Sean: I do think Jensen gotten a lot of bad press. That $500,000 engineer at the end of the year, I'm going to ask him how many token how much did you spend in tokens? If that person said $5,000, I will go ape something else.

Mikhail Parakhin: Yes, but uh I actually think that's undeserved. I think he he is actually right. Uh I do think he's directionally correct.

Mikhail Parakhin: The thing that I do want to say and this is something that we learned through trial and error and very important is like two things. One is that it's not about just consuming tokens. Uh you can consume tokens and and in fact the antiattern is running multiple agents too many agents in parallel that don't communicate with each other. that's almost useless uh compared to just fewer agents and burns tokens very efficiently. uh setting up the right critique loop, especially with the high quality models where one agent does something, the other one ideally with a different model critiques it. It takes much longer. So people don't like it cuz latency goes up. You know, they they have to wait till this debate is happening. But uh the quality of the code is much higher.

Sean: Okay, we're here in a studio, a remote studio with Muel Parkin, CTO of Shopify. Welcome.

Mikhail Parakhin: Thank you. Welcome.

Sean: Your team kindly prepared some slides actually... So here we have uh an internal AI tool adoption chart. What are we looking at here?

Mikhail Parakhin: Yeah, this is very interesting statistics. Uh this is number of daily active workers. You know, think of DAO basically daily active users of AI tool as a percentage of all the people in the company... one is the greenest total... it approaches really 100% by now. It's hard not to do your job now without interacting deeply at least with one tool. You could see another interesting thing is just as many people commented in December was the phase transition when suddenly models gotten good enough that that everything took off and started growing.

Mikhail Parakhin: The other thing I would claim you could see is that uh CLI based tools and tools that don't require you to look at the code becoming more popular... tools that require ID such as uh GitHub copilot or cursor they're not exactly shrinking but they're not growing as fast... we effectively fund unlimited tokens for everybody. Uh we we do we do try to control the models that uh people use but from the bottom not from top like we basically say hey please don't use anything less than us 4.6 (4o).

Mikhail Parakhin: ...the overall budget is just like lines of codes lines of codes are exploding for everybody right now... partially just because AI can write a lot more code you know doesn't get tired and so you have to have to have a very strong narrow waste during PR review otherwise just the number of bugs will go through the roof.

Sean: Totally. Uh, I noticed in your chart you didn't have any review tools.

Mikhail Parakhin: I haven't found a good PR review tool that that does what I think should be done... you want to run the largest models that means I know codex or or cloud code is not going to cut it. You need to have prolevel models. If you really want to uh stand the tide of bugs going into production... I actually don't mind a prolevel model taking an hour or two hours to review my PR because I have dealt with humans who take a week to review my PR, right?

Mikhail Parakhin: ...the real problem is since there's so much more code then uh probability of at least some tests failing going up... it's total time savings if you spend more time on a longer model like thinking for an hour because then then you you don't have to spend all that time during testing and rolling you know rolling back the deployment... clearly the old thing were designed for humans will need to be morphed into something new.

Sean: How would you explain them in together?

Mikhail Parakhin: Tangle first and then tangent is a thing on top of tangle. And the tangle is the third generation I claim of uh systems of uh running any data processing but bit with a skew for ML experiments... Everything is based on content hashes. So even if the version changed, but if the output didn't change, nothing is being rerun. It's very efficient. If you multiple people start experiments the same sort of data prep-processing, it's not repeated multiple times. It's automatically done only once.

Mikhail Parakhin: And tangent is basically an automatic auto research loop that can help and kind of do your work for you... our search recently we moved from it's hard to even quote from 800 QPS to 4200 QPS with the same quality just by pure optimizations and not a research loop that kept running and changing code... tangent in particular are extremely democratizing... PMs are like the highest user right now.

Sean: ...Sim AI... there's a very small cottage industry of people trying to do like the simulate customer thing. I think a lot of people maybe don't super trust this yet...

Mikhail Parakhin: That's exactly actually the thing I wanted to cover because if you don't have the historical data, all you can do is prompt agents in a vacuum... Yes, except Shopify has decades of history of how people made changes and what they what it resulted in terms of sales... our internal goals of correlation of hitting internal goal was to hit 0.7 correlation with add to cart events.

Mikhail Parakhin: ...you want to detect the effects like hey if I make my image just larger will I have more sales or fewer sales and like usually people's intuition here... is that I increase my images I'll have more because they look nicer... usually your cells tank... we run over it and we say, "Hey, here's what predicted values of of uh conversions are and here's how we think you should modify it to increase your conversions."

Sean: I think the surprise to me was that you guys are actively using it already in internally inside of Shopify...

Mikhail Parakhin: ...liquid neural networks are you can think of them as a next step like uh sort of uh state space model square... It's sub quadratic in in length of your context. Uh it's very compact way to represent things... we run it at 30 milliseconds a a tiny model like 300 million parameters in but we run it in 30 milliseconds end to end for search.

Mikhail Parakhin: There's different end of the spectrum where this is maximum through uh bandwidth throughput for things like for example offline categorization... We would take a large model like we would take something huge... we would distill into liquid for a specific task... it's definitely for this scenario of smaller models and distilling into they're second to none.