The Cost Crisis Facing CTOs and How to Solve It with Albert Strasheim, CTO at Rippling

(Intro Narrator at 00:00:00) Today, we're talking to Albert Strasheim, CTO at Rippling, about how he's been tackling cost savings in the age of AI and more. You're listening to Joel Beasley, Modern CTO.

(Joel Beasley at 00:00:17) And when I got presented with this idea of talking about the cost crisis facing CTOs, I said, "Okay, this looks pretty interesting," but I was hoping you could get me caught up on what this cost crisis is.

(Albert Strasheim at 00:00:29) Yeah, I mean, I think probably the big question right now, and we face it at Rippling as well, is just how much should you be spending on AI? And I think more broadly, there's just the ongoing question of how do you manage infrastructure spend? How do you manage spend on SaaS tools generally? But I think the introduction of all of these AI tools and models, both for product development and for internal use—it's just a lot of money. A lot of money is moving. I imagine many CTOs like myself are being asked to keep tabs on all of that. And so just knowing where is the money going? Are we spending it wisely? Do we need to stop spending in certain areas or start spending in others? That old thing is definitely a challenge or a crisis right now.

(Joel Beasley at 00:01:23) And can you remind me of what Rippling does exactly?

(Albert Strasheim at 00:01:26) Yeah, so Rippling, if you're not familiar, we're essentially an all-in-one HR, IT, finance data platform—a general business software platform for running any company from a couple of employees up to many thousands, even stretching into the tens of thousands right now. So imagine payroll, all things related to talent acquisition. So recruiting, performance management, internal surveys—anything you possibly need to run a business. We do that. Full-time employees, contractors, employees based in the U.S., employees that you want to hire internationally, companies headquartered in the U.S., companies headquartered internationally. We do all of that in one platform.

(Joel Beasley at 00:02:20) Okay, see, so I do understand what you guys do. So that's where my mind was at first. It was HR and payroll. But then you're talking about cost reductions for CTOs, and I was trying to figure how do we connect those dots? Is it that you guys grew so much? Have you improved the product and grown what you've done over the past decade?

(Albert Strasheim at 00:02:39) Yeah, absolutely. So there's a few pieces to this. I think beyond HR and payroll, we also have an IT suite. And so as part of that, we offer single sign-on capabilities and access management capabilities. And so we end up seeing a lot of what our customers log into at the end of the day. And as you can imagine, Rippling also uses Rippling to manage all of its internal SaaS tools and granting access to those tools. So that's kind of the internal cost management angle. It's just what tools do you have, who has licenses or seats, who can log into them, et cetera. And then, as you said, the platform is also expanding pretty rapidly. We are behind the scenes a big data and compute platform. And so managing the AWS and Google Cloud bill and Anthropic bill and OpenAI bill underneath this thing is also its own big engineering challenge. So there's kind of Rippling with 5,000 employees using AI in a bunch of areas, managing that internal tooling thing and the infrastructure thing—both challenges.

(Joel Beasley at 00:03:51) Well, it makes sense too because I used to think of the user access and permissions and those types of things over here away from the HR system. And maybe there was an API where we could get some information over so we could keep it. But now they seem to be all kind of converging into each other.

(Albert Strasheim at 00:04:11) Yeah, absolutely. Yeah, it actually ties into how I ended up at Rippling. And so my background, prior to Rippling, I was at Segment, a company that was acquired by Twilio. We built a customer data platform. And Cloudflare before that. But anyway, at Segment, essentially post-acquisition by Twilio, we had to integrate a number of services between the Segment side and the Twilio side, including Google Workspace, Okta, and then we had to move from BambooHR onto Workday. And it took us about eighteen months to complete that migration of about a thousand employees. And at the end of that whole thing, I was like, "Man, this was kind of hard, and the end result still feels quite fragmented." Right? We can't actually use Workday as a system of record to manage any kind of access in Okta. You end up with these multiple models of your organization in multiple tools. It never quite flows. Nobody gets the right access on day one when they join the company or the team. And so I was looking around thinking, "Hey, maybe I should start a company to build something that solves this problem." And as luck would have it, I found Parker, our CEO, through some Rippling investors over the course of that line of thinking. And I was like, "Oh, man, there's already a whole company doing this. I should definitely join them on their mission." And so here I am. It was exactly that kind of frustration with fragmented tools and painful integrations that brought me here.

(Joel Beasley at 00:05:51) Yeah, and it totally makes sense as far as you would want your system to be connected with payroll.

(Albert Strasheim at 00:05:59) Yes.

(Joel Beasley at 00:06:00) Right? Because that's the truth. That's the most source of truth is who's actually employed right now.

(Albert Strasheim at 00:06:05) That is exactly right. Yeah, there's huge incentives that push people to make sure that the payroll system is up to date with what's happening at the company people-wise day to day. Otherwise, people don't get paid appropriately. And so basically leveraging the fact that the payroll system of record has to be kept up-to-the-minute accurate, you can actually build a lot of other useful business software on the back of that. Yeah, so payroll plus HR is kind of the entry point into a lot of other products just working better because they force you to have this comprehensive system of record at the center.

(Joel Beasley at 00:06:45) It's interesting too. So the most employees that my company has ever had was about 20 full-time people. And at that point, it was even hard managing the onboarding and the offboarding and the amount of money we were spending on unused licenses, whether it was from tools like Calendly or our sales seats and all of that. Are you guys managing all seats now in third-party tools? How does that work?

(Albert Strasheim at 00:07:11) Yeah, so there's a lot of that management that you can do through Rippling today. Some more advanced license management stuff is coming as well. We do some things with the Microsoft suite of applications already. But you can also build a lot of management functionality yourself. And so, for example, what we've done for ChatGPT is actually built using a couple of capabilities. One's called Custom Objects and another capability is called Super Groups. Essentially, with those two things together, we are able to build an approval workflow in Rippling so that anybody that needs ChatGPT access can request it. There's a very quick approval process. You can get a very fast approval from your manager. And then anybody that needs ChatGPT access can get it quickly. But we built that in response to watching or looking at usage data inside of the OpenAI platform and realizing, "Hey, we're handing out a lot of ChatGPT licenses, but not everybody is using it." Right? Some people use Gemini, or some people will go directly to Anthropic's, sometimes using Claude or Claude in a more general-purpose way. And so you actually end up with, whether it's ChatGPT or other tools, a lot of unused seat spend every month. And one step towards managing that is just getting people to request access in the first place instead of handing it out to entire teams. And then you can also see very clearly who's using it, why did they request it, periodically expire the access, that kind of thing. And so it really helps to have a platform that allows you to automate that kind of workflow when you're trying to manage costs.

(Joel Beasley at 00:09:06) Oh, absolutely. I like the periodic expiration. I know some people might be hesitant to that, but I love it because it forces you to ask yourself, "Am I still using this?"

(Albert Strasheim at 00:09:17) Yeah, absolutely. And I think the key thing, right, is just as long as, if the access is expired and you can get it back within a few minutes and you do it maybe once a quarter or so, it's not a big overhead. It's a great way to get people to reassert, "Hey, I am in fact using this tool." It depends too. Sometimes with usage data, you can see that more easily and you don't have to leverage the expiration approach. But with a lot of the tools, they're so new, they don't actually even publish good usage data yet, or it can be hard to consume or automate around the usage data. And so expiration done periodically is another management technique that works for some of the tools.

(Joel Beasley at 00:10:05) So you have this blog post that you've written about the $10 million engineering problem. Can you tell me about that?

(Albert Strasheim at 00:10:12) Yeah, so this was definitely a team effort back in our Segment days. And so I think it was roughly 2018, 2019, before the company got acquired by Twilio. And the kind of intro here was, we were looking very closely at our gross margins as a business. And Segment, as I mentioned, pretty infrastructure-intensive customer data platform. We were looking at our margins and they were definitely going in the wrong direction. And one of the key contributors was spend on AWS and Google Cloud. And so we went through this very intensive six-month optimization effort—engineering, product, the finance team, folks from across the business all banded together to fix these various issues. And so we came up with the cheeky title of the $10 Million Engineering Problem. It turns out on Hacker News, people love clicking on that kind of thing. And so the blog post was essentially documenting our whole cost savings effort and everything we learned along the way.

(Joel Beasley at 00:11:28) And so the TL;DR is you reduced infrastructure cost by 30% and grew traffic by 25%.

(Albert Strasheim at 00:11:36) Yeah, exactly. So over that six-month period or so, infrastructure cost came down quite a bit while the business kept growing, which I think was a great result.

(Joel Beasley at 00:11:45) What was the biggest lever in reducing the infrastructure cost?

(Albert Strasheim at 00:11:49) I think the interesting thing there is there was probably never one big thing. I think what you find when you go and lift the covers on an AWS bill or a Google Cloud bill is money is spent on lots of interesting services. Especially if you have a large engineering team of a couple hundred people or more, and operating a complex infrastructure, money goes in lots of interesting directions. For us, it was things like inter-availability zone network traffic—very expensive, but at the same time a pretty high-margin business for the cloud providers. So they're not working super hard to help you optimize that. Spending on things like CloudWatch, which is the observability product built inside of AWS. Spend on things like the CDN product, et cetera. There's lots of these things that add up and through just a bit of attention and optimization, you generally find you can easily shave 10, 20, 30%, sometimes more of these products. Or ask the team, "Why are we even using this thing?" And sometimes that yields useful answers as well. But it's really ultimately kind of a holistic analysis that you have to start with, and then lots of insights get derived from there.

(Joel Beasley at 00:13:20) So you just paid attention to how you were spending money, and then you were smarter about how you spent your money.

(Albert Strasheim at 00:13:27) That is basically what it comes down to.

(Joel Beasley at 00:13:30) That's not as clickable for Hacker News.

(Albert Strasheim at 00:13:32) Yeah, yeah, yeah. Be smarter, waste less. Less clickable. I do think there's maybe a few other things we learned in the process too. I mean, one thing I do religiously now, whether it's an infrastructure cost management project like this or any other similar initiative—you really need the daily dashboard or the daily email or the daily report in Slack to help ground the team in what's going on. And especially if you're trying to do this kind of big optimization effort across a big team with many stakeholders, this kind of single pane of glass or command center view of what's going on was really important. And it takes a while for a team to build muscles around this kind of optimization or efficiency improvement. And so getting everybody into this rhythm of, "Hey, every morning we're going to wake up, we're going to grab our coffee, and we're going to look at the dashboard and drill down and ask questions about things that don't look quite right"—that takes a while to establish. And I think that was another key thing we did early on with that effort was just get everybody used to looking at a daily dashboard. I mean, a lot of stuff actually falls into place from there.

(Joel Beasley at 00:14:54) And that came—we made a change like that similar at our company this year. Actually, we deliver all these episodes, and so we started using ClickUp but using it correctly, with actually looking at the dashboards, reviewing them. That was a hard thing to get us into the habit of doing. How did you guys do it? Was it just difficult and you just woke up every day and you're like, "This is what we're doing"?

(Albert Strasheim at 00:15:16) I think it's difficult. I think it's an opportunity for good old-fashioned leadership. Right? You have to put the daily stand-up on the calendar. You have to have the meeting. You have to be annoying if everybody doesn't show up to the meeting. You have to pull up the dashboard. You have to look at it. You have to ask the hard questions. And I think what I found at Segment and Rippling too is after a little while, you start finding these cost reduction champions or just generally for anything you're trying to optimize these optimization champions. I think some people are just wired to—their parents taught them to be frugal.

(Albert Strasheim at 00:16:01) They love saving money. It's the kind of mission or quest they really get excited about or attached to. And so, you know, you kind of have to find those people, and then you just have to pull the group together on a daily basis. And then you can go pretty far if you have the right visibility into the problem. But just starting with that meeting and getting a couple of key people looking at the numbers is really important.

(Joel Beasley at 00:16:25) Yeah. I found frugal people are great with one asterisk. With one asterisk.

(Albert Strasheim at 00:16:29) Mhmm.

(Joel Beasley at 00:16:29) They have to really understand time value of money. Yes. It's like, it's okay to be cheap, but don't try to save $5 here and actually create a time suck of like $50,000 over here.

(Albert Strasheim at 00:16:41) Yeah. Yeah. You know? Absolutely agree. Yeah.

(Albert Strasheim at 00:16:43) I think, you know, there is definitely — and, you know, we had it in the Segment optimization project as well — you know, there's definitely going to be some things that are below the line. I think the other thing that was very helpful for us, you know, both at Segment and we've actually had a meeting about this today at Rippling, is to have the team understand the broader business context. Right? So explaining to them, you know, what is gross margin and, you know, it might come as a surprise, but like, you know, the average software engineer doesn't always know what gross margin is exactly.

(Albert Strasheim at 00:17:18) You know, explaining that to them, explaining the business goals, you know, "Hey, we're trying to take our gross margins from X percent to Y percent. You know, this is how it relates to our valuation as a company. You know, this is how it impacts the value of your shares in the future." Like, building that whole bigger picture. And I think to your point, you know, also explaining this time component. You know, time is expensive too. I mean, like, you know, once you help the team understand the whole framework around making these decisions, you know, things flow a lot better. And, you know, nobody loves just like looking at a spreadsheet with a bunch of dollars on it with no context, being told to make it go down. Like, the business context is really important.

(Joel Beasley at 00:18:06) It is. And actually, very impressed. You stated that very well. Most — it takes a long time for people to figure out that you have to connect the business outcome you're trying to achieve with how it directly impacts you as an individual.

(Albert Strasheim at 00:18:18) Yep. Yep. Yeah. Connecting those dots, I think, are so important. You know, like, I think aligned incentives are a really powerful force in the world.

(Joel Beasley at 00:18:29) They are the most powerful force.

(Albert Strasheim at 00:18:30) Yes. This book I read long ago called Freakonomics has a lot of thoughts on aligning incentives or what incentives do to human behavior. So, you know, I apply that when it comes to infrastructure cost as well.

(Joel Beasley at 00:18:46) I did see that they had a book, but I think I found them through like YouTube videos or something like that or a podcast or something. Yeah.

(Albert Strasheim at 00:18:54) Yeah. But, uh, yeah. Yeah. I — the book was, um, I remember it. I think it had like an apple with like a green —

(Albert Strasheim at 00:19:00) Yeah. Yeah. Yeah. Orange. Orange and green.

(Albert Strasheim at 00:19:02) Yeah. Yeah. I don't know when I read that book, but it might have been before the era of YouTube, which might be dating me slightly. Um, but, yeah, that was like the OG —

(Joel Beasley at 00:19:14) In the interest of the Ararat by an economist. Yeah.

(Albert Strasheim at 00:19:17) Yeah. Great book. Yeah. Life is good. Mhmm.

(Joel Beasley at 00:19:21) Uh, what was the most surprising place that you found cost savings?

(Albert Strasheim at 00:19:26) Um, I mean, probably — I mean, there's network optimization. There's like, you know, inter — what AWS calls availability zones. So like different, you know, sub-regions inside of the same bigger region. Like, there's a lot of network transfer there that is actually quite expensive. And depending on how you configure your infrastructure, you know, that can really add up.

(Albert Strasheim at 00:19:53) Uh, so that was probably one. Um, I think another one was actually from the CloudWatch metrics system. And the fix there was actually to go into Datadog and configure it to pull the CloudWatch system less. And then, you know, we also had some services send less metrics. But there's, you know, funny little things like, "Hey, do we need to pull this service every minute, or can we pull it every five minutes or every 10 minutes?" You know, if you can find that kind of slider, suddenly something, you know, is 90% cheaper. Um, and so, you know, these were the little bits of analysis where, you know, things that you think should be approximately free actually add up when you do them at the high volume or high frequency.

(Joel Beasley at 00:20:41) Have you ever seen startup CTOs be kind of shy about delaying cost management, or are most people like pretty into it?

(Albert Strasheim at 00:20:49) I think probably if you have seen it done before, you jump in much more proactively the second time. And so I think having been through the Segment journey where, you know, the — I forget exactly when Segment was founded. But, you know, the company was like four, five years in and had managed to make it, you know, pretty far without having to pay super close attention to infrastructure costs. Like, but, you know, and then we went through this like intense six-month period where we got really disciplined about it. I think having seen that, you know, literally the day I landed at Rippling, you know, we procured a cloud cost management tool called Vantage. It was one of the first things I rolled out after starting. And so I think once you've seen it done well once before and once you've experienced the pain of like being behind on your homework here, I think people are a lot more proactive. Um, and what I'll say too is I think as a, you know, early-stage company, it can be hard to understand, like, "Hey, when should I prioritize this kind of work? You know, hey, should I build my product for another week or worry about what it costs?" Um, you know, my general advice there is like, you know, the best time to start is like, you know, six months ago.

(Albert Strasheim at 00:22:14) The second best time is probably today. It's actually much better, uh, or much easier to get organized early on. You know, find a good tool. Uh, we love Vantage as an example. Implement it. Like, the configuration and setup takes, you know, an hour or two.

(Albert Strasheim at 00:22:32) And then, you know, all it needs for a while is probably like weekly care and feeding just to look at the reports, you know, do a bit of automation, et cetera. Like, retrofitting that stuff when your internal SaaS tooling and, you know, infrastructure has been sprawling for a couple of years can be really painful. Um, and one thing that comes up to quite often is like backing out the rules for how you want to reallocate costs from a big bucket, like, "Hey, this is all of our AWS spend" to, "Hey, this is the spend being incurred by team A or product A, and this is the spend for, you know, product B." Like, those kinds of rules take a while to develop. It's always harder to do later.

(Albert Strasheim at 00:23:17) You know, you can frequently make short-sighted decisions in how you structure your cloud accounts or, you know, where you run the services exactly. People forever will forget to tag things appropriately. A lot of that stuff is very hard to clean, you know, a few years down the line. And so, you know, I think based on my experience, I'm a fan of starting early. Start small, just get something going, and then, you know, iterate on it as with everything else. But I think deferring that homework is a recipe for pain ultimately.

(Joel Beasley at 00:23:48) Yeah. Sometimes they need the pain to make better decisions.

(Albert Strasheim at 00:23:53) Yeah. At your next company, you know.

(Joel Beasley at 00:23:54) At your next — yeah. So how many years have you been at Rippling?

(Albert Strasheim at 00:23:59) Uh, coming up on about three and a half. So I joined back in August 2022. Uh, yeah, we were about 500 engineers back then, uh, more than a thousand now. So, uh, it's been a wild ride.

(Joel Beasley at 00:24:11) How was that on a scale of one to ten, ten being the most disciplined you've ever seen in engineering team? How disciplined was the team when you joined?

(Albert Strasheim at 00:24:20) I think from a like execution, you know, just like getting things done perspective, like, really impressive. But, you know, the company at the time, and I mean, it's still true today, it is both a large company and also this amalgamation or, you know, aggregation of a bunch of smaller startups. And so although it was a 500-person engineering team, it was actually much more like, you know, 10-ish, like, 50-person startups, uh, you know, each charging ahead. And so, you know, although we executed very quickly, both very quickly, like, the challenge was putting in some like over-arcing operating systems, uh, or processes, uh, for the team. And so, you know, cost management was actually a good example of this.

(Albert Strasheim at 00:25:08) You know, each of these, you know, ten, 50-person startups hadn't yet, you know, gotten to the point where they needed to worry about infrastructure costs. But in aggregate, it was becoming a real issue. And so that was the kind of, you know, discipline or, uh, or framework type stuff that, you know, I focused on in my first year or so at the company.

(Joel Beasley at 00:25:29) Interesting. You know, discipline has come up a lot in this conversation. Uh, I'm a huge fan of the topic. I'm curious to know when did you first start learning about this? Is this early in your career?

(Joel Beasley at 00:25:41) I mean, I don't think you were born disciplined and like perfect. I think it's something most people learn.

(Albert Strasheim at 00:25:46) Uh, yeah. I think I probably through my early engineering career, I came to appreciate just the value of being organized. Right? Like, the — and I think as you mentioned earlier, like, you can over-rotate on this kind of thing. But like the very simple, you know, stand-up meeting in the morning or the very simple, you know, async, like, Slack conversation about what we're trying to get done on a given day or the quick meeting to review the dashboard, like, you know, just a little bit of coordination like that, I think, can go a long way and can really help you.

(Albert Strasheim at 00:26:25) I mean, for me, initially, as an individual contributor and then later as a manager and, uh, you know, and so on. Like, I think it helps people feel connected. It clarifies what needs to get done. You know? And so I think sometimes in the name of moving fast, people neglect the tiniest bits of coordination, and I actually think it's, you know, to your detriment. You know?

(Albert Strasheim at 00:26:46) The answer is not, you know, hours and hours of waterfall process and, you know, multi-hour sprint retros. I think you can move really fast, but you can move fast with just a little bit of documentation, a little bit of coordination, a little bit of structure. So feels like I learned it early on. I remember one of the first things I did as an intern at my first company was set up an issue tracker. So and I think I also configured the build system because I was like, "Man, we're all building, you know, on our laptops, but the builds never work when someone else tries to build.

(Albert Strasheim at 00:27:20) Like, let's set up a build server." Yeah. It's, uh, for whatever reason, I don't know what it was about my childhood, but I gravitate towards, uh, a bit of that organizing. And, you know, I think it really saved a bunch of time in the end.

(Joel Beasley at 00:27:33) You probably had a pretty chaotic childhood.

(Albert Strasheim at 00:27:36) It was quite calm, actually. I —

(Joel Beasley at 00:27:38) Was it?

(Albert Strasheim at 00:27:39) I like to tell this story where, you know, I would sometimes be sitting in my room at night, uh, looking at my very neatly organized bookshelf, and I would be so bored that I would unpack all of the books onto my desk and reorganize them. Um, maybe, you know, Cape Town was not a very exciting place or something. But I really enjoyed the process of bringing order to chaos. Um, you know, the battle against entropy is what, you know, keeps me going, I guess.

(Joel Beasley at 00:28:08) There's something in the water down there. Musk came out of there. You've come out of there. Yeah.

(Albert Strasheim at 00:28:13) He — I mean, you know, Elon is certainly an interesting character, but he is definitely doing, you know, uh, battle with entropy on a daily basis, uh, with remarkable results. So, yeah, maybe we just don't like entropy south of the equator.

(Joel Beasley at 00:28:30) How long have you been stateside?

(Albert Strasheim at 00:28:32) Uh, moved to ACEF in 2013. So, uh, I guess, like, twelve, you know, twelve years and a bit. Uh, yeah. Straight to San Francisco. I honestly had no idea what Silicon Valley was when I showed up here.

(Albert Strasheim at 00:28:45) And, um, uh, it was a pleasant surprise. So, yeah, I'm really enjoying it.

(Joel Beasley at 00:28:51) All those companies that you mentioned that you worked for, were any of those based in South Africa? Were they all once you moved here to San Francisco?

(Albert Strasheim at 00:28:58) No. So I moved for Cloudflare back in 2013. They were —

(Joel Beasley at 00:29:03) By the way, I'm a fan.

(Albert Strasheim at 00:29:04) Oh, yeah. Yeah. So, uh, really excited about what they're building over there. It really feels like the product is going from strength to strength. But, yeah, I moved.

(Albert Strasheim at 00:29:11) Um, they were about 50 people at the time. Um, so, like, really small and, uh, you know, very explicitly were recruiting internationally. Um, so, uh, Michelle, one of the founders, uh, with Matthew, you know, it was part of their strategy was to basically say, "Hey, we're going to recruit from anywhere. We will work through all of the visa hassles and immigration hassles, uh, to get people." And I think it really enabled them to build a very strong team.

(Albert Strasheim at 00:29:40) Uh, it's probably even harder to do that these days, uh, with everything going on, but, uh, you know, I really think there's amazing talent, uh, internationally.

(Joel Beasley at 00:29:50) Well, how could you not have amazing talent with the lava lamps?

(Albert Strasheim at 00:29:55) Yeah. The talent love the lava lamps.

(Joel Beasley at 00:29:57) Can — can you just for like two seconds, can you just tell people who don't understand the Cloudflare lava lamps what they are?

(Albert Strasheim at 00:30:03) Um, so I actually don't remember all of the details here. But essentially, in the Cloudflare — well, Cloudflare had an office, and I think, uh, John Graham-Cumming, who was the CTO at the time, and maybe some other folks, I think Nick Sullivan on the security team, they had this idea for like, "Hey, we need some kind of big installation in the office." And so they built this, uh, wall of lava lamps. And then I believe there's like — I don't know if this is still true today. Uh, Dane, who's the CTO now, might know.

(Albert Strasheim at 00:30:37) But, uh, they essentially take a video recording of the lava lamps and, you know, the lava lamps are essentially generating entropy, and they feed that into some random number generator somewhere at the core of the network. And so, you know, it is both just a very cool installation to look at, but then it has this, you know, technical angle where it's actually generating entropy for a random number generator somewhere. So pretty cool installation. I think you can basically walk past, uh, that building and see it through the window if you're in San Francisco.

(Joel Beasley at 00:31:11) But that's just got to be the fake one so no one can grab it. They actually have a — they have a hidden — a hidden lava lamp.

(Albert Strasheim at 00:31:17) Second. Yeah.

(Joel Beasley at 00:31:18) Yeah. Where they're actually using that one. Yeah.

(Albert Strasheim at 00:31:21) We should, uh, you should get Dane on, uh, on here and ask him what's happening with the lava lamps these days.

(Joel Beasley at 00:31:27) Is Dane the current CTO?

(Albert Strasheim at 00:31:29) Yeah. Yeah. Dane took, uh, took over from, uh, John Graham-Cumming or JGC, uh, recently. So yeah.

(Joel Beasley at 00:31:36) Nice. Are you still in San Francisco?

(Albert Strasheim at 00:31:37) Uh, yeah. In, uh, kind of like North Beach, Russian Hill area right now. Uh, actually moving to the Inner Sunset soon. So excited about that.

(Joel Beasley at 00:31:46) Oh, is that closer to work?

(Albert Strasheim at 00:31:48) It's actually further from work, but it has a great view of the sunset, as the name suggests. So looking forward to that.

(Joel Beasley at 00:31:56) Nice. Nice. You strike me as a guy who's not at home during sunset time. You're like late at the office. No?

(Albert Strasheim at 00:32:04) Yeah. There's probably sunsets on Saturdays and Sundays and maybe during the summer. You know?

(Joel Beasley at 00:32:11) Okay.

(Albert Strasheim at 00:32:12) Yeah. Looking forward to it.

(Joel Beasley at 00:32:15) Alright. So this is one of the first recordings of twenty twenty-six. I'm excited. Are you looking forward to any big developments in tech this year? And if so, what are they?

(Albert Strasheim at 00:32:28) Yeah. I mean, I think it's going to be another big year for AI. I mean, I don't know if you heard of AI, but it's all the rage right now. So that's coming. I think probably the thing, two things I'm most interested in right now. Maybe three things. I think one is world models. And so I don't know very much about world models as I sit here, but just for entertainment and gaming purposes, this ability for the AI to generate a 3D world that you can navigate seems like a very compelling capability in its own right. You know, I can just imagine the games we're going to be able to play from a prompt. So I'm avidly consuming world model content on Twitter right now. And then I think it ties into the other part of the equation, which is robotics. You know, I think there's a lot of cool stuff coming there. And robots training inside of simulations feels pretty critical, you know, as I look ahead. Really cool to see what's coming there. I was looking at the CES announcements as well. I think LG announced this laundry folding robot. It was interesting. Everyone's commenting like, "Man, this thing is so slow." It is the slowest it's ever going to be. You know, it is going to be faster next year and much faster the year after that. So just like that whole field of robots, world models, robots training inside of world models, you know, physical AI, I'm really excited about. I don't know if we're going to give birth to Skynet here. That's probably its own problem. But yeah, that's the one thing I'm really excited about.

(Joel Beasley at 00:34:25) We got Elon and Bezos putting satellites up every day. Once all the satellites are deployed, then we'll get some more Skynet.

(Albert Strasheim at 00:34:32) Yeah. Yeah. So that's—

(Joel Beasley at 00:34:33) I like the world models. Yeah. I like the world models. I think that's fascinating because you've been seeing hints of that since before AI went mainstream and was going to be the one. You know, we were even looking at that just standard programming before the AI models came about.

(Albert Strasheim at 00:34:50) Yeah. Absolutely.

(Joel Beasley at 00:34:51) As being a possibility. I really think—

(Albert Strasheim at 00:34:54) You know, if you kind of think about what is the Netflix for world models. Right? It's almost like, I don't know where it all ends. It is basically the holodeck from Star Trek, or it is Westworld, you know, coming to life to some extent.

(Joel Beasley at 00:35:10) Hopefully.

(Albert Strasheim at 00:35:11) Yeah. Pretty, yeah, pretty exciting. I mean, also lots of interesting, you know, societal implications. I don't know if anyone is ever going to leave their couch again when this stuff gets good.

(Joel Beasley at 00:35:21) We just plug in with the IV, man.

(Albert Strasheim at 00:35:23) Yeah. Yeah. So I mean, I think that's one. And then the other thing that seems to be gaining a bunch of traction right now is essentially multi-agent orchestration in software engineering. It feels like we went from one good agent writing code with some thinking models and some human input to that process last year to this multi-agent orchestration thing really working well now or starting to work well. And I think people are also learning how to build these kinds of platforms that have agents supervising agents, you know, these sandboxes where the agents can safely work with code without deleting everything on your laptop. Like, you know, it's almost like world models for training robotics. There's like sandboxes for coding agents. Those things seem to be going together. And so I'm very interested to see where that all goes. Steve Yegge, who's at Sourcegraph, or I guess Amp these days, potentially, he wrote a great blog post in the last couple of days about this agent orchestration framework he built called Gastown. If you haven't read about it, the post is titled "Welcome to Gastown," if I remember correctly. Really interesting. He basically documented his entire journey on building this multi-agent orchestration system that they're using to generate some ideas for their product. So really cool stuff. And I think it's going to completely transform how software engineering works in the next couple of quarters.

(Joel Beasley at 00:37:11) Nice. I'll add that to my reading list here. "Welcome to Gastown." I found it on Medium here.

(Albert Strasheim at 00:37:17) Yep. Yep. Yeah.

(Joel Beasley at 00:37:19) Have you been reading him for a while?

(Albert Strasheim at 00:37:21) I've definitely followed his career. He has a few seminal posts from his time at Google and Amazon. And so, you know, I keep an eye on what he writes. It's very entertaining and also thought provoking. So—

(Joel Beasley at 00:37:39) The LG folding laundry robot, you think we're there? Are we there yet? Like, if I bought one, would it fold my laundry?

(Albert Strasheim at 00:37:48) I mean, I didn't carefully scrutinize the demo, and I do not own an LG laundry folding robot or stock in LG yet. But it seems to be working. I mean, if you look at other companies like Boston Dynamics, and I mean, there's a whole bunch of startups in this space as well, it feels like the capabilities of the models and the tasks at hand, you know, they're reasonably well matched. And so I feel like we're making pretty good progress. You know, the moment where these systems can start learning semi-autonomously feels pretty close. My own, you know, maybe many years from now after Rippling, my one dream is to build a robot that picks up trash on the streets of San Francisco. Might be a very expensive way to solve that problem, but I think that would be great to wake up to clean streets every morning. So—

(Joel Beasley at 00:38:49) That's not a hard thing to do. My town has clean streets every morning. So yeah.

(Albert Strasheim at 00:38:55) Yeah. I think, you know, here the scale of the problem is a little bigger. And I was thinking about it too. Right? It's like, you know, if you're trying to build a robot that deals with dry trash, wet trash, there's actually a lot of modes of trash collecting to deal with. So it is actually an interesting problem. And then I imagine people are going to try to vandalize the trash collecting robots, so they will need to deal with that as well.

(Joel Beasley at 00:39:24) Just don't paint them. Just let the vandals decorate them. It looks like it's intended.

(Albert Strasheim at 00:39:29) Yep. You know? Yeah. Absolutely. So but anyway, that whole thing feels primed for a very exciting year ahead.

(Joel Beasley at 00:39:37) Yeah. I'm excited. I want the laundry folding robots. I saw Elon's Optimus doing some laundry folding. That got me pretty excited.

(Albert Strasheim at 00:39:46) Yep. Yep.

(Joel Beasley at 00:39:46) I saw Boston Dynamics a few years ago. Three years ago, dude, they were doing backflips with machine guns. And I'm like, can somebody just make it do laundry? That would be great.

(Albert Strasheim at 00:39:58) Yeah. No, I think they've evolved their platform a lot. Optimus, I think, is interesting because, you know, Tesla certainly has strong muscles when it comes to large-scale production of these robots. There's also a big difference between building small amounts of these versus building millions or tens of millions. And so I think it'll be interesting to see how, you know, Tesla, LG, and others ramp up production, you know, what kind of cost efficiencies they can achieve as they do that.

(Joel Beasley at 00:40:32) Yeah. Right now, I'm watching Optimus and Figure. Have you seen Figure?

(Albert Strasheim at 00:40:37) Yeah. I've seen Figure around too. Yeah. It looks pretty cool.

(Joel Beasley at 00:40:41) Yeah. Yeah. Now this multi-agent orchestration has definitely been on the rise the past six months, at least as far as my radar goes. Makes sense as far as the direction we're headed. It wasn't like a big surprise. It's just cool to see how things are getting to production a lot faster, it seems.

(Albert Strasheim at 00:40:59) Yes. Absolutely.

(Joel Beasley at 00:41:01) Do you have any multi-agent orchestrations going on at Rippling right now that you're using in production?

(Albert Strasheim at 00:41:06) So we're busy building essentially an agent framework that allows you as an administrator to interact with various parts of the product. And the goal here is to automate, you know, largely complex flows for admins. So folks that are hiring people, making changes to payroll, you know, doing complex reports. You know, a lot of that historically would have been done by some kind of HR analyst or systems analyst. Essentially, building that as a set of agents now to make it easier to use almost like the most complex parts of the product. And a lot of teams are currently integrating into that framework, and we're seeing some very good results. Yeah. So excited, you know, using it internally. Excited to roll that out to customers over the next little bit. And yeah, I think, again, looking back twelve months, the models probably weren't quite there yet. But I think, you know, since the summer, now last year, things have really come a long way. And so it feels finally possible to build good products for these kinds of admin personas that are doing pretty complex business workflows.

(Joel Beasley at 00:42:21) Yeah. One of the things that surprised me about this past year is it's called MCP. Right? Is that the model where you're hooking up your API essentially to be consumed by—okay. So I've been building software systems for twenty years, and it seems like every time there was a new emergent technology, it was almost resisted by the incumbents. I'll give you very specific examples. One would be banking. I was building real estate or, I'm sorry, real estate software. But I was building financial software fifteen years ago. And the banks were almost fighting this idea of letting you log in. And then eventually, they kind of acquiesced, and then they begrudgingly provided APIs. And then it got a little bit forward and it moved forward. And that, to me, sums up my experience and what I've seen happen multiple times through multiple different stages of technology. However, my experience currently, although it's small, when watching AI come about is people almost seem proactive and excited about it. Even the incumbent people, the people that have been in there, they seem to be like, "Okay. Let's embrace it." And to me, that feels good. It feels like we almost got over this technological hump. Do you—am I crazy? Do you feel that way? Have you seen something similar?

(Albert Strasheim at 00:43:43) I definitely think, you know, the market, there's a lot of collaboration right now. I think there's probably a couple of different forces. Right? I think, initially, you know, with OpenAI pretty far ahead, it made sense for everybody else to try and collaborate and embrace open standards to catch up. I think that's one thing. You know, different teams have built out these capabilities at different rates. And so, you know, I think there's going to be a bit of a hype cycle here too. And so, you know, to stay part of the conversation, you had to just announce anything with AI, and, you know, so much better if you could announce something that implicated OpenAI as well. So I think that's maybe the slightly pessimistic view. At the same time, I think it's great that people are experimenting and, you know, embracing open standards as we figure out how to use these models and these capabilities. On the other hand, I am curious to see, you know, what happens. Like, there's always this question of, like, do you move the compute to the data or the data to the compute? And I think right now, you have, you know, the models running in OpenAI's cloud or Anthropic's cloud, you know, pulling data from various systems through APIs. I think that works well for less data-intensive applications. But I think for more complex data analysis, you know, like truly dealing with large volumes of data, you might find that pushing a lot of the work that the models are doing closer to the data is going to be pretty important. And so, you know, I look at the rise of Databricks, for example. I think they're very much about bringing compute to the data. And so it's going to be interesting to see which use cases are addressed by moving data to compute and which use cases are addressed by moving compute to data. And so it feels like initially, it was all about moving data to compute because the compute was scarce and new and special. But it's starting to shift back where you can now run lots of different models in your own cloud or even on your laptop. And so I think the whole ecosystem is going to transform a couple more times.

(Joel Beasley at 00:46:06) Yeah. It reminds me, as far as the advancements in the compute, reminds me of when I was a child and I would drive between cities as, you know, my parents would drive me between cities and I'd see the billboards change every two weeks. Pentium 2, Pentium 3, Pentium 4. It's like every other week, there was this new advancement. And that's what it feels like today. The, you know, a 10 billion parameter model can now be the size of a—you know, they just keep making all these advancements and it's happening so fast. I can just imagine over the next year, things are just going to become cheaper, faster, better.

(Albert Strasheim at 00:46:41) Yeah. Absolutely. I mean, very interesting to see, you know, what Google is doing with their TPU silicon, you know, NVIDIA with their deal to roughly acquire Groq. I think all of those silicon level—

(Joel Beasley at 00:47:00) Did you see that? Do you know those guys, by the way?

(Albert Strasheim at 00:47:02) Not personally. Just, you know, long-time follower on Twitter. I definitely spotted Groq a while ago, and I was like, "This is going to be a big deal." TBD how they exit. So cool to see the success there.

(Joel Beasley at 00:47:18) Yeah. I interviewed Jonathan, the CEO and founder of Groq, like four or five years ago.

(Albert Strasheim at 00:47:26) Wow. Awesome.

(Joel Beasley at 00:47:26) Yeah. Yeah. Right when they were getting started. And boy, were they smart. Because he was the guy that designed the TPU over at—

(Albert Strasheim at 00:47:34) Yes. Yes. That's right.

(Joel Beasley at 00:47:35) And then he went out and did that. But man, he was smart. I had such a stupid suggestion for him because they were selling chips and sending them off at the time. Their model's changed quite a bit. They're doing more inference now on-site, but they were selling their chips. And I told him he should package them in like little Dorito bags, like little fun chip bags. Because we're having like a creative session on the podcast. It was a lot of fun.

(Albert Strasheim at 00:48:01) Nice.

(Joel Beasley at 00:48:01) Yeah. He's super smart people.

(Albert Strasheim at 00:48:03) Yeah.

(Joel Beasley at 00:48:03) But also fun. That's what stuck in my mind about Jonathan was he was so much fun to nerd out with as an engineer, but he was also super creative, like somebody I could make music with.

(Albert Strasheim at 00:48:13) That's awesome.

(Joel Beasley at 00:48:14) Yeah. Yeah. Yeah. I'm a fan of his. Yeah.

(Joel Beasley at 00:48:18) So I've got one last wrap-up question if that's okay. I'd like to ask all my guests here this. Looking across your career, what's the most useful piece of leadership advice? But here are the constraints: you've received it, you put it into practice, and it stuck with you for over a decade.

(Albert Strasheim at 00:48:37) Great. I think probably from two mentors, folks I've worked with, different versions of the same advice. And I think it basically comes down to, if you have feedback for someone, don't pull your punches. Give it to them very clearly and very directly.

(Albert Strasheim at 00:49:00) And Matt, who's our COO and CPO at Rippling, actually helped me understand this quite viscerally. What he said is, holding back on feedback is ultimately something you're probably doing because you feel unsure about how the feedback will be received. And that's actually a very selfish thing to do, right? You're optimizing for your own safety and security and comfort and not for the growth of someone else.

(Albert Strasheim at 00:49:33) And so just embracing the fact that you have some hard news or hard feedback to deliver and coming out with all of it—obviously thinking about how you deliver the message—but it's very important not to shy away from those difficult conversations and not be selfish. And so I think that whole "don't be selfish, out with the feedback" approach to management has been really valuable to embrace. And so I think if you're coming up in management, that's one of the key things to think about. A big part of the whole game there is don't shy away from giving hard feedback.

(Joel Beasley at 00:50:15) Nice. And if people want to learn more about Rippling, if they want some of this cost management, reduction, licensing, software-type stuff—that was well said, right?—where would they go?

(Albert Strasheim at 00:50:26) We write about it on Rippling's blog, so rippling.com/blog, if I remember correctly, and we'll certainly post about it on LinkedIn too. So you can follow Rippling on LinkedIn or follow me on LinkedIn, and yeah, we'll post there.

(Joel Beasley at 00:50:44) Awesome, man. We did it. Thank you so much for listening. And if you found this episode useful, please share it with a friend or colleague who you think would get value from it. And if you have topics that you'd like to hear discussed on the podcast, either add me on LinkedIn or send me an email: [email protected]. Every time I get an email or LinkedIn message, it absolutely makes my day and inspires me to keep going.