How to use shopify flow AI to create workflows?

Updated 2026-07-15880 searches/moRanked #353 of 519· AI jobs and future of work
Short answer

Install the free Shopify Flow app, open it at admin.shopify.com/apps/flow, and describe the automation in plain English to Sidekick — for example, "when a customer spends over $500, tag them VIP and alert me in Slack." Sidekick drafts the trigger, conditions, and actions in the Flow editor for you to review before activating.

Why — the first-principles explanation

Every Flow workflow is the same three-part sentence: trigger, condition, action. The trigger is the event that wakes it up (order created, customer account made, inventory dropped). The condition is the filter that decides whether to continue (order total over $500? product tagged fragile?). The action is what happens (add a tag, send an email, call an API). Once you see that structure, the whole product stops being intimidating — you're writing when this, if that, do this, and nothing else.

The AI part exists because translating a sentence in your head into that structure used to be the tedious bit. You had to know which trigger existed, what it was called, and which fields it exposed. So Shopify built an agent that does the translation: you describe the outcome, it picks the trigger, wires the conditions, and fills in the actions. Shopify fine-tuned an open model — Qwen3-32B — as a tool-calling agent for exactly this job, running it 2.2x faster and 68% cheaper than their previous approach.

There's a genuinely interesting engineering detail that tells you how to use it well. Shopify found the model was much better at generating Python than at generating Flow's native JSON — 22 points better on syntactic correctness, 13 points on semantic correctness — because Python looks like reasoning and dominates pretraining data. So the model writes Python and a transpiler converts it to JSON. The lesson for you: it's strongest on clean logic, and weaker on fiddly configuration.

Which is exactly where you should expect it to stumble. Shopify reported a 35% gap between benchmark scores and real activation rates, and named the weak spots: editing existing workflows, email configuration, and third-party integrations. So the reliable pattern is to let AI draft from scratch, then verify and tune by hand — never assume a generated workflow is correct because it looks correct.

An example that makes it click

Think of Sidekick as a contractor who's built a thousand kitchens. You say "I want an island with the sink in it and stools on one side" and they come back with a drawing in twenty minutes instead of you learning CAD. The drawing is usually good. It is not automatically right — they might have put the sink where your plumbing isn't.

So you look at the drawing before anyone knocks down a wall. Same with Flow: read the trigger it picked, check the condition's number is really 500 and not 5,000, and run it on one test order before you let it loose on every customer you have. The AI removed the drafting work, not the checking work.

How to do it

  1. Install Shopify Flow — it's a free app from the Shopify App Store, available on the Basic, Grow, Advanced, and Plus plans.
  2. Open Flow in your admin at admin.shopify.com/apps/flow, or find it under Apps.
  3. Describe the automation to Sidekick in plain English, stating the event, the filter, and the outcome: 'When an order over $500 is placed, tag the customer as VIP and send me a Slack alert.'
  4. Review the draft Sidekick produces in the Flow editor — confirm the trigger is the event you actually meant, and check every threshold and tag name character by character.
  5. Add or edit steps by hand where needed. Flow's building blocks are triggers, conditions, and actions; you can add conditions to narrow when the workflow runs.
  6. Test on a single low-stakes order or customer before turning it on, since the workflow will run automatically on live store data.
  7. Turn the workflow on, then check the run history after the first real triggers to confirm it fired when expected and skipped when it should have.

Key facts

Infographic: How to use shopify flow AI to create workflows — short answer and key facts
Visual summary — How to use shopify flow AI to create workflows?
▶ The 60-second explainer (script)

Here's how to build a Shopify Flow workflow with AI. First, install Flow — it's a free app on Basic, Grow, Advanced, and Plus. Open it in your admin under Apps. Now, before you type anything, understand the one structure that makes all of this click. Every workflow is the same three-part sentence: trigger, condition, action. The trigger is the event that wakes it up — an order gets placed. The condition is the filter — is that order over five hundred dollars? The action is what happens — tag the customer VIP, ping Slack. When this, if that, do this. That's the whole product. So you just describe that sentence to Sidekick in plain English: 'When a customer spends over five hundred dollars, tag them VIP and send me a Slack alert.' Sidekick drafts the trigger, the conditions, and the actions right in the editor. Under the hood, Shopify fine-tuned an open model called Qwen3-32B for this — it's two-point-two times faster and sixty-eight percent cheaper than what they had before. But here's the part you actually need. Shopify themselves measured a thirty-five percent gap between how the model scores on benchmarks and how often its workflows really get activated. It's weakest at editing existing workflows, email setup, and third-party integrations. So: let the AI draft, then you verify. Check the trigger is the event you meant. Check the threshold says five hundred, not five thousand. Test on one order. Then turn it on.

What authoritative sources say

Shopify Help Center, Shopify Flowofficial — Shopify Flow is a free app available on the Basic, Grow, Advanced, and Plus plans, built from triggers, conditions, and actions, with the Send HTTP Request action limited to Grow/Advanced/Plus and custom partner app tasks limited to Plus. source ↗
Shopify Engineering, 'Flow generation through natural language: An agentic modeling approach'official — Shopify fine-tuned Qwen3-32B as a tool-calling agent to generate Flow workflows from natural language, running 2.2x faster and 68% cheaper, retrained weekly, with a 35% gap between benchmark performance and real-world activation and known weakness at editing workflows, email configuration, and third-party integrations. source ↗

People also ask

Does Shopify Flow cost extra?

No. Flow is a free app on the Basic, Grow, Advanced, and Plus plans as of 2026-07, though usage limits track your plan's API limits and some actions are gated to higher plans.

Can Sidekick edit a workflow I already built?

It can try, but Shopify's own engineering team identified editing existing workflows as one of the model's weakest areas in production. Expect to make edits manually and verify carefully.

What if Sidekick builds the wrong workflow?

You review the draft in the Flow editor before activating it — nothing runs until you turn it on. Always confirm the trigger and every threshold value, then test on a single order first.

Can Flow connect to tools outside Shopify?

Yes, via the Send HTTP Request action, but that requires the Grow, Advanced, or Plus plan. Third-party integrations are also a known weak spot for AI generation, so wire those up by hand.

What's a good first workflow to try?

Something reversible and low-stakes, like tagging high-value orders or flagging low inventory. Save irreversible actions such as customer emails or refunds until you trust the setup.

Related questions