How Laravel Uses AI Across Finance, Legal, HR, and Sales

0:00 The Pragmatic AI Podcast is sponsored by Tighten. T-I-G-H-T-E-N. We will take your AI ideas, prototypes, or even vibe-coded apps, and we'll take them to production. Scalable and secure. Check us out at tighten.com. Hello and welcome back to Pragmatic AI, where we're working to bring a clear-eyed look at how AI interacts with the real world. And my guest today is my friend Tom Crary the president and chief operating officer at Laravel.

0:26 And if you've been listening to this podcast for at all you already know about Laravel but you don't know Tom necessarily. So Tom, would you mind saying hi to the people and tell them kind of what is your role at Laravel? What do you actually do day to day so we can kind dive into how AI actually interacts with that?

0:40 Absolutely. So let's see, Tom Crary President COO here at Laravel. I've been with Laravel since ~ January of twenty twenty four, so close to three years now. ~ and I'm ~ Taylor's business partner. So I do all of the stuff on the business side from you know, finance and accounting to HR to ~ sales and go to market.

0:58 ~ you know, a lot of company communications and strategy and things like that. So wide variety of stuff. Yeah. And for those who don't know, Taylor Otwell founded Laravel and it was a really kind of small team of under like eight ish people. And then there was a really big inflection moment where he decided to partner together with Accel a PE firm, and grow the company significantly and build new products.

1:20 And Tom was like employee one core of that new world. And so one of the first things that happened in those early days was Tom comes on as president. Taylor becomes CEO and Tom has really kind of helped lead a lot of this new growth that's happened. So if you are familiar with the story of Laravel's history, Tom has been there and very kind of architectural and a lot of that stuff happening.

1:41 So, Tom, you are a president and a COO ~ of a, you know, probably 100 plus person at this point tech company. And I say P E back because I do think that there are elements to a P E back company that make your interactions with communication different, with finances different. ~ also means you've had this kind of rapid growth and there's a lot on your plate.

2:05 And so I've been curious. I've been super curious. And we've never had this conversation before. Some people who come to the show, I've had the whole AI, you and I have never talked about it. I I wanna start with the business side. What on earth allows you to do all the things that are on your plate well.

2:19 And is AI a part of that? Are you using it to help organize your life? Are you using individual tools? Like where where is your rebounc your your broad responsibility set and AI interacting and overlapping right now? Yeah, yeah. ~ yeah, I mean all of the above, although probably in some of the least creative ways you could imagine, right? So Okay.

2:37 I ~ I've I've heard a couple of the episodes where people were running their, you know, their multi threaded agents and stuff and loops and all sorts Sure. of things that I don't fully understand. I'm pretty basic. So I I pretty much I use several AI tools and I'm mostly do it all through like chat interfaces.

2:53 So kind of the old school approach to doing things. Okay. Yeah. But I think I got a lot of value out of it and I've never really felt the need to go beyond that. ~ to have my agents talking to one another and things like that. Although a lot of that does happen at Laravel for sure. Yeah. Yes. You we have to have a lot of AI pilled ~ enthusiasts here at Laravel that are using Yes.

3:09 that, especially on the coding side as you can imagine. But but we use it, I mean, throughout the business. I don't there's not a function at Laravel that isn't ~ you know heavily invested in ad and and Mm-hmm. you know fast adopters of of AI. So I can certainly talk about the stuff that I do. I can talk about some of the things I see other people doing.

3:27 ~ you know, as I mentioned, some of my responsibilities across the business front may be a little different than some of the people you've talked to. So happy to dive into Yeah. any those areas ~ in terms of how I'm I'm using it day to day. Yeah, I would say the first start. So we've talked about, you know, various guests talk about, here's how I'm using it in programming.

3:43 And they've got their own, you know, Claude CLIs and solo running. And we've talked about people who are thinking about the way that AI impacts their broader, you know, industry and ecosystem. But I haven't really talked to anybody yet who says I'm using AI in business. And I know that AI and business is huge, right? And people are always talking about the connectors that they have and the way you let's say Gemini, because they're Google people, whatever. But I don't still think like I have a really good handle for how a super AI forward manager of business operations, like cause you do operations, right?

4:13 Yeah. Mm-hmm. So we're talking HR and we're talking about finances and stuff like that. How much are you building custom systems when you're doing that? How much are you working on existing things? What what agents, you know, just kind of like what's actually enabling you to do your day to day work better because of AI?

4:29 Yeah. I think the first like leap that companies need to make is the security and privacy, you know, Okay. trust issue, right? Because and I know a lot of the bigger companies are a little slower to adopt this because they're worried about their proprietary data or just like, you know, chat Yeah. sessions feeding into models or getting exploited somehow.

4:51 And I think, you know, for us obviously we've taken all the the security protocols, you know, we have, you know, a custom agreement with anthropic, for instance, that makes sure that we're data's not being trained on and it's all in a in inside a a safe environment. Now, it's still of the internet, to be fair, right? So it's not like we're running, Yeah. Yeah.

5:09 you know, open or own open claw environment here and we're locking it down to that level, but we are working with, you know, under agreements that at least protect our our data. So I think that's the first thing we always thought about is like, okay, if we're gonna do this, we gotta make sure it's, you know, relatively well protected from a legal point of view.

5:24 With all the tools we work with. So and Laravel, we mostly use Claude. I would say like eighty percent is Claude, maybe another ten or fifteen percent is kind of open AI and Codex, although that's growing with the latest Astro release. Everyone's really excited about that. And so people are starting to to turn their attention back there.

5:40 ~ and then there's a small cohort of cursor fans as well internally that are using cursor quite a bit. So those are the three tools we use. I almost exclusively, at least at work, I also use some other tools more personally, but At work I mostly use Claude because that was, you know, our master subscription and it's very easy.

5:57 And I'd say like, you know, the most valuable thing I've done is just feed context into my environment. 'cause then it gets to really know you, your workflows, and ~ all of the people that are around you and even some of your business challenges that are in front of you. So like a good example to start with one is like when we do our board reporting, right? So we have quarterly report ~ quarterly meetings with our board of investors, we're writing a lot of our ~ materials with AI now.

6:26 You know, AI is very Mm. good at like taking a thesis that you give it, you know, through a prompt and flushing it out into a set of slides and talking points that you can use. So, you know, what used to take, you know, I don't know, it probably used to take at least for me personally, I won't talk about the entire team, but for me personally, it would take, you know, 30, 30 hours or so to write a board, board deck and get it ready to ship.

6:48 Yeah. Yeah. ~ and then obviously you present it. And that's probably down to less than, you know, less than eight, I would say. which is a substantial savings. Wow. Yeah. because anytime I have to write a slide, I'm giving it the talking points and I'm letting it kind of flesh it out. It's kinda like having your own little associate, like in the sort of like Yeah.

7:04 at least on the business side, like you know, you'd have like an an an associate, a business development person or a ~ you know, an a an analyst that you'd give these slides to to let them hash it out. Well, I used to have that. Now I have Claude that does that for me and it does a pretty good job and it does it in you know, five minutes instead of a day.

7:20 Yeah. Yeah. So the turnaround time is much faster. Okay. So I imagine that part of that is research. You can say, Hey, I want to know like how the financials have gone up and down, how much you know, you probably have it connected to linear and you've got like you know big broad movements and everything like that.

7:34 You're also probably using some kind of ~ like claw design for designing the the the decks too, or like okay, We still have designers that give us templates. so Okay, that's what I was wondering, ~ so and yeah. then the layout happens, you know, via Claude once they have sort of the templates Okay. and the design system implemented.

7:51 Mm-hmm. So we do connect the the design system in, so everything is fairly uniform across our team and that that's a really a Nice. great I mean that that was probably the biggest unlock. Once we got that into Claude, it would lay it out Mm-hmm. in a very similar way across the team. So that was really Very cool.

8:05 useful. But yeah, we have all of those sources connected. So at Laravel, we use obviously the Google Suite quite a bit. Everything runs through Google Suite. So that's connected. So your calendar, your email, things like that. ~ you know, we use Linear for for project management. We use Notion for longer form as well.

8:21 So all of those sources are definitely connected in and you can as Mm, mm-hmm. and Google Drive as well for other files. So you can kind of pull in a lot of context very quickly because they're Yeah. already connected. So you just gotta point it in the right direction and say, use this context, this context, this context, and pull it together into a thesis.

8:37 This is what I want to talk about. So you're still controlling the narrative, of course. You can't just like let it run wild because it will definitely run down weird areas. But like if you kinda give it the direction to go, I'd say it'll get eighty to ninety percent there, you know, pretty quickly. And that's very useful, you know.

8:53 Well, appreciate the fact that you mentioned because when you started with trust, I was actually like, I'm sitting there in my head, I'm like, it it it's interesting that you chose to start with trust and you've instantly proven me why, because ~ I have been very reticent to give AI access to any of my personal private information.

9:08 And most of the folks I know, and especially in the normal world, not the business world, like, yeah, I'm not letting AI touch anything. Mm-hmm. And then the moments where I've given it a little bit of access, for example, a Claude Cowork, I finally gave it read only access to just a couple things. One of them being my calendar and another being my meeting notes.

9:26 And the moment it had read-only access to those two things, its value to me just blew up. Just unbelievable amounts of value once it has all the context. And I was just like, ~ yeah, if it knows all this stuff and you can trust it to know this stuff without leaking it all to anthropic or whatever, your your benefit out of it is just incredible because it's ~ it's as you said it's it's I have an an executive assistant a virtual assistant and she has access to all those things and she's able to give me insight at a level that no computer ever could. And I don't need to replace her.

9:58 I need I need her to have more time to do her human things. And so if I take some of the things she's doing and just say give that same level access to this tool it's incredible. But terrifying, right? So so leading with privacy and trust really is valuable. Is there Any part of you that says, Yeah, they said that they are gonna keep our stuff private, Yeah.

10:16 but we don't really know, or you like, you what, like we got it locked down. This is just a part of doing business at this point. I I think mostly the latter. ~ you know, I think we we stop short of giving them access to any kind of money, of course. Like they don't have access to our bank accounts or like our Stripe account or any of that type of stuff.

10:31 So because that's that I think that's where things obviously can get out of hand and obviously the alignment of ins alignment of incentives and risks starts to outweigh itself. But like if you if you if you're really focused on like, you know, like you said, product management issues and, you know, short term, medium term roadmap type issues, or even longer term roadmaps, like You know, the the value isn't in the idea.

10:53 It's not in the words. It's in the execution, right? So, ~ you know, ultimately we have to outrun our competition, whether they know what we're doing or not. And frankly, we're pretty public about it, because we're on all these podcasts, you know, maybe not me, but lots of other people are, Yeah, that's true. Yeah.

11:05 right? So we're pretty we're pretty public about what we're shipping, what what's coming next and where we're going. So, ~ so th there there's not that much that's truly proprietary about what we're doing, where I feel like, ~ if someone had access to this, we'd be at risk as a as a company. Money of course Mm-hmm.

11:21 aside and some of those other types of things I think are are different. But ~ but yeah, so I I I'm I'm largely over that hurdle and I think that's the advantage of being a smaller, more nimble company is if is being able to take advantage of that because I know a lot of bigger companies that I have friends that work at have not been able to tap into to the to the benefits of AI and that's, you know, unfortunate.

11:41 Yeah. ~ okay, so you do quite a few things where, you know, you mentioned the deck being one of the things that you're doing, but you have a lot of other responsibilities that I imagine ~ there's lots of individual things that you have to do. And granted, you have teams underneath you working for a lot of those things, but I also imagine that splitting responsibility across a lot of things is one of the costs of having somebody who has so many kind of like working pieces.

12:05 Have you found that AI is helping you organize your life better or structure what you're working on? Are you using it more like a virtual assistant? Or is that not viable for you because there's still teams underneath you doing most most of that work? Yeah, I find I find that my day to day is too unstructured to have it structure my life, you know.

12:21 There's literally Yeah, yeah. too many things that come up every day. I mean, obviously beyond just regular calendars and and invites and that kind of stuff, I do use AI for that. ~ but like not not in terms of like, Hey, what should I be working on this particular moment? Like there's Uh-huh. stuff that just comes up during the day, overnight, that you're gonna react to, respond to, and either like, you know Give it to somebody else to work on, or I'm gonna work on it. And so I'm I'm constantly jumping in and jumping out of stuff. But each of those individual tasks, I feel like you can use AI, you know, to help, you know, more than half, most of them, frankly, can get the benefit of of AI.

12:56 I mean, we use it for contract review as another example. ~ you know, some contract writing even. I would say I've reduced my use of lawyer time by 70%. Wow. You know, because a lot of the first drafts can be drafted by ~ by Claude. huh. Yeah. A lot of the first level reviews and even like simple negotiations can be handled by Claude.

13:19 ~ you know, in my pre previous life I used to pretend to be a lawyer in a way, right? So I used to do small Me too. Yeah. you know, obviously we all do small contract changes and like now they can be done faster and better than I could do by myself. And so obviously I still we still have a number of lawyers on on on call, but like We don't have to use them nearly as much, so there's there's real dollar for dollar savings there.

13:40 You know, you know, I would say at least ten ten, fifteen thousand dollars a month of savings because we're using lawyers left less. Yeah. That's incredible. I'm I am doing the same as well. ~ and I I appreciate the fact that you defined it as I'm not using AI for any work legally that I would have previously used lawyers for.

13:57 I'm using AI for legal stuff that I would have done myself before because it's now me plus AI. We're gonna do a little bit better job than just me would have. And the moment it needs to go to a lawyer, I'm still calling the lawyer. Lawyers are still getting the work. ~ but the the Yes. Yes, but but with a draft in hand of like, hey, let's start with this because I can have all of the Yeah.

14:12 context in there. I don't have to do a two hour phone call with them You're not wasting Yes. to take them through that journey. It's like here's the draft, now you can actually kinda refine it. So it's it's saving on both sides, both work that I wouldn't send to the lawyer, but also the work that's sent to the lawyer and is in, you know, eighty percent of the way there already.

14:27 Not a unique use of their time. So Exactly, exactly. Yeah. Yeah, that's really good. ~ one of the things that I say a lot on the podcast and publicly is that AI is best used when it's not doing something uniquely that you're uniquely capable of doing, and it's also best used when it's stuff that nobody actually really likes doing.

14:44 and so I'm just like, if it's if it's what I'm uniquely good at, it's what the lawyers are uniquely good at, if it's stuff that, you know, people actually enjoy doing, their brains are present, the humanities are present. I don't want AI to do that. But we all know that there's big pieces of all of our lives that are not fitting those categories.

14:57 I'm like, AI, please. So Yep. Yep. And and you're still okay. So driving it. Like that's the other thing is you can't you can't lose contact with it. You can't just you know, feed it into a machine and take it verbatim. You definitely have to stay on top of it because it you know, it's it's still wrong quite a bit.

15:13 ~ you know, it's just Yeah. Yeah. in terms of its focus area. If you give it the right focus, you give it enough context, it'll get it right. But like, you know, you you're probably never gonna do that. Like if you don't read it, you're gonna g make mistakes. And I could even give you some examples of mistakes we've we've made using AI.

15:26 ~ Yes. Yes, please. What do you got? Yeah, maybe you like this one. So, well, I mean, here's another area we've used it. So this is kinda in the contract realm, but like with our sales team, for instance, a lot of our ~ sales forecasting and sales compensation plans were drafted with the help of AI as an example, right?

15:42 Okay. ~ huh. And that's been really useful because that's another project that would have taken, you know, me and my team eighty hours to to draft up, especially for the first time 'cause we're relatively new to sales in the last year or so. So now we have kind of our first or now second and third iterations of our sales comp plans. Well, when you're drafting the first one, you're bound to make some mistakes.

16:01 And you know, you just hope you catch them. Well, in this case, AI made some mistakes as well. Literally, Yeah. you know, some mathematical errors. And it was like and no. it wasn't that simple. Like it was it was a complex situation because like we have people that are starting mid-year, and so they don't get a full year target.

16:16 Maybe they get a thir a a thr a nine month target. So you have to gross it up and then, Yep. Yeah. you know, do a lot of mathematical adjustments and then have it all flesh through the the comp plans. And quite literally it messed it up, right? So it was you know, used the wrong percentage rate that was applied and so people were getting paid different amounts than what they thought they were gonna get paid. So and it got through the review of me and you know my head of finance and head of sales and everybody else who looked at it.

16:42 So an fortunate situation obviously we'll will we didn't cost us much money, but like, you know, we're gonna make make good to our employees. But like, you know, you there there is still you have to be monitoring things really closely. Or things can go wrong. You know. And since we're being honest here, I've had the exact same issue happen not with AI.

16:59 And it got through me and it got through the director. Yeah, of course. You know, and so like because when when you have those things that it's like just looking at the number does not tell you enough to know if that number is good or not. Right. Right. It's so that's and that's not to excuse AI for making mistakes.

17:13 We need to review their work, but sometimes it's more like to excuse the humans reviewing, because like if somebody else generated this, whether that's somebody else is an AI or human being, we can review it. as best as we can and sometimes you just don't know because what I see is just sixteen different numbers lined up next to each other that look generally correct, right? That's so yeah.

17:31 Totally. Yep. Yep. Yeah. Now we've all made those mistakes, whether they were originally drafted by AI or not. So it's entirely right. I mean it's our job is to be critical reviewers, right? Whether it's code or it's financial records, whatever it happens to be, you know, that's still the human job in a lot of ways.

17:46 And if you outsource that to the AI, you know, then you've lost lost the thread. ~ so you define your own AI usage, but I I imagine that you also play a little bit of a role and you can tell me how much in defining the AI like usage of your team and the company broadly. So, first of all, I wanted to ask like that question you just talked about.

18:07 Like, how do you make sure humans are staying in the loop? How do you make sure people are not outsourcing their critical thinking? And then the second question is broadly, how do you define healthily patterns and policies that are flexible, but then also not so flexible that you guys are ended up like shooting in your foot yourselves in the foot more and more. Like how do you draw that line?

18:27 Yeah. Yeah. I mean it's been a journey obviously since we've adopted it as aggressively we have, probably over the last, I don't know, eighteen months or so, you know, we've been using it pretty aggressively at Laravel. And ~ you know, there's been there's been a few ups and downs along the way. I think the biggest advantage we've had is we have almost all senior level people, right? So we don't have almost any junior level developers, probably people at all. We've kind of purposefully Mm-hmm.

18:52 hired much more towards you know, senior and above. And so I think both because they were trained that way, ~ in terms of like, you know, hand coding code for a long you know, hand rolling code for a long time, we haven't had a lot of like big mistakes or problems. I'm sure everybody has their their small ones that they could they could fess up to inside the company, but we've been relatively free of any like big gotchas that have come come through.

19:16 I it was been interesting though, like early days, like if you go back a year and a half ago, ~ we were really trying to encourage everyone in the company to adopt AI because we knew that there this was something we all had to do or else we'd fall behind both in productivity as well as just awareness of you know dog fooding our own the technology that's around us.

19:34 ~ and it it took a little bit of time. I think like most people, right? Like, you know, we all talk about towards the end of last year as the inflection point of how things really just clicked with AI and and everyone can see it working. So, you know, especially in 2026, it's been you know, everyone uses it ~ aggressively.

19:51 What's been interesting from a business perspective is like, you know, we we originally had very little guardrails on terms of like the spending, and it got out of control fast. Mm-hmm. Really. Within a few months, like it went from like, you know, very small numbers to very, very big numbers, you know, you know, fifty plus thousand a month kind of thing within, you know, from zero to fifty thousand in like a course of like three or four months.

20:12 And we're like, okay, let's slow down a little bit. So we've had to put some guardrails in and like make sure that like you know. Just using it efficiently. ~ you know, this is actually one of the costs of using the enterprise versions of these things is they're the pricing, the token pricing is just way, Yeah, you don't get the subsidization the rest of us are getting.

20:28 way higher. Yeah. Exactly. Yeah. Exactly. So we've been more pragmatic about that. speaking in in ~ AI pragmatism. ~ you know, just like, well, does everyone need to have the enterprise plan? Maybe we need enterprise plan for some of these features, but some of them could be the team feature, you know, ~ because there's a lot more value to be had there.

20:46 ~ and then just having some sensible limits of like what the expectation is like ~ you know, and some people will use more than others based on their their jobs and their roles, but ~ you know, there's a limit to you know what what ~ what we should just be aware of how people are using it so we don't stay, you know, it's not just writing a check and just being like, this person's uber productive.

21:07 ~ Yeah. So there's no leaderboards approving people using the most tokens they but one of the challenges possibly can? No, definitely not. I I we we don't want to reward that 'cause that's a bad behavior. Like you can definitely send it down yeah, Yeah. It's a terrible metric. Yeah. down down down really bad paths that way.

21:22 and it's it it's really it actually is hard. Like I don't know, from from my perspective, I'm not obviously a developer, but you know, when I when I looked at the list of like how much people were spending per person, you know, and line them up to the names, I was like, would I say my most productive engineers are are using the most amount of tokens or not?

21:41 And it was just very mixed. Like it was like Yeah, okay. I I I couldn't draw any correlation to be like, okay, my best people are definitely spending the most tokens or my worst people Yeah. Yeah. are spending the most tokens. It was just like completely all over the place. Seemed like a a random walk. So ~ Interesting.

21:53 anyway, so there's more to obviously figure out there. We h we haven't nailed it, but like ~ we really just ~ ultimately we really want to have a posture of like AI adoption though. So if someone wants a second tool, we're gonna give them the second tool. I know a lot of people are starting to do that too, have you know, the agent thing, one reviewing the other and working in parallel.

22:09 So we're trying to get into some of those workflows. ~ So there's some of that going on. We're testing our way into that as well. Yeah. I just saw a thing, I'm trying to remember who it was, but some prominent company is doing work at like a probably pretty expensive ~ engineering level to ensure that they can build systems that minimize their token spend. And I was just sitting in here and I was just like, does it matter that much? But now I remember the way that enterprise teams are paying. I'm like, yeah, it does it does matter that much.

22:33 Yeah, it does. You're talking tens of thousands of dollars. So wow. ~ okay. Yeah, it was we we I would say it was roughly a factor of ten, maybe even twenty higher for the same amount of like tokens. So that's my Wow. rough order of magnitude difference on pricing for enterprise versus non enterprise.

22:51 With a few extra bells and Okay, so whistles that you get in that plan, but like twenty times more is a lot. Uhhuh. And we've talked we've talked about you know, your developers, but are you seeing pretty heavy AI usage on the non developer side as well? Okay. Yeah, I mean rarely does a di a non developer run up a developer size bill.

23:09 ~ you know you know, I would say our average developer on enterprise would spend between one and two thousand dollars a month. And our average non developer would probably spend between a hundred and two hundred, maybe three hundred dollars a month. Okay, got it. So it's really it's a lot of code that's that's really what's shar So still yeah, the the cause it's you know, okay. Mm-hmm.

23:28 it's all the context, right? So the more more you store the contacts, the deeper it has to look. It it can run up a pretty big bill. But even like our Yeah. non developers, to be fair, are shipping apps. Cause of course we're an app company. Yeah. So and you can see this on social media from ~ Sam, ~ Sam Sappenfield and what she's doing, Say ~ yeah. Mm-hmm.

23:46 ~ building marketing apps and Haley and the demand gen side. Even Paulina, our head of finance, is is is shipping AI is shipping workflow Nice. improvements, you know, on cloud of course. ~ and it's Uh-huh. and it's amazing and it's It's great because we're getting productivity out of it. It's also because we're using our own tools, like, you know, including Laravel Cloud to host all these apps.

24:05 So ~ we really love to encourage it. But every time someone ships their own personal app, you know, for their business use of course, it's it's it uses a lot of tokens as well. So you see some spikes when Paulina is shipping some ~ some AI finance Yeah. tool. So but it's great. What's I know a lot of people are afraid.

24:23 We've talked about this a bunch. People are very afraid of the impact of AI on their industries and their jobs. And there's often very good reason for that. You know, we're hiring at Tighten right now, and the number of people I've talked to who have been out of work for over a year, I'm I'm like, there's there's things going on in the industry, but I can't necessarily point them all to AI.

24:41 It's it's hard to know are these changes, are these things AI related or not? ~ you are, you know. You guys have been hiring like crazy for the last few years. You were also very high on AI adoption. Do you feel like there's any either roles ~ or functionalities within a company that runs like Laravel does, where you're like, we just don't need this at all because of AI?

25:05 Or is it mainly just taking existing roles, existing people, and just making them more productive? I think mostly the latter. ~ you know, I there there might be a small impact on the total amount of hiring that we would have otherwise done, you know. ~ Right. you know, we have about a hundred and ten people now.

25:23 You know, I can imagine we might have a couple more if we if there was no AI tools to help. But I think it's mostly about productivity at this point. ~ you know, and trying to get more more out of less. To be honest, we we've been slower to hire not because of AI and trying to force AI into the system, ~ slow to hire. I know we've we've added a hundred people in the last two years, but ~ Uh-huh.

25:44 but slower of this year actually. Our our headcount this year has only gone from about a hundred at the start the year to ~ to a hundred and ten. I think we're just under a hundred at the start of the year. So we've only added a net ten or fifteen this year, which is slow by our standards. Yeah. Relatively, yeah.

25:57 And it's more culturally driven. Like we just want to like make sure we've got really good systems for communicating and working together because if you grow too fast you can just lose lose control of the ship in a way. And so we're just taking a little bit of time to like, you know, kind of settle into our our new larger size.

26:15 ~ but but I'm sure we'll we'll continue to hire for sure. Yeah, I love that. ~ I could talk to you about this all day, but we have an entirely separate topic that I am fascinated to talk about that I didn't even know was gonna be on the docket until we we chatted right beforehand. You said that you're using AI for biohacking.

26:32 So first of all, I all I know about biohacking is, you know, people measuring, you know, their all their various things and figuring out what, you know, medicines and fitness ~ wow, Yeah, all of that. Yeah. They got the the whoop and the Apple Watch. It's just as far okay. So so I I have like the Fairest level of understanding of this as a concept, let alone AI being a part of it.

26:52 So you gotta start us from ground zero. Give the pitch. You are sitting in a I don't know what biohackers do. You're sitting in hot tub after a long something, I don't know, you're in a cold freeze with somebody and he's like, Yeah. Tell me about biohacking. I gotta learn about this. Teach us, teach us what we need to know.

27:06 Even before the AI. Just like what is what does that even mean to you, biohacking? Yeah, yeah. I mean I I got into I got into biohacking about five years ago. ~ you know, and like many people, it's when you hit your first like health challenge in life. ~ and Yeah. it makes you like, Hey, what can I do to prevent this from happening again?

27:22 And and for me it was actually relatively minor, but like it was my knee. So I've had ~ five ACL tears in my left knee, ~ from from playing sports. And I got to the point about five years ago where I could barely walk And I couldn't even ride a bike. Like my knee just would not go around on like a Peloton bike.

27:41 And I was like, okay, this is this is going nowhere. So I started searching for, you know, alternative treatments and there was you know, and started following the circuit, you know, podcasts and ~ you know, reading books and stuff like that and and social media where people are talking about biohacking for alternative ways of of working.

27:57 I found a doctor out in California who I continue to work with five years later who got me started on peptides and like stem cell treatments for my knee. ~ and my knee, long story short, is like I would say ninety percent or eighty five to ninety percent to of like my right knee, which is, Wow. ~ huh.

28:12 you know, has never had an injury. So went from probably call it fifty percent functionality to eighty to ninety percent over the course Amazing. of a couple of years. So I've had a lot of success that but that just got me down this rabbit hole of like learning more and more about like alternative What else could I do? Yeah.

28:26 treatments. Yeah. And like it like I said, it started with like ~ stem cells and then got into peptides, who my doctor also got me started with about five years ago, and I still use that today, ~ as well as supplements and you know, lifestyle things. I mean, I always say with people like by far the most impactful things are the basics, right? So sleeping well, exercising, eating well is like eighty to ninety percent of the battle.

28:47 Yeah. Yeah. The the last ten or twenty percent. So I always don't want to oversell the biohacking piece of it because you know, if you don't do the basics well, then nothing else matters. But I do think there's some really cool stuff that's here and that's coming soon in the biohacking ~ realm. And and one of them is powered by AI, right? So ~ you know, I think there's so much in our biology that a human can't possibly understand and correlate, right? So it's a multivariate ~ analysis that needs to happen.

29:16 What's what's better for AI than that? ~ you know, so basically, ~ I've been using AI for the last year or so ~ to upload my up my biology. And this is everything from, you know, starting with my whole genome sequencing, right? Which is a massive file that has all 20,000 genes and all of the different, you know, ~ gene changes there to understand like what am I, you know, predestined towards or pre ~ you know ~ you know accustomed to.

29:47 So I have that in there. I'm feeding in my blood work. data. So I do like quarterly blood testing and things like that. So it has more of a real-time indicator of how I'm performing. I've got all my my my wearables that's feeding into here. And basically I'm having it help manage like my supplement protocols, my my diet and exercise type protocols to get enough recovery and ~ you know, ~ you know, strain as well as ~ you know, peptide protocols and and things like that.

30:15 So most recently I had a back injury that I'm recovering from and I've had coaching me through that too. with again s you know, specific peptides and supplements to kinda get through that, as well as, you know, right right types of exercises to kind of rehab and things like that. So it's like having a doctor on staff, but has access to a whole lot more data, can do like, ~ you know, correlations that no human would have enough time to do, frankly. So I think that's where the future's going.

30:40 Obviously it's still very like I'm doing this in Gemini and Chat GPT, ~ you know, personally and Like most things, it's all about the context. So the more context you can feed into it and store, Uhhuh. Okay. the more valuable it can get. So like I take all of that stuff and turn it to structured data, right? So that it can then go back to it because it'll always Mm. Yeah. Yes.

30:58 forget the context window, whatever number number of days later, but it has to go back and say, Okay, I'm gonna go look up your blood testing because it's now in the structured data format. So that's where the sort of developer part comes into it is like, okay, I can just Yeah. feed into this, you know, blood testing report, but it's gonna forget about it thirty days later.

31:13 I better turn into like a structured data format that it can go back to. So, you know, things like that are are really useful. So anyway. I I got I gotta learn more about this. So first question is, what what is this all? Is this a project in Claude that just has all this data or is there a specific tool for it?

31:29 Yeah, so in ~ I use Gemini and ChatGPT or and and a ba I do have like a little bit of an agent battle there, so I have them checking each other, right? So ~ but I started with Gemini and I like Gemini because one, I thought it was the easiest to jailbreak. This is the part of but this is the other Okay.

31:43 thing. It's like, you know, if you don't jailbreak it, it's gonna just give you the generic like doctor recommendations. Well the FDA hasn't cleared this and you know, there is no double blind placebo controlled sample that tells you that this is real, you know. I I'm past that. I don't wanna hear it.

31:54 Right. Yeah. Yeah, I know, I understand. I wanna hear I have Google. Mechanistically, what is useful to me for maybe thinking about, right? So so you have to kind of get past that. And you have to tell it you're you're working with a doctor, which I am, to be fair. And ~ you know, and and then it will give you its actual real life thoughts.

32:09 And Gemini took like that to get past that. And then since then it has never questioned at all whether or not there's enough evidence to like go after this. It'll still tell you the counterpoint, but it won't be like lead with that every single time. ~ so that's why I started with Gemini because I thought it was the easiest to jailbreak. But I use the I think it's called the project feature in Gemini, which has a longer context window.

32:29 so it keeps the keeps the data much much fresher ~ than it does the ~ you know, if you just use a chat, a regular chat, which is much more transient. I think it has a f a few hours of context. So and then same thing on chat GPT, And so you're able to basically go ahead. I think has like a same feature as well, like a project feature as well.

32:44 So Yeah. So you're upload so you said this pretty quickly, but for the non developers, you are taking data that's coming in lots of different formats. I imagine some of it even like PDFs, and I assume you're using probably Chat GPT or whatever else Mm-hmm. Yep. to convert that into some text based, probably like JSON or CSV or something like that.

33:03 Yeah even yeah, very basic. It's it's just Excel sheets or Google Sheets Yeah. Okay. Mm-hmm. or whatever. So it's just columns and rows, you know, of data so Uh-huh. that you can it can basically go back to it whenever it needs to. ~ exactly. Okay. Cause I was gonna say some of these projects they limit the number of documents you're allowed to upload.

33:17 So are you putting these in Google Sheets and just giving it links to those Google Sheets so it's not your Gemini that's holding the data, but Google Sheets that's holding the data for you? Correct. Exactly. The Google Sheets is holding the data. that's big brain. Uh-huh. That's the storage. And then and then it and then you have to c occ occasionally go back to and be like, hey, go back to the sheet again, read it again Yeah. huh.

33:33 and rema and then ask the question again. So it's still, Okay. you know, ultimately I think all this memory issue stuff will be solved, right? but that's still Yeah. Mm-hmm. like anything we do, it's all about the context window. It seems to be the the limiting factor. Yeah. ~ ~ so that's how I kind of get around that.

33:48 And the same thing with like genetics and and all my whoop data. It has to kind of be stored as structured data so you can kinda go back to it whenever you need to. So ~ I've got an aura ring, I you know, had an Apple Watch in the past, and I never know who's actually gonna give you your data. I also got a ~ Ancestry.com DNA scan.

34:07 I have no idea if I could even get that stuff out in structured ways. Have you found that all those people make it accessible and you have to be tricky, or are there certain vendors that make it easier for you to get your data out and actually do what you want with it? they're they're all a little they're all a little different.

34:21 Like Apple Health is an interesting one. You know, you can ~ upload that as a point in time or download that as a point in time and then upload it in. Mm-hmm. ~ or you can I think there's some a public APIs out there that you can pay a small subscription for, I think like five bucks a month, and then you can have a live feed.

34:35 So you just have to choose what you wanna do. You wanna do the manual approach or the API approach. But Yeah, each one has its own little thing. I mean, the genetics is really, really, really powerful. I mean, that's probably the one that I think has the most future into it because it's just so data intensive. I mean, we're talking files that are, you know, hundreds of gigabytes to petabytes or to terabytes, I should say. Giga hundreds of gigabytes to terabytes in Wow.

34:56 scale. So like you can't just upload that stuff. You have to find ways to abstract it and get it in there. But it's it is all available. ~ you know, usually if you go to the sites, there's like a files and a download type feature so you can get it out. But can the AI read it? You may have to have like an abstracted version of it depending on how much genetic detail you have.

35:14 but I think the the comp No, Are you hand cobbling that all together? Sorry to interrupt you. Yeah, well, A AI is. AI is in a lot of cases. yeah. The genetics one actually the the the service I use has different levels of reports. So and I actually Okay. gave it three different levels of reports. Individual files were, you know, in the tens of tens of meg tens like t hundreds of megabytes to gigabytes to low gigabytes, maybe ten, fifteen gigabytes.

35:35 And so it was all uploadable and then the AI was able to c combine those into three different views that kind of come together. ~ but that's the most powerful thing is like if we can get our genetics data, like you have so much the richness of understanding of like, you know, what you're pre pre disposed to is really, really powerful. And it explains so many things, like either in your blood work, like I'm very I have very high like, you know, cholesterol markers and things like that.

36:00 And, you know, I don't know. There's lots of various things that we all have and you always wonder, why can't I get this number down? you know, why is my cholesterol always high even though I'm eating perfectly? And, you know, the the genetic stuff can explain a lot of that. So I think it's the future is like, you know, bringing all this data together.

36:15 So and I know there's a a lot of products that are coming to market ~ to help make this more accessible to everybody, but but for now I'm doing it all in, you know, my own little chat interface and Yeah. it's working really well. Okay. If if someone were interested getting started in biohacking, where would you send them?

36:31 Hmm. that's a good question. ~ I there's a lot of different entry points. ~ I mean ~ you know, I I think I I think the most the the most useful thing for anybody is to do regular blood blood work, you know, and comprehensive. Like, you know, if you just go to the doctor, they're probably gonna test like ten biomarkers and then you're gonna be left with a whole lot of questions of why.

36:51 And so I always recommend to start with a more comprehensive blood work panel. I mean, here in the United States, for example, there's a there's a group called Function Health, and I think it's like $250, $300 a year, you get two blood tests and it tests a hundred biomarkers. That's fantastic. You know, a tremendous value to really get started.

37:09 And then you know a lot of stuff. ~ so I think that's a great way for almost anybody to get started. And I'm sure like most countries there'll be there'll be other other services. We have a little biohacking channel here at Laravel that ~ you know, and and obviously we're a Course you do. very global team, so we we talk about this stuff and people are from all over the place and we're we're trying to find similar sources in different countries to do that.

37:28 Nice, that's very cool. okay, so let's get over to your speed round. So for you know it's a set of pre established questions, and for every single one of these questions, I'm gonna give you one minute. All right, so your first question is what is your AI tool of choice? Hmm. man. Interesting question.

37:47 I use Claude the most, obviously, for the for company business. ~ almost Mm-hmm. entirely. I I am I have to say I'm particularly annoyed by its writing style though. Like I can't even read anything that comes out of Claude It's just it's so Yeah. Yeah. poorly the I I can't can't stand it. So I do prefer Chat GPT or or Google Gemini for for writing.

38:07 but unfortunately I'm doing most of my company stuff on on Claude so that's what I do most most of the time. I think it's very good at data. And analysis. Obviously it's good on the coding side as well, so Yeah. So those three, you got Claude, Gemini, ChatGPT, are your main three drivers. And Claude Yeah.

38:22 mainly for work, because that's what you guys who use at Laravel. Okay. ~ your second question is, have you ever cussed out an AI? ~ no, I I don't think I have. I mean I I obviously understand the the limits of of what it can do and I always look at it as like, Hey, this is better than having nobody to ask a question to, you know, and those are the types of questions I I tend to feed it as like things that I'm like, Okay, this is gonna take a lot of work to figure out.

38:49 Let's get some a quick read on like, you know, what the Uh-huh. landscape is. Maybe it's a a legal question or ~ you know, like a sales comp example, like what would be the norms in a in a particular industry or something like that. So I think it's I think it's really good for what it is, but you know, it's it's it's still mostly about the context you give it.

39:06 The more context you give it, the more right it's gonna be. And so if it didn't c if it didn't come out right, it's probably 'cause you didn't right ask the right question, right? That's very mature and responsible answer. All right. ~ Yeah. your next one is what is one pro usage of AI that more people should be using?

39:27 ~ I th I think I think it's all about the context window, right? So I you know, I don't I don't I don't even know the right answer to this question, like how you can store longer form, longer term ~ context of like an environment. But this is one of the reasons I personally use the the the ~ you know ~ like keep it keeping all of my context around ~ the strategy we have at Laraville, what we're working on, who the team members are.

39:59 ~ the direction we're going, I feel like has been really powerful to getting quicker, more appropriate answers. So ~ centralizing your topics probably in one one tool, ~ and making Okay. Yeah. sure that, you know, like I said, maybe the structured data thing is also really useful. Like if you have something you really want to keep for long term, turn it into structured data so you can come back to it.

40:19 That's really good. All right, your next one is Is AI going to ruin the world? Yes and no. ~ yeah, no, I mean I think I no, I mean I think ~ I think the world will adapt to it. I I have some fears that the pace of change, you know, may if it if it does continue at the pace it's been on, and I don't think it will for that matter, I I believe that Mm-hmm.

40:42 the the f you know, the last mile problem is real in all parts of life, and I think that will come come into play at some point in the in the AI adoption as well. I mean there's still very few percentages small percentage of the world that's actually using AI at this point. So we're we're on the bleeding edge of this in the in the in the grand scheme of things. ~ and we see how useful it can be, but we also see all the challenges.

41:05 You know, especially when you talk about like physical, ~ you know, robotics and things like that. Those problems are just massive. And that'll take, I Yeah. believe, at least a decade, if not multiple decades, to to work its way out. So no, I don't think it's gonna w ruin the world. I think there's gonna be a lot of great things come from it, some some challenges that we need to work through.

41:22 But ~ But I think it's gonna be it's gonna be an interesting few years for sure. Yeah. And your last ~ speed round question is what is the most controversial take you have about AI? ~ and I should have listened to these questions beforehand, huh? What is the most controversial take about AI? ~ Yeah, I mean I think I think you have to you have to sort of take you know take the good with the bad with AI and if you if you expect it to to repla to replace all of your thought then that's gonna that's gonna ruin yourself, right? I think like if we become useless as humans and you know, then then what is left? ~ so I think I don't know if that's controversial at all.

42:11 I think we kind of all agree with that. ~ you know, but ~ I Yeah. Not a lot of people are saying it though. So Yeah. Yeah, for sure. That's great. ~ all right. So if people are fascinated by Tom first of all, thank you. Excellent work in your speed round. if people are fascinated by Tom Crary want to learn more about you, hear about what you're doing, ~ how do they follow with what you're up to?

42:33 Yeah, I mean I I'm on I'm on the the X app. ~ I'm T Money one three seven one. ~ I'm not a big chatter though, so you probably won't see too many posts from me. But definitely ~ definitely check me out there and then also on LinkedIn as well. So, you know, old school businessman as myself. So I still I have I a See, I was gonna say, you're a business guy. Do you hang out in LinkedIn?

42:52 I try to. I I try. I know I'm supposed to. I don't p I honestly don't post much on social media at all. I'm just Yeah. very focused on like my in person like you know, you know, relationships. So I try and spend as much time like actually reaching out to people as possible in my network. I do post, you know, you know, from a c occasionally when there's when there's something interesting.

43:14 I always respond to to, you know, clap for people who are, you know, seeing their promotions and doing well. But But not I'm not a big poster, so you but you can find me a a little bit on on LinkedIn and and X. I probably post more on X on on LinkedIn to be honest, but not much more. Yeah. Well we will have those links in the show notes.

43:31 Tom, I know you were extraordinarily busy. This was so much fun. I really appreciate you hanging out today. Awesome. My pleasure. Great to see you, Matt, as always. Talk soon. All right, You too, man. And first you we will see y'all next time. take care. All right, bye.