Shopify's continual learning and reinforcement learning flywheel has our models outperforming frontier models on the tasks at a fraction of the cost and at higher tokens per sec. Check out @cmazzaanthony speak to it at @PyTorch conference
How does Shopify outperform frontier AI models at a fraction of the cost? During #PyTorchCon North America keynote, @cmazzaanthony of @ShopifyEng explained @Shopify's "Flywheel" a state-of-the-art continual learning loop built on top of @PyTorch and @vllm_project.
As enterprises scale generative AI, the bottleneck quickly shifts from building an initial model to maintaining and improving its performance in production. Shopify solves this by turning real-world failures into a self-improving engine.
Watch the full keynote video on YouTube to learn how this continuous learning loop operates in production: https://t.co/4qxlQ4efNr
Introducing Steve: the World’s first AI CEO for fully Autonomous companies.
Powered by World Models, Recursive Self Learning, a Policy Engine and a team of specialist AI agents, it runs your business end-to-end 🔥.
Proof? Steve currently runs 3 businesses we acquired and they are all crushing it (numbers in thread 🧵)
We’re now launching the Steve Startups Program of $2M to acquire 5 additional startups for Steve to automate.
Join us in the Autonomous Enterprise revolution 🚀.
Sidekick going from a conversational agent to a coding agent as unlocked some killer scenarios. Canvas is one of those things. @Shopify is incredible at re-thinking things from the ground up to make Merchant lives better. I've been editing my store with canvas for months with Sidekick and it's just a delightful experience.
When we launched Canvas, I said I couldn't wait to see what everyone would make. Wow, did they deliver.
More than 2,500 Canvas stores are already live. Here are a few of my favourites, with links in the thread.
https://t.co/n1unp7xMRP
ChatGPT. Gemini. Claude. Copilot. Perplexity. Grok. Grok Bot. Muse. Claude Duet.
@Shopify connects to all of them.
And when a new AI tool comes along, we’ll build a connector there too
This one has been near and dear to me from both working on Sidekick, but also as a Merchant. A lot of my needs as a merchant end up finding their way into Sidekick, super proud of the team on this one. It was tough to get right!
@Shopify sidekick has come along way
I’m starting to create new “skills” to see what kind of tasks it can take over to save me time
Here’s a few that are useful so far. Ill continue to update this thread as i add more 🧵
Going to chat about productionizing agents, post training, and inference optimizations on @nebiusai live at #NVIDIAGTC Berlin in 35 minutes or so. Tune in!
Talking at 3pm at @nvidia GTC Berlin on how @Shopify post-trains open source models and optimizes inference using Nvidia's models and tools. Read our blog post about it here: https://t.co/bkKnkzhCCg
Nice work by @cmazzaanthony and @Shopify ML team on our flywheel. Cody is walking through our continual learning framework at @PyTorch conf that optimizes our task specific agents beyond frontier models at lightning speeds and at a fraction of the cost.
Now: @Shopify’s continual learning loop with PyTorch and @vLLM_project takes the stage at #PyTorchCon 2026 with Cody Mazza-Anthony @cmazzaanthony.
Using Buyer Profile as an example, Cody walks through how a calibrated judge flags low-scoring outputs in production. Critics diagnose each failure, and a replay with a critic hint is scored again by the judge. Accepted repairs expand and improve the training data.
Cody shows how a small specialist model can outperform a frontier model on this task, then turns to optimizing inference through prompt compression.
Watch the keynote livestream: https://t.co/6TtaEkw0x7
@Alex06425829@spisallyouneed yes we will building assistnts at Maluuba in 2011, been doing only that for last 15 years until today even! Though, i guess we call them agents now.
Had the pleasure of embarking on the wild journey of Maluuba with @spisallyouneed building agents over 15 years ago before people were even talking about Siri or ML. Pumped about Sky 001, the team is cooking!
Introducing SKY-001, the first Engineering World Model.
On CAD and BIM tasks, it beats GPT-6 Astra and every frontier model we tested:
-> 51-125x faster
- > 150-210x cheaper
-> Higher IOU than all frontier models
SKY-001 lets you dream a physical object before you build it.
18 months of work. Built by the team out of Maluuba's deep learning lab (acq. by @MSRResearch).
The first ‘AI General Engineer’ will be the backbone for Autonomous Manufacturing.