Sidekick answered everything we asked. That was the problem.

Field notes from an infrastructure company that tested the platform's free AI hoping to be smug, and wasn't.

We spent a week trying to catch Sidekick failing. It kept answering.

Ask for last month's numbers in plain English and it writes the query itself. Ask for a flash sale and the discount exists before the meeting ends. The product description it drafted was fine, which annoyed us more than we'd like to admit.

We run catalog infrastructure for Shopify brands. Sidekick ships free with the plan. You can see why we wanted it to stumble.

Every quarter it gets better, and every quarter someone asks

It is not slowing down. The Spring '26 Edition shipped 150-plus updates, with Sidekick sitting in the front row of them. It now reaches into installed apps, starting with Judge.me, Klaviyo, Loop, and Smile, answering questions and taking action inside them. It keeps working in the background while you do something else. It takes voice. You can ask it about your business from a wristwatch.

Shopify Magic covers the writing side: product descriptions, pages, blog posts, generated right in the admin, free on every plan.

So every quarter someone asks us the reasonable question: doesn't this make you obsolete?

Our first theory was the comfortable one. We went hunting for a capability ceiling, some quality line the model couldn't cross. That theory did not survive the week. The output is fine and improving, and if your moat is "the AI writes worse copy than we do", you're renting.

The ceiling wasn't in the model

It was in the conversation.

Sidekick is an assistant, and Shopify's own docs describe every ability the same way: a merchant asks, Sidekick acts, the merchant reviews. The unit of work is a request. Even the extension system is tuned to that tempo. An app plugged into Sidekick gets three seconds to come back with an action. The whole reply has to fit in 4,000 tokens. Snappy, chat-shaped, and correct for the job it was built to do.

Now look at a catalog the way we see one. Across client stores we've processed 642,000+ SKUs, and the defects that cost real money were almost never something a merchant would think to ask about. Nobody asks which variants quietly lost their size attribute. Nobody asks which products contradict their own metafields in a language they don't read. You cannot ask about a problem you don't know exists.

The expensive problems never open a chat window.

What infrastructure does when nobody's watching

Infrastructure is the opposite shape. Nothing prompts it. It runs on a schedule and owns the outcome, with humans reviewing exceptions instead of typing requests.

Our version of that shape is 147 production workflows. On a normal day they move about 5,000 products through checks, enrichment, and verification, and no one typed a prompt for any of them. That is the whole difference. Nobody has to remember to ask.

Where the line actually sits

Here is where we honestly put it. Under roughly $250K a month in store revenue, don't hire us. Sidekick plus Shopify Magic, both included in your plan, will cover more of your operations than most owners expect, and buying an infrastructure team at that size is buying a snowplow for one driveway. Above the line, the catalog stops fitting inside a conversation, and the question changes from "what should I ask" to "who is watching all of it, all the time".

Everything above about what Sidekick does today comes from Shopify's own docs. What happens next is ours. Our bet, on the record: Sidekick keeps compounding, runs longer jobs across more apps, and by mid-2027 handles most in-the-moment merchant requests end to end. What it won't become is a standing system that watches an entire catalog continuously and answers for the result, because that is not what an assistant is for. If Shopify ships exactly that, we were wrong, and we'll write that post too.

Three notes for the road

• Test the free AI before you pay anyone, including us. If it covers you, it covers you.
• Judge AI tools by interaction shape, not demo quality. Ask-and-answer scales with your attention; infrastructure scales without it.
• The dangerous catalog problems never arrive as questions. Budget for detection, not just generation.

We never did catch Sidekick failing. We found the question it will never be asked, because at catalog scale nobody knows to ask it. It answers everything. It can't hear silence.

If your catalog is past that line and you want to hear what it would say if it could talk, OKART's Infrastructure Stress-Test reads it end to end, read-only, delivered in 48-72 hours.

Sources

1. Sidekick: Shopify Help Center: what Sidekick does, background tasks, memory, voice, and app integration.
2. Shopify Sidekick: official product page: plan inclusion, capabilities, admin actions, language support.
3. Sidekick app extensions: shopify.dev: how apps expose data and actions to Sidekick, response-time and reply-size limits.
4. Shopify Magic: Help Center: Magic feature coverage and free availability regardless of plan.
5. Automatically generating product descriptions: Help Center: in-admin description generation workflow and its limits.
6. Shopify Editions Spring '26: 150+ updates: Sidekick app integrations, background multitasking, voice, Apple Watch.