Shopify is quietly reworking the AI behind its platform, and the change is already starting to show results. The company has replaced a closed, third-party model with a fine-tuned version of Alibaba’s Qwen3-32B.

On the surface, it’s a technical shift. In practice, it simply means the system is now faster and far more cost-efficient.

The update is closely tied to Shopify Flow, a feature many merchants use to automate routine work without writing code. Simple actions—like tagging certain orders or triggering notifications—are handled through Sidekick, Shopify’s AI assistant that understands plain English and turns it into working tasks.

Earlier, Sidekick depended on external AI models to interpret these requests. Now, Shopify is building that intelligence in-house, using data based on how merchants actually create and use workflows on the platform. This gives the system a more practical, real-world understanding.

One of the bigger challenges came from Shopify’s own setup. Flow runs on a complex JSON-based structure, which isn’t easy for AI models to handle. To get around this, the team trained the model using Python instead—a language it already understands well. Once the task is completed, the output is converted back into Shopify’s original format so everything works as expected.

The first rollout, however, revealed a few gaps. When tested with a small group, fewer workflows were being completed than before. A deeper look showed that email-related tasks and more complex conditions were causing most of the trouble. It also became clear that many users wanted to edit existing workflows, something the model wasn’t initially trained to do.

Shopify has since started improving the system through regular retraining, using fresh data and feedback from real usage. Instead of relying only on basic usage metrics, the company is also focusing on more detailed ways to measure how accurately the AI performs.

Behind the scenes, the system runs on high-performance GPU infrastructure and is supported by Tangle, Shopify’s open-source machine learning platform. This allows the team to test, learn, and roll out improvements quickly.

Overall, the move shows Shopify is becoming more focused on building its own AI strengths rather than depending on outside tools. By using its own data and learning from real merchant behaviour, the company is laying the groundwork for more reliable and tailored AI features across its platform.

Also Read: Airtable Steps Into the AI Agent Game With Superagent