Shopify Tech Week was a lot of fun! They had four different talks going on, each very interesting, so I wanted to write about my thoughts on them.

About Universal Commerce Protocol

The first was about Universal Commerce Protocol, which is a protocol they've been working on with other e-commerce platforms to standardize the way that information is stored in commerce, making it easier for both APIs but mostly AI agents to do shopping, because they see that as the future. It enables flows like a user asking for a certain item. They send a request to what's called the catalogue, which then returns a list of the most relevant items, let's say ten, and from there they can ask for more details on a specific item. They can also buy directly through UCP, enabling cleaner shopping flows.

It was an interesting talk. I think mostly what I got out of it is how they're actively thinking about the future of commerce, what they see the future to be, and realized that they need a massive change in the way things are done. Because without a universal protocol, it will be near impossible for AI or any other automation to seamlessly interact with shopping. Even as someone who isn't as invested in e-commerce as them, I liked the push for standardization to help everyone.

Integrating with other systems

I talked with one of their product managers afterwards about applications that could surround UCP and the possibility of sponsored listings. It was unclear from the presentation if each shop owns it own catalogue, but I got clarification that Shopify is the universal source of truth for the catalogue, or at least their catalogue and all shops on it. So sponsored listings won't actually occur. The only time you might get a sponsored ad or product is if you're using someone else's application, where they built it to fetch from Shopify but then inject some sponsored items into whatever they retrieve for you. I'm sure people will build their own commerce platforms that basically just pull from Shopify so that other people don't have to build the code to do it themselves. Hopefully we get some official options that have no ads, but we'll see.

SimGym

The second talk was about prototyping with SimGym, their simulation engine for shops. It simulates how real people react with the shops based on years of data and allows vendors to experiment with how changes to their website will likely affect shoppers. There were some good questions asked there about how they deal with data for new shops with no pre-existing data. Shopify then pulls data from similar shops to help model the simulations for the new shop, and it works very well from what they've said. Super cool application of machine learning that I never initially expected from an e-commerce platform.

River - Internal Slack AI

The third talk was about an internal tool called River, and that one was incredible. They have been migrating everything to a monorepo to make it easier for AI to work across apps. Something I thought about, though, is whether a monorepo is really the best idea — and I say that not because I don't understand where they're coming from, but Git has submodules which do the same thing with more organization. The idea behind a monorepo is: one, you need everything under the same folder structure, so you need some root folder where everything the agent could possibly need is somewhere under that root; and two, for working with others, you need version control. My idea with submodules is that you can have everything under the same root, and it would make things easier because when you're pulling, if you're doing some isolated work, you don't suddenly get 10,000 lines of code changed every time you need to pull — you just get what's relevant to your repo, or the repos you're working with within the monorepo. Maybe that's actually more complicated and I simply enjoy the idea of dividing things because that's how I like organizing. But a monorepo can actually divide things in its own way, just not the way I have in my mind.

Regardless, River is really cool. They have it as a Slackbot and it enables everything you could imagine for development. You ask it for changes, it includes a whole memory layer for all the information they have. They transcribe their meeting minutes, it can fetch stuff from meeting minutes, it can fetch stuff from Slack channels.

Help for one, help for all

The coolest thing about it is the fact that you can only use it in a public channel. By making it public only, everyone benefits from everything you say to River. You ask it to clarify something, you make a bug fix — that's all accessible to everyone to learn from. They can also learn how you're using it, and they can chime in. They said it was really interesting how they didn't initially imagine this, but people started doing a kind of multiplayer with River — one person would start asking something, then others would chime in and help and build off of it in the same thread. That's really cool to me, because since it's built within Slack, the context is simply the conversation and all the other Shopify data like the codebase and meeting transcripts. It doesn't really matter who sends the message — it's just a free-for-all, and River simply prioritizes the current thread it's in for what it's actually responding to, as opposed to what context it uses to inform that response.

Quantum computing

The last talk was about quantum computing, which was really interesting because I had just talked to the Wealthsimple CTO about quantum the day before. The talk was given by the CEO and founder of his own company, and what made it distinct is that while typical quantum computing uses electrons, they use photons. He said that with photons they can operate at room temperature — around 300 Kelvin — as opposed to electrons, which need to be kept at essentially near-absolute-zero temperatures, something like 0.00001 Kelvin. Crazy stuff.

The chat after was great. I talked a lot more with Victoria, who is a product manager at Shopify. Shopify's PMs are a lot more technical — they hire very technical people, and her job is heavily focused on direction: what is good, what will work. She thinks the future belongs to product managers, especially because if it's easier to code, a lot more people will start moving into product. I'm curious how that will actually work out, because obviously not everyone can go into product. If you have 10,000 people all chiming in with different ideas, how do you actually make a decision? At that point you're almost crowdsourcing the direction of your company to a bunch of PMs, and they all go out, do their own thing, use AI to prototype stuff, and eventually someone with the necessary seniority picks one out of the hundred ideas. Still, a very interesting conversation, and a very, very good day overall.