace can now handle much of the routine theme work brands used to brief an agency for. The decisions that make a store sell still sit with people, and the code underneath matters more than it did.
Within a few days of Shopify launching Canvas, our Shopify team had it running on a test store. It could build a theme from a description of a brand's style, mood or season. It could not tell us whether any of it would sell.
That gap is the most useful way to understand what Shopify released on 1 October, and what it changes for brands building on the platform.
Canvas is a visual workspace that lays out every page of a store in one place. A merchant can edit pages directly, or describe what they want and have Sidekick, Shopify's AI assistant, build it, including changes to the theme code itself. Shopify's announcement is open about the ambition. For most of the platform's history, a store that looked distinctive meant writing code, paying a developer to write it, or wrestling with a settings panel. Canvas is Shopify's attempt to take that barrier away for every merchant.
It's a significant step, and one we've been expecting. In April, when Shopify released its AI Toolkit, Tim wrote in Offcuts that Sidekick was the merchant-facing layer of Shopify's AI, with clear limits on what it could touch. The piece that was missing, he argued, was governance: drafts, audit trails, rollback and approval before anything reaches a live store. He expected Shopify to close that gap by giving Sidekick more execution capability with guardrails attached. Six months on, Sidekick can edit theme code. The guardrails are still catching up.
Our Shopify team tested Canvas as soon as it reached one of our test stores. Access was limited in the first rollout, but the functionality we could reach was strong. In practice it covered most of what a brand would once have briefed an agency to deliver on a custom theme:
The team's summary was blunt: in short, Canvas can do what a traditional agency does to create a custom theme for a brand.
Our developers also looked at it from the code side. Themes built on our framework already work with Canvas, so changes on a client store can happen very quickly. The catch is that the code is written by Shopify's own AI, so we don't have full control over what it produces. If Sidekick can be made to follow our style guidelines and coding standards, it becomes a very fast way to make changes. If it can't, someone has to check every line it writes. That question decides how much time Canvas actually saves, and it's the one we're testing now on our own themes.
The list of what Canvas can't do was just as specific, and more interesting. It can't tell a brand whether a new section will lift conversion. It can't judge whether the structure of a store gives customers more value, or whether a design will bring in the right customers at all. And it can't brief itself. Getting Sidekick to build the right thing depends on the person writing the instruction knowing what the right thing is.
Those are the questions a brand is really paying for when it hires a team to build a store. Laying out a page has rarely been the expensive part of a build. The cost sits in knowing which product information belongs at the top of the page in a category with high return rates, how a subscription option should sit next to a one-off purchase, or which pages a merchandiser will need to change every week through peak season. That knowledge comes from having built and run stores, and from watching what customers actually do on them.
Canvas makes building a layout fast. It also makes building the wrong layout fast. A brand can now publish several untested variations of a page in the time it used to take to brief one, and without a way to measure which one works, more output simply means more guesswork reaching customers.
Canvas is early. It's in early access, available only to certain stores and only on desktop. Shopify has also been clear about what it doesn't support yet. Theme app extensions don't work in Canvas, so app blocks and embeds aren't available in a theme created or duplicated there, although Shopify says support is coming within weeks. Rollouts, Markets and Translations aren't supported either.
For many of the brands we work with, those gaps cover the parts of the store that matter most commercially. Reviews, loyalty schemes and size guides usually arrive as app blocks. Brands selling across several countries depend on market-level content and translated storefronts. Until Canvas supports them, integrations and multi-market setups will stay with a development team.
Then there's the question of where changes get made. An AI editing theme code directly on a live store, with nobody reviewing the result, is exactly the scenario Tim warned about in April. Our advice to clients is straightforward to follow: use Canvas on a duplicate theme, never on the live one, and have someone check what it has built before it's published.
Brands should still use it. The right way to think about Sidekick is the way we think about Claude in our own work: an employee or a collaborator, not the boss. It's fast and capable, and its work gets checked before customers see it.
There's long been confusion in the market about custom themes versus off-the-shelf ones, and we try hard not to frame client conversations that way. Our position has been that ROI², the framework we build client stores on, gives brands complete freedom over their store while making our own work more efficient. Canvas will bring the old question back into sales conversations in a new form. If AI can generate a bespoke theme, why pay anyone to build one?
Our view is that the foundation underneath decides how useful AI is on top of it. Stores built on ROI² already work with Canvas, so brands on our framework get the speed of AI editing from day one. They also get a codebase with consistent patterns, documentation and standards for an AI to follow. We made the same argument recently about our headless builds: an AI agent dropped into an undocumented codebase will guess, and guessing in a checkout is expensive. AI amplifies whatever engineering culture it lands in. Strong standards make it faster. A mess gets scaled.
Canvas puts that dynamic in the hands of every Shopify merchant. A brand with a clean, documented theme will find Sidekick's changes easier to review and safer to publish. A brand sitting on years of patched-together customisation will find out quickly how much the AI is guessing. Brands planning to use Canvas seriously should get their theme code tidy and documented first, before Sidekick starts building on it.
Routine storefront work will move away from agencies, and we think that's a good thing. A new landing page section, a seasonal refresh, a small layout fix: a merchandiser with Canvas will increasingly be able to make these changes without opening a ticket or waiting on a sprint. That frees development time and budget for the work that actually moves commercial results.
That work is changing shape. As AI takes on more of the routine edits, the value of a build team moves towards the parts AI can't judge:
For brands already on ROI², Canvas is mostly good news. They get a faster way to make day-to-day changes on a foundation designed for AI to work with. For brands on themes nobody fully understands any more, it's a prompt to fix the foundation before handing the keys to an AI.
Shopify will keep adding to Canvas, and the gaps we've described will close over the coming months. The underlying shift won't reverse. When every merchant can build a theme, the advantage goes to the brands that know what to build, and that have a foundation solid enough to build it on.
