Daniel Whitenack: Welcome to another episode of the Practical AI Podcast. This is Daniel Whitenack. I am founder and CEO at PredictionGuard, and this is a pretty special and fun episode for me, because I get to kick back with an old friend of mine.

We went to the same university, for those that haven’t heard of it, Colorado School of Mines, in Golden, Colorado. Of course, a shout-out to all the ore diggers out there that are listening… But yeah, we have with us today, Bengsoon Chuah, who is a data scientist now, and working in the energy sector. I was really fascinated to talk over the years with Bengsoon about all the things he’s doing, and in particular his approach and learnings around active learning and NLP models… And I wanted to invite him on the show to talk through some of that and learn a little bit from him. Welcome to the show. How are you doing?

Drew Wilson: Yeah. Clerk.com/billing is where you’ll go for that page… And what’s really cool about Clerk is that it makes authentication super-easy. Kind of like Stripe did 15 years ago, or whenever they started. They made this pitch that “We want to own the credit card data at your company, and you’re no longer going to have ownership ownership of it. We’re going to own it, but we’re going to build some really cool software around it to make it really easy to use, and build in a bunch of fraud and safety and monitoring protections for you.” And they actually won that argument. People stopped using whatever they were using to reconcile credit cards on their own, and they gave all that data to Stripe. And Stripe built some really cool APIs and stuff around it to make it really easy, and actually easier to access than if you’d done it on your own.

And so Clerk’s doing the same thing with the user table… Which is for a lot of engineers a harder thing, but as we scale engineering into, and as it goes downstream into less and less technical folks, with AI now happening, they don’t really care. They are happy to give up their user table to Clerk, to say “You handle the user record, because you’re going to build this amazing software around it, these cool APIs, these cool UIs… You’re just going to make it where I don’t even have to think about this stuff.”

So Clerk’s done a really good job to date on that argument, and winning folks over, and showing how much better it can be to build your app on Clerk that not build your app on Clerk. The next major step to that user, from a company perspective, is like “I want to make money off this user.” And so that’s where this new product billing comes in. So it’s really the second product that Clerk will have ever offered off, and now they have billing.

[01:21:59.11] So everything that we’re building at Clerk in terms of billing is all native to Clerk. So we’re not actually using Stripe’s billing product. We’re just using Stripe for payments, and eventually PayPal for payments. And you’ll be able to pick between them. For right now though, all of the recurring billing logic, all of the invoices, the payment attempts, the emails, all that stuff is native and built into Clerk. And what that means is we can tailor these plans that you create, these subscription plans you’re going to create to SaaS, and we don’t have to account for every other type of business like Stripe has to. So we can get very narrow-focused to make the UI and the UX way better.

Then separately, there’s this idea of entitlements. Okay, someone signed up for a bronze plan, and this is a person who signed up for a gold plan. Well, what do they have access to? Well, usually, if you’re using Stripe or some other system, you’ve got to then build out these entitlements in your app internally, and you have to somehow reconcile them and keep them in sync with the plans that they’re signing up with. And you have all these if statements all over your code saying “If bronze plan, and it has this feature, and if that entitlement, do this, that, and the other thing. But if this…” etc. So it gets crazy. And so at Clerk, what we’re doing is we’re merging those all into one thing. So when someone signs up, or when you’re creating your plan, you can assign features to it and entitlements, and you can set the parameters on those, and you can have them be the same on gold, and on platinum, by just adding the features to both of them. Or you can slightly tweak that feature to have this number of credits, versus that one has this number of credits… You could do all that right from the UI inside of Clerk. And then in your codebase, you just use our has function [unintelligible 01:23:45.20] the name of the feature. And then it will automatically check to see who are they logged in as, and if you’re in an organization situation - because clerk also supports orgs - where you have roles, this has function will automatically look at the role this user is logged in as. Are they logged in as an admin, or a member? And maybe admins have the thousand credits and members only have the 50. And it will just automatically figure that out for you.

So then the entitlement scenario on actually building out your app becomes way easier than it ever would anywhere else. And I mean that sincerely, because that is the truth. So it ends up becoming just a way better experience as an engineer, and you can get to making money so much faster.

Kendall Miller: This is a whole new world, and the people who are missing out on the AI gold rush - because a lot of us are Luddites over in the corners, going “Oh, this is one step farther than I want to go. OpenStack was enough.” “We tempted fate when we trained the computers.”

Nick Nisi: This is more than nuggets gold here…

Jerod Santo: For sure. I think it’s a really good point. And to just to draw a gross analogy, or a basic – just to get back to basics… If we talk about a pickaxe, imagine somebody who spent years trying to dig gold with their hands, and then the pickaxe comes along. And it’s like “Hey guys, I’ve been digging gold with my hands, and now you get a pickaxe.”

Anurag Goel: So this is something that came out of Amazon, where back in the day Bezos or someone at Amazon was like “Look, team sizes should not exceed the number of people who can consume two pizzas.” And so if you have a team of 50 people, they’ll require way more than two pizzas. That’s not great. We want your team to be small enough, and we want their scope to be small enough. And the whole thing around service-oriented architecture - so everything is a service, and then all these teams build these well-defined interfaces that talk to each other, or that people have to go assemble themselves… I mean, in some ways, that’s also the Unix philosophy; you have tools that do one thing well, and then you find ways to configure them or use all of them together. So that is essentially how AWS - and not just AWS; I’m sure other parts of Amazon - that’s how they thought about organizational design.

[00:19:45.10] And it’s true that you ship your org chart, right? You’ve heard that one before, I’m sure. And when you think about it that way, that’s why AWS isn’t able to do an end-to-end application stack well. That’s why they falter. And it’s not something that they’re going to be able to change. But when you think about Render, we’re always thinking about stuff at the very top, at the very layer where we’re focused on what application developers want. We don’t actually build for DevOps engineers. We think they’re great, they’re valiant soldiers in this battle that they rage against complexity every day, but we live in a world, Render lives in a world where application engineers are responsible, and want to control their compute. So that’s a very different mindset and philosophy, and it requires thinking about products in a way that is fully integrated. Because as an app engineer, I don’t really want to both integrate Lambda, and API gateway, and something else, and then EC2, and then Kubernetes, and install all of this. No. I just want my app to be out there, to be resilient, to just work, and I want reliable, scalable compute.

So that’s how we think about it, and that leads fundamentally to very different products. And I think it’s pretty evident in how AWS has tried to build these products at the top level. We’ve all heard of Elastic Beanstalk, we’ve all heard of other attempts that AWS has made at building these higher-level products… And they fail. They fail – well, failure is in quotes, because at AWS’s scale, any product you launch will get some adoption, but it doesn’t become the gold standard for how to do it. And that’s because they always have these trapdoors. When you start out with something like EC2, then your incentives for making a high-level product aren’t as strong, because you can say “Well, if this product doesn’t work for you, you can always use EC2, and you still make money that way.” For Render, that’s not true. We’re not selling VMs. We’ll never sell VMs. Our platform is what we are selling. So if this doesn’t work for you, then we lose a sale. So we make it work for you. So that’s, again, a philosophical difference.

Jerod Santo: What up, nerds? I’m Jerod and this is Changelog News for the week of Monday, February 24th, 2025.

My fellow Severance fans must know that Apple TV has published eight hours of music for us to refine to. They call the ODESZA set, “perfect for your innie’s workday.” So good. So gold.

Ok, let’s get into the news.

Kris Brandow: Yeah… It’s sort of painful, I will say. I see that “Export to a Premiere” button or menu item in Audition, and I’m like “I should just be doing this…” But I’m on a Mac, and there’s a lot of really deep integration that Final Cut has with the hardware that I wanted to take advantage of if I could. And unfortunately, Apple does not have a similar level DAW, digital audio workstation, for podcast editing. Like, they have Logic, but that’s definitely more of a music-based DAW, that you can sort of squint at and turn it into like a spoken speech kind of thing. But it’s like, Audition is definitely the gold standard for this type of audio content. And so it’s just like “You know what? I’m just I’m just going to use Audition, and I’ll just deal with this weird flipping between Final Cut and Audition.” I think I’ve gotten it down after editing a few episodes now. It was kind of a mass jumping back and forth between the two… But I think I’ve figured it out now. Gotten it down.

Jerod Santo: Well, I do think that we’re very much in the early days, where the pioneers are pioneering, and that’s why so much excitement in our particular industry was like “Ooh, a new gold rush.” And as tech enthusiasts and people in the software world, of course, we’ve been interested in toying with it and trying it… But it certainly hasn’t arrived yet in any packaged way where it’s mainstream.

I mean, Apple Intelligence is probably the closest thing, and it just kind of sucks at this point. I mean, are you using any of –

Jerod Santo: Mat, when you took this open source award, this medal, this gold medal, did you mention all the little people that helped you along the way? All these contributors?

Adam Stacoviak: So gold. Yeah. Thank you, BMC, for those beats, and for just the remixes, and making this show a little more special. A little more special. Thank you.

Mitchell Hashimoto: There’s a lot of layers, and it is to some artistic, the way fonts look, and stuff. And there’s not a pixel-perfect specification that says that given this font and these characters, whatever, this is how it looks. It really is mathematical, with a lot of wiggle room in terms of how much anti-aliasing you use, the spacing, where it floats on the line… There’s a lot of wiggle room.

And so I think, in addition to sort of the core technical challenges of so many different phases of fonts, there’s also like people that just come up and say “It doesn’t look good. Nothing’s wrong, it’s just I don’t like the way it looks.” And then you kind of have to dig into why that is, and Ghostty provides a ton of knobs for those people to get fonts looking the way you want. And then our gold standard has basically been we want our fonts to look like Mac default. So like text edit, honestly, is our gold standard. The way text looks in text edit is the way it’s going to look in Ghostty by default.

[00:52:09.03] So yeah, that’s been that… But I think also just finding all the layers I never thought about. I think the first one that shocked me, that I had never considered as an engineer, was “How do you pick a default font?” You have a monospace application; in a web browser, you say “Font family monospace”, and the web browser picks it for you.

Adam Stacoviak: [00:31:54.29] And CentOS used to give it - I’m quoting, “used to give it” - prior to being acquired by Red Hat. Now it’s upstream from RHEL in terms of a visual diagram. It was acquired as an open source, acquired by. Now it is where the active development happens, which ultimately lands in RHEL the product. And so the angst there is the folks want what is enterprise-grade Linux, RHEL. You’re considered the standard, the gold standard of enterprise-grade Linux. They want it for free. That’s the angst.

Justin Garrison: And I’m really tempted to go and just buy the jacket, and just go to re:Invent and just wear it. Just, I have no certifications, but I was like “I’ve got a gold jacket.”

Adam Stacoviak: That’s good. I think you have – Keenan had good instincts as well, but it was like, everything he touched turned to gold. It was a Midas touch. So he couldn’t almost go the wrong way. His VR company, just to set the stage a little bit, eventually sold to Hooli for billions.

Adam Stacoviak: Even as I look at the screenshots, what I like about it is when you go to the GitHub repo, it has the link to the app store, download on the app store, which is - this is a Mac app. And as I’m going through the screenshots, it has similar characteristics. I mean, how much different can you make a chat app, really? So I’m not even knocking them necessarily, but like it has similar characteristics of what I like about Raycast AI chat. And Thomas, if you’re listening to this show, or anybody from your team’s listening to the show, you may be doing this already, but - gosh, strike the gold while it’s hot. Be the single application. And I guess maybe Enchanted might have a leg up, because they – what’s the license? AL v2? Apache license version 2? So…

Adam Stacoviak: I assume that, because any good hacker that writes a CLI that’s worth its weight in gold would always include a -n, right?

Jerod Santo: Harvesting, fishing, panning for gold

I like this set of metaphors for how to think about bucketing challenges into different categories. Some problems are like harvesting:

Harvesting problems have straightforward solutions and no shortcuts: You just get a big basket and pick every damn strawberry in the field. You solve these problems with pure perseverance, slogging away for weeks, months, or years until they are done.

Some problems are like fishing. You know that there are fish out there in the ocean, but you don’t know exactly where. If a great fisherman knows where the hungriest fish are and how to set their lines just right, they might catch everything that they need in a few hours. Fishing problems can sometimes be solved shockingly fast by motivated teams with a bit of luck.

Some problems are like panning for gold – going out to a river or stream where there might be gold, getting your pan out, and seeing if you can find traces of the shiny stuff in the sediment. If you find gold, you can become generationally successful – think of the massive moats created by Google Search or the AirBnB network.

If you can categorize the problem you’re trying to solve into one of these buckets, applicable strategies become more clear.

Autumn Nash: [00:40:11.01] With all of the multi-cloud and all that good stuff, and then AI, I think the people that are really going to make money in the next couple of years are going to be the people that make really good tools for multi-cloud, or things to lift and shift… Because that’s what people seem to be doing right now. And the chip companies. Because – you know when they say that like during the mining gold rush back in the day, the people that made the money were the ones that sold the shovels…

Shawn Wang: Maybe for the engineers listening, I would say the progress here has been at the model layer. So Claude Artifacts is built on top of 3.5 Sonnet, which is the current world best model… But also, there’s a significant amount of AI engineering that was required to build Devin and to build Artifacts. And I think that if you want to see what the future of AI engineering should look like, you should be trying to build a clone of this thing. That’s what I’m trying to do. Because I think a lot of AI engineering will look like this. It will look like “How do you wire up a model to the real world to produce projects of significant value, that you would otherwise have had to assign to a junior engineer?” I think that is absolutely the sort of gold trophy that people are going for right now. And obviously, the step beyond that is artificial general intelligence. But this is a pretty good second place.

Jerod Santo: What up, nerds? I’m Jerod and this is Changelog News for the week of Monday, June 10th, 2024.

The AI gold rush has Nvidia breaking profit records left and right. Turns out at their current market cap they’re now valued at $102M per employee. I think it’s time for some serious raises…

Adam Stacoviak: Round number eight. My definition. I mean, that was gold, right?

Amal Hussein: It’s interesting. Yeah, that’s interesting. It’s always really nice to see people exploring different avenues. I think there’s always gold that comes out of it irregardless. So that’s very, very cool. Thanks for sharing. So yeah, so Cory, what do you feel like is a big takeaway that you want to kind of share with folks around using Web Awesome, in terms of it bringing value into their stacks, and also just soothing over a lot of the common problems that we all have when designing web interfaces?