How to buy software in the age of AI

00:00:00,600 --> 00:00:32,750 [Eric] [upbeat music] Welcome back to the Token Intelligence Show. AI is changing the way we work, and Token Intelligence is here to help you understand the state-of-the-art, cut through all of the noise, and apply wisdom to become an effective leader in the age of AI. John, I hear a little bit of background noise, and that's because you're on-site at Snowflake Summit. 00:00:32,750 --> 00:00:32,870 [John] It's loud. 00:00:32,870 --> 00:00:34,600 [Eric] Uh, how has Snowflake Summit been this year? 00:00:36,160 --> 00:00:44,070 [John] Man, we were just talking about this before we hit record, and, um, it's a weird time. It's a weird time. I, I think 00:00:45,240 --> 00:01:05,860 [John] it's been probably as you would expect, a lot of announcements that, um, of AI further in the product, right? Um, and it's just a... It's really interesting to be here and see all of the vendors building these AI features into their product, and then see Snowflake from the stage building the same I- AI feature into their product. And it's just a wild time. 00:01:05,860 --> 00:01:06,039 [Eric] Yeah. 00:01:06,040 --> 00:01:08,460 [John] It's hard to, hard to describe. 00:01:08,460 --> 00:01:52,540 [Eric] Well, I-- that actually is a great segue into what we wanted to discuss today, which is how you buy software in this new world of AI. And for those listeners who have never been to Snowflake Summit, Snowflake's one of the, like, premier providers of data warehousing solutions, so think about things like analytics and, um, you know, data storage. Snowflake is, uh, one of the big providers there. They had the biggest tech IPO in history at the time they went public, uh, so a very important company. And when you walk around Snowflake Summit, it's a gigantic conference, and there's all sorts of different software vendors there, an entire... multiple ballrooms actually, um, 00:01:53,580 --> 00:02:26,340 [Eric] worth of software vendors who, um, you know, who provide software. And so I thought it would be helpful, John, for us to talk about how to buy software today. Because if you walk around and look at all those vendors, it was overwhelming before the age of AI, but really, the calculus has kind of changed. So what I'd like to dig into is, one, can you build software today that you would have previously purchased? Because, you know, AI helps you do that. Um, 00:02:27,800 --> 00:02:28,220 [Eric] second, 00:02:29,560 --> 00:02:42,049 [Eric] what is the new architecture? Like, are there new architectures that are important in the age of AI? And then lastly, who's selling the actual software, which I think is important. Um- 00:02:42,049 --> 00:02:42,049 [John] Yeah. 00:02:42,049 --> 00:02:48,480 [Eric] So I'm going to, I'm gonna have you give us a, a mini master class in how to purchase software today. 00:02:50,240 --> 00:03:08,390 [John] Yeah. Um, like I was saying earlier, it's really been on my mind this week 'cause you walk around and there's... I mean, hundreds is not an exaggeration. Literally hundreds of, you know, software vendors here. Um, so I think the first thing we have to think about, which is an age-old question, is build versus buy. 00:03:08,390 --> 00:03:08,440 [Eric] Yep. 00:03:08,440 --> 00:03:53,600 [John] Like, that is even more so on the table than it ever has been because building is easier than it ever has been, um, for companies to make that decision. Um, and then I think, I think there's three components that I wanna break down for build versus buy. Um, and this would be true of, of... This was true 20 years ago too. Like, so this is more of a timeless framework. Like, number one, like, really think through how this, uh, software fits your company, whether you're building or buying. Number two, um, cost, and you have to think through hidden costs, not just the, the ticket price. And then risk, um, security, vulnerabilities, you know, et cetera. So I, I think those are the, the three main characteristics I'd still think about when talking build versus buy. 00:03:53,600 --> 00:03:53,880 [Eric] Yep. 00:03:53,880 --> 00:04:05,000 [John] But, um, Eric, I think you had a couple examples of, um, where you faced that kind of build versus buy decision in the last year. Um, love to hear how you've kind of a- approached that. 00:04:07,020 --> 00:04:20,760 [Eric] Yeah. I think, uh, you know, Vercel is very, um... We like to live on the bleeding edge, um, and so we use AI a ton. I'm on the marketing team, specifically lead the content team, 00:04:22,120 --> 00:04:22,900 [Eric] and we have, 00:04:24,740 --> 00:04:29,510 [Eric] we have built a lot of tooling that we previously would have purchased, right? 00:04:29,510 --> 00:04:29,520 [John] Mm-hmm. 00:04:29,520 --> 00:04:50,820 [Eric] So you can think about, you know, sort of AI-enabled, you know, uh, document creation, drafting, you know, analysis, all those sorts of things. Um, so we've built a lot of that ourselves, but even, uh, I would say, like, you know, that's probably, uh... You know, Notion probably packages a lot of that stuff for the average company out there- 00:04:50,820 --> 00:04:50,830 [John] Mm-hmm 00:04:50,830 --> 00:04:52,950 [Eric] ... because their AI is really good. 00:04:52,950 --> 00:04:52,960 [John] Yeah. 00:04:52,960 --> 00:05:06,080 [Eric] But on the more practical side at Vercel, um, our customer support agent is incredible. We used sort of a, a combination of vendors previously, and then we brought it in-house and rebuilt it with AI, 00:05:07,180 --> 00:05:11,570 [Eric] and it handles over 80% of, of tickets. Um- 00:05:11,570 --> 00:05:12,320 [John] Nice 00:05:12,320 --> 00:05:27,770 [Eric] ... which is incredible. And then we also have, uh, another popular one that our COO talks about a lot is, um, our lead agent, and so it basically does the work of an outbound SDR, but it can do it 24/7. 00:05:27,770 --> 00:05:27,820 [John] Right. 00:05:27,820 --> 00:05:35,840 [Eric] And they were actually able to decrease the size of that team from, like, 9 or 10 people to 1 or 2 people- 00:05:35,840 --> 00:05:36,570 [John] Wow 00:05:36,570 --> 00:05:44,940 [Eric] ... which is wild. Um, and it really does an outstanding job. And they basically trained it by shadowing the top performing- 00:05:44,940 --> 00:05:45,500 [John] Mm-hmm 00:05:45,500 --> 00:05:57,170 [Eric] ... SDR, uh, understanding what do you hate about doing your job, what are the manual pieces, et cetera, and then sort of mapped those out to tasks that an agent, um- 00:05:57,170 --> 00:05:57,170 [John] Mm-hmm 00:05:57,170 --> 00:06:04,910 [Eric] ... you know, tasks that an agent could accomplish, and it works great, and it, you know, we're, we've been running it in production for quite some time. 00:06:04,910 --> 00:06:04,920 [John] Right. 00:06:04,920 --> 00:06:20,550 [Eric] And so these are software, you know, these are software vendors who you would have paid... you know, hundreds of thousands of dollars to, um, previously, right? Um, you know, I mean, depending on the software, right? Um- 00:06:20,550 --> 00:06:20,680 [John] Yeah 00:06:20,680 --> 00:06:23,930 [Eric] ... but customer support software can get very expensive if you have- 00:06:23,930 --> 00:06:24,180 [John] [laughs] 00:06:24,180 --> 00:06:36,630 [Eric] ... a large customer base. Um, and so I've seen this play out in, you know, in real time and actually work. So, and companies are proactively trying to figure out, "Can I offload 00:06:37,650 --> 00:06:42,190 [Eric] software vendor costs by building these tools myself- 00:06:42,190 --> 00:06:42,330 [John] Right 00:06:42,330 --> 00:06:56,200 [Eric] ... with AI?" I would say the flip side, which you probably think about a lot with your customers, is there, it's not like you just flip a switch and, you know, you prompt, you prompt Claude, flip a switch, and, you know, you have a, like a piece of enterprise software. 00:06:56,200 --> 00:06:56,250 [John] Right. 00:06:56,250 --> 00:06:56,970 [Eric] A lot of the, 00:06:58,050 --> 00:07:05,790 [Eric] you know, there's a lot of ongoing maintenance and, um, feature development, right? So it is actually real software that you have to maintain. 00:07:05,790 --> 00:07:15,390 [John] Right. And the thing about build versus buy, like just looking at, looking at this framework, so like fit, cost, and risk are three of the big things. Um, 00:07:16,750 --> 00:07:25,630 [John] it's always been true, and it's even more so true that if you're gonna build something, it will probably fit your company more closely to exactly what you need. 'Cause you won't build- 00:07:25,630 --> 00:07:25,640 [Eric] Yes 00:07:25,640 --> 00:07:27,640 [John] ... features you don't need, obviously. Um, and- 00:07:27,640 --> 00:07:27,640 [Eric] Yeah 00:07:27,640 --> 00:07:41,410 [John] ... that's always been true. Like in data land, I think I, I was joking with somebody the other day, like how are local Claude automated task different than like access databases and macros like before? 00:07:41,410 --> 00:07:42,080 [Eric] [laughs] 00:07:42,080 --> 00:07:44,070 [John] Like, it's branded differently. 00:07:44,070 --> 00:07:45,090 [Eric] Yeah. It's- 00:07:45,090 --> 00:07:45,240 [John] But, 00:07:46,250 --> 00:07:46,529 [John] uh- 00:07:46,530 --> 00:07:46,560 [Eric] Yeah 00:07:46,560 --> 00:07:47,310 [John] ... you know, it's- 00:07:47,310 --> 00:07:47,320 [Eric] Yeah 00:07:47,320 --> 00:07:49,660 [John] ... if it's a local thing you built on your computer that, that- 00:07:49,660 --> 00:07:49,660 [Eric] Yeah 00:07:49,660 --> 00:07:52,190 [John] ... that kind of runs and does data stuff, like it's- 00:07:52,190 --> 00:07:52,890 [Eric] Totally. 00:07:52,890 --> 00:08:05,690 [John] Yeah. So just, it's a funny thing that came up. But um, it's gonna fit your company the best. Like, like you, you know if you've got a, a power user that understands your business that can build something like that, that will fit your company the best. 00:08:05,690 --> 00:08:05,700 [Eric] Yep. 00:08:05,700 --> 00:08:08,960 [John] I think cost and risk are where you get, 00:08:10,270 --> 00:08:11,630 [John] where it's trickier. 'Cause again- 00:08:11,630 --> 00:08:11,750 [Eric] Yes 00:08:11,750 --> 00:08:20,239 [John] ... back to that access database thing, like over, you know, that, that kind of had its peak and went back down, and a lot of that was risk. Like, "Uh- 00:08:20,239 --> 00:08:20,239 [Eric] Yep 00:08:20,239 --> 00:08:23,110 [John] ... we don't really want all our data sitting on one guy's computer," for example. 00:08:23,110 --> 00:08:23,890 [Eric] Right. Exactly. 00:08:23,890 --> 00:08:40,179 [John] Um, and, and there's other things. But on the flip side, and we'll, we'll talk about this in a minute, there's, there's absolutely risk with, um, you know, using 20 different SaaS providers, 15 of which are startups that like maybe aren't following the best, you know, security policy. 00:08:40,179 --> 00:08:40,730 [Eric] Yeah. 00:08:40,730 --> 00:08:41,030 [John] Um- 00:08:41,030 --> 00:08:42,210 [Eric] For sure. 00:08:42,210 --> 00:08:42,350 [John] But, 00:08:43,390 --> 00:08:45,090 [John] but again, like with that framework, 00:08:46,250 --> 00:08:48,610 [John] I think cost is the most interesting thing. 00:08:48,610 --> 00:08:48,630 [Eric] Mm. 00:08:48,630 --> 00:08:54,430 [John] Because cost, the thing about like total cost of ownership, I think everybody's heard a lot about. Um, 00:08:55,550 --> 00:09:11,630 [John] cost is confusing because tokens now versus tokens in the future, I think everybody agrees there's gonna be different. Like, it's gonna be different. Some people are like, "It's actually gonna be way cheaper 'cause we'll all use open source models." Other people are like, "We'll be using model providers and it's gonna be more expensive." 00:09:11,630 --> 00:09:11,750 [Eric] Right. 00:09:11,750 --> 00:09:14,570 [John] Nobody knows, but there's for sure gonna be a cost component. 00:09:14,570 --> 00:09:14,670 [Eric] Yep. 00:09:14,670 --> 00:09:14,820 [John] Um, 00:09:15,970 --> 00:09:36,450 [John] and then the other component of cost is the maintenance piece, right? Of what, what's the plan? Like, there's a lot of people that thinks it's, think it's really fun to vibe code something or to get a, a little app working and done and, and what I'm gonna call it 90%, maybe 85%. 00:09:36,450 --> 00:09:36,710 [Eric] Yep. 00:09:36,710 --> 00:09:41,110 [John] It's way less fun to get it to like 99% or 100%. Like that is the- 00:09:41,110 --> 00:09:41,280 [Eric] [laughs] 00:09:41,280 --> 00:09:42,850 [John] ... unfun work that takes forever. 00:09:42,850 --> 00:09:44,570 [Eric] Yeah. It's brutal. 00:09:44,570 --> 00:10:04,790 [John] And it's still, and it's still true. And then you have to have s- And I think what it comes down to is a lot of the people building... 'Cause I, I mean, you know, my background, like a lot of like ops and data stuff. Um, so we're, we're, uh, you d- typically workflow-wise, the development team would hand stuff to us like, "Hey, run this," and if it goes wrong- 00:10:04,790 --> 00:10:04,970 [Eric] Right 00:10:04,970 --> 00:10:07,690 [John] ... like we have to fix it on the op side- 00:10:07,690 --> 00:10:07,700 [Eric] Right 00:10:07,700 --> 00:10:24,820 [John] ... from my background. Um, and I think you're gonna see a lot of that, especially with like less technical users writing the code. Even if they did a fine job, and even if it's like, let's just argue they're using, it's six months from now and the models are even better and the code's actually not that sloppy, for example. 00:10:24,820 --> 00:10:24,820 [Eric] Yep. 00:10:24,820 --> 00:10:34,290 [John] Um, you still have that maintenance problem of who's responsible, and I think it'll be interesting, like are non-technical users gonna feel responsible or are they gonna feel like, "Well- 00:10:34,290 --> 00:10:34,510 [Eric] Hmm 00:10:34,510 --> 00:10:39,270 [John] ... I built it," and like, "I don't know, you guys deal with it now." Like I, I don't know. Um- 00:10:39,270 --> 00:10:53,430 [Eric] Yeah. I think there's a middle ground though, and I think this is one of the most interesting developments over the last year because [clears throat] it's... You know, when I talk about running, you know, building your own customer support, 00:10:54,610 --> 00:10:59,310 [Eric] you know, agent and system and running that in production- 00:10:59,310 --> 00:10:59,450 [John] Mm-hmm 00:10:59,450 --> 00:11:00,730 [Eric] ... you know, that's a big deal, right? 00:11:00,730 --> 00:11:01,720 [John] Yeah, for sure. 00:11:01,720 --> 00:11:07,030 [Eric] And a big project and a super critical f- you know, a team has to maintain it, and you have, you know, you gotta- 00:11:07,030 --> 00:11:07,040 [John] Yeah 00:11:07,040 --> 00:11:09,280 [Eric] ... have people on call and it's, it's- 00:11:09,280 --> 00:11:09,280 [John] Yeah 00:11:09,280 --> 00:11:11,490 [Eric] ... enterprise software running in production. 00:11:11,490 --> 00:11:12,290 [John] Yeah. Yeah. 00:11:12,290 --> 00:11:15,860 [Eric] But I think, as I think about evaluating 00:11:17,450 --> 00:11:22,510 [Eric] build versus buy, it's not just that choice, which I agree th- with you that- 00:11:22,510 --> 00:11:23,310 [John] Yeah. There's really- 00:11:23,310 --> 00:11:23,520 [Eric] This, it's not- 00:11:23,520 --> 00:11:24,280 [John] ... a third option here. Yeah 00:11:24,280 --> 00:11:27,060 [Eric] ... it's not so different than 20 years ago, right? 00:11:27,060 --> 00:11:27,069 [John] Yeah. 00:11:27,070 --> 00:11:30,330 [Eric] It's just that you can do it way, way, way faster. 00:11:30,330 --> 00:11:30,490 [John] Yeah. 00:11:30,490 --> 00:11:31,350 [Eric] Um- 00:11:31,350 --> 00:11:32,110 [John] Yep 00:11:32,110 --> 00:11:35,310 [Eric] ... uh, you know, and you don't have to have a full engineering team to sort of build it- 00:11:35,310 --> 00:11:35,319 [John] Yeah 00:11:35,319 --> 00:11:41,920 [Eric] ... right? But you still have to think about the maintenance side. But there, the third option is, there are actually sort of two flavors of this. 00:11:43,270 --> 00:11:50,380 [Eric] One is that a lot of the tools that you're probably already using are adding AI functionality. 00:11:50,380 --> 00:11:50,380 [John] Yeah. 00:11:50,380 --> 00:11:59,610 [Eric] We talk a lot about Notion on the show, but another example that comes to mind is Linear, which is a very popular project management tool similar to like a Jira- 00:11:59,610 --> 00:12:01,380 [John] Like a Jira alternative, yeah 00:12:01,380 --> 00:12:04,650 [Eric] ... um, yep, a Jira alternative or Asana. You know, you can kinda think about it- 00:12:04,650 --> 00:12:04,660 [John] Mm-hmm 00:12:04,660 --> 00:12:18,638 [Eric] ... as issue tracking, project management, project planning. And, um-You know, they've been aggressively adding AI functionality, and now it can review code diffs, you know, for software- 00:12:18,638 --> 00:12:18,648 [John] Right 00:12:18,648 --> 00:12:25,558 [Eric] ... engineers, right? Like Linear is a, you know, an issue tracking tool. And so I think there's a world where, 00:12:26,658 --> 00:12:48,298 [Eric] um, in many ways you don't have to build because the vendors that you're already using are augmenting their platforms with AI capability in a way that you can just adopt and will probably do a lot of things for you that are great. Um, but then the other part is that there's this new world of personal softwares is what some people call it. 00:12:48,298 --> 00:12:48,798 [John] Mm. Mm-hmm. 00:12:48,798 --> 00:12:51,068 [Eric] You know, or ephemeral software. 00:12:51,068 --> 00:12:51,078 [John] Mm. 00:12:51,078 --> 00:13:02,578 [Eric] And so one of the, one of the reasons that a company will buy software is to align everyone on the same process, remove inefficiencies, collect data in one place, um, 00:13:03,658 --> 00:13:09,878 [Eric] you know, solve some sort of business problem or create, you know, take advantage of some sort of business opportunity, right? 00:13:09,878 --> 00:13:09,898 [John] Sure. 00:13:09,898 --> 00:13:13,998 [Eric] And a lot of times that's, that's some sort of alignment or process fix. 00:13:15,198 --> 00:13:18,138 [Eric] But one of the other things that I would encourage people to think about is, 00:13:19,498 --> 00:13:31,128 [Eric] you know, if you have your security and permissions, like, managed correctly, your employees can build their own tools on their own computer and just use them. They don't even have to be- 00:13:31,128 --> 00:13:31,128 [John] Perfect 00:13:31,128 --> 00:13:37,978 [Eric] ... necessarily deployed, like, you know, to a URL. Although we do that. This is something we do all the time with Vercel- 00:13:37,978 --> 00:13:37,988 [John] Mm-hmm 00:13:37,988 --> 00:13:39,348 [Eric] ... 'cause it makes it so easy, right? 00:13:39,348 --> 00:13:39,378 [John] Right. 00:13:39,378 --> 00:13:45,958 [Eric] So other people can potentially use it. But I have lots of little utilities that I just run locally- 00:13:45,958 --> 00:13:45,968 [John] Right 00:13:45,968 --> 00:13:51,548 [Eric] ... through Claude Desktop, you know, to do my team planning, to do project tracking, to do- 00:13:51,548 --> 00:13:51,548 [John] Yeah 00:13:51,548 --> 00:14:00,918 [Eric] ... whatever, right? And it connects to all these other systems, and it's my own little process and, you know, it, and it can change as much as I want. I can tweak it as much- 00:14:00,918 --> 00:14:00,928 [John] Mm-hmm 00:14:00,928 --> 00:14:07,168 [Eric] ... as I want. Um, it's not disturbing any other systems, and it makes me dramatically more productive, right? 00:14:07,168 --> 00:14:07,178 [John] Right. 00:14:07,178 --> 00:14:21,737 [Eric] And so that's another thing of build versus buy, which it's not just a enterprise software maintenance decision. You can actually give people the ability to build their own personal tools that make them dramatically more efficient. Now, that's a little bit- 00:14:21,737 --> 00:14:21,747 [John] Yeah 00:14:21,747 --> 00:14:27,138 [Eric] ... different than, like, you know, aligning a business process or different teams. Um- 00:14:27,138 --> 00:14:27,828 [John] Yeah. Well, I think- 00:14:27,828 --> 00:14:29,518 [Eric] But I think it goes overlooked a lot. 00:14:29,518 --> 00:14:39,318 [John] Yeah. It does, and I, and I think that's the right motion. Like, here's how I would order it. Like, I think the future is that first. 00:14:40,338 --> 00:14:40,378 [Eric] Yes. 00:14:40,378 --> 00:14:47,548 [John] Which you could call that a version of building, but it's a new version of building. And, like, there's no question that you should just start there. 00:14:47,548 --> 00:14:47,558 [Eric] Yep. 00:14:47,558 --> 00:15:01,428 [John] Like, even if you decide to buy something, you should... Like, I think that's the new world. Like, you, you start there, then you have, like, roughly what you want. It's not feature complete. There's bugs. It doesn't matter. You just have roughly what you want. It, it's almost like requirements gathering in a new way. 00:15:01,428 --> 00:15:02,798 [Eric] Mm-hmm. Yep. Absolutely. 00:15:02,798 --> 00:15:25,218 [John] Then you can approach build versus buy, like, in a much more intelligent way. Um, and for the personal productivity thing, I think the challenge is gonna be, um, you're gonna have people like you that are, that are power users that just get so much value out of it. How do we then take that, and you're like, "All right. We've got a team of 100 people in marketing. Like, we wanna, like, everybody to be able to, like, get this value out of it"? 00:15:25,218 --> 00:15:25,418 [Eric] Yep. 00:15:25,418 --> 00:15:32,488 [John] And, and it'll be interesting to see, 'cause there's pieces of, like, your workflow where it truly is, like, personal software, and it doesn't make sense- 00:15:32,488 --> 00:15:32,488 [Eric] Mm-hmm 00:15:32,488 --> 00:15:34,978 [John] ... for somebody in a slightly different role to do it the way you're doing it. 00:15:34,978 --> 00:15:36,278 [Eric] Agreed. Yep. 00:15:36,278 --> 00:15:41,818 [John] But there's for sure, like, skills or things that can be pulled out of your workflow that could be useful for a lot of people, and I don't think that's- 00:15:41,818 --> 00:15:41,898 [Eric] Yep 00:15:41,898 --> 00:15:46,978 [John] ... a solved problem yet, and that's something that we're, that I'm thinking a lot about for, for my team and for our clients. Um- 00:15:46,978 --> 00:15:47,108 [Eric] Yeah. 00:15:48,278 --> 00:15:49,117 [Eric] Okay. Next- 00:15:49,118 --> 00:15:49,188 [John] Cool 00:15:49,188 --> 00:15:58,038 [Eric] ... topic, which, uh, I'm really excited to get your take on this one because as someone [laughs] who's worked in data for a long time- 00:15:58,038 --> 00:15:58,058 [John] [laughs] 00:15:58,058 --> 00:16:08,438 [Eric] ... you are no stranger to Salesforce. And Salesforce recently announced this big architectural shift for their platform, and they said they're going headless. Now, 00:16:09,458 --> 00:16:23,378 [Eric] in the world of, uh, like websites, headless is a very old concept, and it basically means that you separate the backend system that holds all of the information- 00:16:23,378 --> 00:16:23,388 [John] Mm-hmm 00:16:23,388 --> 00:16:34,667 [Eric] ... from the front end, and this allows you to do a number of things. Uh, you know, um, have more control over how content is displayed. I think about this from the world of, like, content and blogging, where it's like- 00:16:35,708 --> 00:16:35,708 [John] Right 00:16:35,708 --> 00:16:48,948 [Eric] ... your content management system is separated from how, you know, the blog that the content is displayed on architecturally under the hood. And so there's all sorts of benefits to that. Speed, you know, design flexibility. 00:16:48,948 --> 00:16:49,038 [John] Yeah. 00:16:49,038 --> 00:16:54,417 [Eric] Uh, WordPress is sort of the traditional, um, monolith you would call it, where everything is- 00:16:54,418 --> 00:16:54,428 [John] Mm-hmm 00:16:54,428 --> 00:16:55,828 [Eric] ... tied together and, you know- 00:16:55,828 --> 00:16:55,828 [John] Yep 00:16:55,828 --> 00:16:57,458 [Eric] ... there are security issues, et cetera. 00:16:58,478 --> 00:17:05,188 [Eric] So I think that's probably familiar to people in, you know, who have experienced the world of content and websites, right? Like headless. 00:17:05,188 --> 00:17:05,798 [John] Mm-hmm. 00:17:05,798 --> 00:17:11,278 [Eric] But Salesforce is a CRM, so what does headless mean in the context of Salesforce or other enterprise software? 00:17:12,358 --> 00:17:13,958 [John] Yeah. Well, I mean, I think 00:17:15,118 --> 00:17:38,917 [John] whether you use the term headless or not, I think the general strategy for almost all enterprise software, CRM, ERP, like whatever it is, you know, HR planning, it's all gonna be this headless strategy. And what all it means is, like, we wanna be the database. We wanna be the source of truth, and then we're going to not force you to use our user interface. 00:17:38,918 --> 00:17:39,018 [Eric] Mm. 00:17:39,018 --> 00:17:54,258 [John] We're gonna have this, um, ability... 'Cause, you know, Salesforce user, user interface I think will be around for a while, but, but you get access to this new little bit lower level layer that agents and applications can access the data 00:17:55,378 --> 00:18:04,198 [John] in, in a format that makes sense for them to build whatever you want and integrate it into your systems and processes, like however you wanna use it. 00:18:04,198 --> 00:18:04,238 [Eric] Yep. 00:18:04,238 --> 00:18:17,868 [John] So it, it's... I would almost think about it, [clears throat] A, like an agent interface into it or programmatic interface, but two, it's also, um, drastically increases the flexibility. So back to the, like, personalized software. 00:18:17,868 --> 00:18:17,878 [Eric] Yep. 00:18:17,878 --> 00:18:38,621 [John] When you give access at that lower layer down, um-You can, um, you can really make these systems fit your company better and your people's workflow, which I think that is the sales pitch. And from Salesforce perspective, and I, I think I could speak for all of these, I think they're taking the approach of like, as long as we have all their data, like, they're not gonna leave. 00:18:38,621 --> 00:18:39,142 [Eric] Right. 00:18:39,142 --> 00:18:39,522 [John] And, uh- 00:18:39,522 --> 00:18:39,652 [Eric] Yep 00:18:39,652 --> 00:18:43,842 [John] ... we'll [chuckles] And some of it, honestly, I think it was like, "Eh, we'll figure out pricing later. Uh, we just-" 00:18:43,842 --> 00:18:43,852 [Eric] Yeah 00:18:43,852 --> 00:18:45,622 [John] ... "wanna retain all these people." 00:18:45,622 --> 00:18:45,652 [Eric] Right. 00:18:45,652 --> 00:18:49,132 [John] Like, "Keep all your data with us. We'll give you that, like, new access." 00:18:50,282 --> 00:18:50,412 [Eric] Yeah. 00:18:50,412 --> 00:18:50,412 [John] Um. 00:18:50,412 --> 00:19:10,472 [Eric] Yeah. I, I have a specific example of this actually that I, I ran across the other day at Vercel. There's a... This is probably gonna be a familiar process to a lot of people, uh, who run Salesforce or a CRM, but there's this constant need to do data cleanup or, um- 00:19:10,472 --> 00:19:10,472 [John] Mm-hmm 00:19:10,472 --> 00:19:18,822 [Eric] ... even just data updates. So a good example would be like, we need to switch this account ownership from one rep to another rep, right? 00:19:18,822 --> 00:19:19,082 [John] Mm-hmm. 00:19:20,222 --> 00:19:33,562 [Eric] And that's typically, uh, it sounds so simple, but in order to ensure that you don't have some sort of collision, right, or you assign the, you know, the wrong account, the wrong account to a rep and, you know- 00:19:33,562 --> 00:19:33,572 [John] Mm-hmm 00:19:33,572 --> 00:19:37,992 [Eric] ... people get multiple emails and you create a bad experience or, you know, you have reporting issues, 00:19:39,582 --> 00:20:00,862 [Eric] that's usually a pretty manual process, right? You have a Salesforce admin, you have, you know, whatever. All, all, you know, and you have approvals and, and all these things. And, um, long story short, our, uh, team on the go-to-market, uh, engine- we have a go-to-market engineering team, and 00:20:02,202 --> 00:20:06,022 [Eric] they built an interface on top of Salesforce Headless for- 00:20:06,022 --> 00:20:06,222 [John] Mm 00:20:06,222 --> 00:20:09,202 [Eric] ... for making those updates that includes a human in the loop- 00:20:09,202 --> 00:20:09,212 [John] Mm-hmm 00:20:09,212 --> 00:20:09,542 [Eric] ... right? 00:20:09,542 --> 00:20:10,372 [John] Yeah, yeah. 00:20:10,372 --> 00:20:17,272 [Eric] So they basically request the update. An agent goes and looks in Salesforce to say like, "Okay, does this make sense? Do we have the right data?" Et cetera. 00:20:17,272 --> 00:20:17,281 [John] Mm-hmm. 00:20:17,281 --> 00:20:26,302 [Eric] "Yes, this change can be made." And then it asks a manager for approval, "Yes, do you wanna make this change?" Right? And it all happens- 00:20:26,302 --> 00:20:26,312 [John] Right 00:20:26,312 --> 00:20:40,002 [Eric] ... in Slack. And so yes, you're interacting with Salesforce, which still drives the reporting, the process, you know, all those things, and is the central repository for information. Um, but the interface is actually just Slack, not the Salesforce- 00:20:40,002 --> 00:20:40,382 [John] Right 00:20:40,382 --> 00:20:41,162 [Eric] ... UI. 00:20:41,162 --> 00:20:57,582 [John] Right. Well, I think that, I think that's an awesome view into the, a trend that I think Headless is going to unlock. Like we're seeing this, people are wanting this from us. So in data land, people traditionally access data through analytics tools. Tableau, Power BI. 00:20:57,582 --> 00:20:57,862 [Eric] Mm-hmm. 00:20:57,862 --> 00:21:13,302 [John] There's newer tools, you know, Sigma or Omni or some of these newer tools. So that is the traditional access layer. As these... And, and then you were, and then in the back end you're like pulling in data from Salesforce. You're pulling in data from wherever, wherever to your, um- 00:21:13,302 --> 00:21:13,502 [Eric] Yep 00:21:13,502 --> 00:21:34,682 [John] ... your database that is then connected to these like front-end tools. Um, I, I think as this like Headless experience continues to get better, like more and more people at least are going to try, and I think it will work in some circumstances, just to directly connect agents to these Headless command line interfaces and, and, um, 00:21:35,902 --> 00:21:38,302 [John] uh, MTP. Like there's, there's multiple ways to do it, but- 00:21:38,302 --> 00:21:38,462 [Eric] Mm-hmm 00:21:38,462 --> 00:21:46,272 [John] ... the point is the agents will talk to the tools directly, and you will be able to get analytics out of it. You will be able- 00:21:46,272 --> 00:21:46,332 [Eric] Yep 00:21:46,332 --> 00:22:07,592 [John] ... to do, like, do admin type steps. And then the big question on my mind for, for data, um, is what happens... And I think that works great for a couple of applications, but when you get to 15 or 20 or a bunch of applications, like at, at least for now, do, do you still need that, that modeling layer that, that- 00:22:07,592 --> 00:22:07,682 [Eric] Mm 00:22:07,682 --> 00:22:51,912 [John] ... that, um, database to basically collect all the data together and have the agent access that and then still give the agent access to like go make an update in, in Salesforce or something? So that's kind of our bet right now is, is it's not quite there if you have a, a, a good number of different tools. But if you, uh, like everything's in Salesforce, if all your data's in, in Salesforce that you need for, for marketing and sales and you really don't have many other tools, or you figure out a way to get all your data in Salesforce from the other tools, like I think that's a legitimate use case where you don't necessarily need, um, depending on your organization size, a bunch of, um, business intelligence layers and complexity with, with, um, storing the data for the business intelligence layer when you're like, "I just wanna talk to the data directly from an agent." 00:22:51,912 --> 00:23:08,582 [Eric] Right. [clears throat] A- and I think that the... So net-net, my takeaway is if you're evaluating software in this era of AI, I would vet for a Headless architecture or plans for a Headless architecture- 00:23:08,582 --> 00:23:09,002 [John] Yeah 00:23:09,002 --> 00:23:15,442 [Eric] ... even just as option value. But the other thing which actually leads into our, to the third thing I wanted to ask you about is, 00:23:16,462 --> 00:23:23,122 [Eric] uh, it's very easy to jump to the conclusion of like, well, I mean, Salesforce is just a, a database, so like- 00:23:23,182 --> 00:23:23,572 [John] Yeah. Right 00:23:23,572 --> 00:23:25,802 [Eric] ... and I'm paying them a bunch of money, and now- 00:23:25,802 --> 00:23:25,832 [John] Right 00:23:25,832 --> 00:23:29,772 [Eric] ... I'm Headless, so we're not even logging in anymore, so why would I use it? And one of the- 00:23:29,772 --> 00:23:29,772 [John] Yeah 00:23:29,772 --> 00:23:40,242 [Eric] ... reasons actually [chuckles] is because they have decades and decades of experience dealing with enterprise security- 00:23:40,242 --> 00:23:40,272 [John] Mm-hmm 00:23:40,272 --> 00:23:44,772 [Eric] ... scale, uptime, all of these different, um, 00:23:45,882 --> 00:23:53,192 [Eric] challenges, right, that are non-trivial. And so one of the reasons you pay them is because they've spent decades doing that and you haven't [chuckles] and- 00:23:53,192 --> 00:23:53,192 [John] Right 00:23:53,192 --> 00:24:06,122 [Eric] ... their system is reliable, which is why they, you know, are still the dominant player in the market, right? And so that also kinda goes to the ne- this ultimately circles back to fit, which is, you know, 00:24:07,502 --> 00:24:15,522 [Eric] is that a requirement for you? And if so, then, you know, buying proven software is important, right? Even if it's- 00:24:15,522 --> 00:24:15,722 [John] Right 00:24:15,722 --> 00:24:21,362 [Eric] ... you know, from a, from a maybe not like as cool of a company, right? Um- 00:24:21,362 --> 00:24:21,902 [John] Right 00:24:21,902 --> 00:24:24,582 [Eric] ... but that can actually be strategic. But what say you? 00:24:26,558 --> 00:24:51,718 [John] So, uh, a couple of things, and actually it relates back to the headless versus traditional. If you go with a startup for whatever you're, you, you want, odds are you're more likely gonna get the headless as being an option. And, and headless I think is hel- helpful to define. Like, people might not call it that. People might... It might be under command line interfaces. They might be calling it MC, you know, MCP interfaces. There's multiple ways to talk about it, um, for headless. 00:24:51,718 --> 00:24:52,298 [Eric] Yep. Agentic. 00:24:52,298 --> 00:25:02,958 [John] But I do think... Agent- Yeah, agentic interface. Um, but I do think startups are more likely to, to have, to have that as part of their product, right? It's a differentiator- 00:25:02,958 --> 00:25:02,968 [Eric] Yep 00:25:02,968 --> 00:25:09,898 [John] ... for them. And enterprise is potentially more likely to, to be, be a little bit behind. But 00:25:11,078 --> 00:25:20,598 [John] I mean, the other thing to think of too is, is, I mean, think about people. If you, if you, um, hire somebody with 20 years of experience versus 00:25:21,678 --> 00:25:25,458 [John] 3 or 5, like what do you expect to get? And out of- 00:25:25,458 --> 00:25:25,558 [Eric] Hmm 00:25:25,558 --> 00:25:35,278 [John] ... the, the person with 20 years of experience, A, like you expect to get, um, knowledge about how things should work in their domain. Like, they just know- 00:25:35,278 --> 00:25:35,288 [Eric] Yep 00:25:35,288 --> 00:25:48,768 [John] ... how sales teams work, and everything's built in for like what you need for a sales team. And I think one of the s- selling points for enterprise is I don't... Especially for a small business, I don't know what I don't know. And- 00:25:48,768 --> 00:25:48,768 [Eric] Right 00:25:48,768 --> 00:25:55,938 [John] ... what if we get this customer that like needs this thing that I don't even know that we needed? Like, I know Salesforce will have it. 00:25:55,938 --> 00:25:55,997 [Eric] Exactly. 00:25:55,998 --> 00:25:59,558 [John] Whereas if I go with a startup, they're like, "Oh, that's on our roadmap," you know? 00:25:59,558 --> 00:26:00,238 [Eric] [laughs] 00:26:00,238 --> 00:26:01,858 [John] So I think that's the big- 00:26:01,858 --> 00:26:02,158 [Eric] Yeah 00:26:02,158 --> 00:26:06,958 [John] ... fear. Um, it, and it dep- I mean, there's so many factors to whether, you know, 00:26:08,078 --> 00:26:14,168 [John] uh, if it's a good idea [laughs] to use a startup or enterprise, but I think that's the big fear on like why you would go enterprise. Then for the startup piece- 00:26:14,168 --> 00:26:14,178 [Eric] Yep 00:26:14,178 --> 00:26:40,897 [John] ... I think if you have a really clear vision of like what you're doing and what you're gonna need out of, again, let's just use, uh, Salesforce as the example as the customer relationship management, what you're gonna need out of that, and you have a really good vision for like, "I know I'm not gonna need all this other stuff," um, there's a lot, there's a lot there to... The cost savings, sure, but like better interfaces with agents like Cloud or, or GPT, um- 00:26:40,898 --> 00:26:41,478 [Eric] Yep 00:26:41,478 --> 00:27:05,858 [John] ... that, um, that is worth considering. Um, and then I think the last piece, startup versus enterprise, is the security piece. Like, I'm pretty convinced that it... Let's call it risk, but like risk and security is a component of risk. Um, the security piece is, I mean, it's just gonna be really, really hard for a couple of years because of how powerful the models are getting at being able to- 00:27:07,018 --> 00:27:07,028 [Eric] Yep 00:27:07,028 --> 00:27:12,658 [John] ... to, to break things. And, and I mean, Salesforce spends millions and millions of dollars on security every year and s- and startups- 00:27:12,658 --> 00:27:12,668 [Eric] Yep 00:27:12,668 --> 00:27:48,828 [John] ... just don't have that kind of budget. Um, now, Salesforce has a bigger service area to secure, of course, um, and they're a bigger target. But I think that's gonna be an interesting one, and, and I think one of the core pieces when I would evaluate a startup is to ask some questions about like what are they building on? Because if, if they've kind of... Um, and Vercel is a great example, like it has a lot of good security features built in. Um, you know, may- maybe it's a different type of startup. They're building in AWS. So part of, part of the things there is just a little bit of vetting from a technical person of understanding h- how they've built it. And y- you, you'll never have- 00:27:48,828 --> 00:27:48,828 [Eric] Yep 00:27:48,828 --> 00:27:56,518 [John] ... perfect, um, insight into their security, but, but that, that can help. And then, of course, are they in the... Are, you know, there's 00:27:57,618 --> 00:28:07,078 [John] SOC 2 is kind of the c- the classic security standard. Like, that helps, but, but just being educated yourself is, is also pretty important of like how are they thinking about security? 00:28:07,078 --> 00:28:11,518 [Eric] Yep. Agreed. Another vector I think a lot about is switching cost, and that- 00:28:11,518 --> 00:28:11,938 [John] Hmm 00:28:11,938 --> 00:28:14,397 [Eric] ... varies even within the same company. 00:28:14,398 --> 00:28:14,468 [John] Yeah. 00:28:14,468 --> 00:28:22,638 [Eric] So how painful would it be... So for example, let's say you, you know, try sort of a hot new CRM startup. 00:28:22,638 --> 00:28:23,397 [John] Mm-hmm. 00:28:23,398 --> 00:28:27,367 [Eric] And you realize it doesn't work. How painful is the switching cost for you? 00:28:27,367 --> 00:28:27,378 [John] Sure. 00:28:27,378 --> 00:28:29,388 [Eric] It's like, "Oh, well, it's not a big deal. Like we have a-" 00:28:29,388 --> 00:28:29,388 [John] Yeah 00:28:29,388 --> 00:28:32,038 [Eric] ... "technical team. We use AI a lot." 00:28:32,038 --> 00:28:32,048 [John] Right. 00:28:32,048 --> 00:28:34,978 [Eric] Like, great, we could jump around and that's okay for us, right? 00:28:34,978 --> 00:28:35,528 [John] Mm-hmm. 00:28:35,528 --> 00:28:35,528 [Eric] Or- 00:28:35,528 --> 00:28:35,528 [John] Mm-hmm 00:28:35,528 --> 00:28:40,558 [Eric] ... like, no, that's gonna be extremely disruptive to our business, right? And so- 00:28:40,558 --> 00:28:41,198 [John] Right 00:28:41,198 --> 00:28:45,337 [Eric] ... if that's the case, you may want to look for a more proven- 00:28:45,337 --> 00:28:45,777 [John] Yeah 00:28:45,777 --> 00:28:54,278 [Eric] ... you know, solution with, you know, again, with, with vetting for things like, um, you know, headless architecture that sort of give you future- 00:28:54,278 --> 00:28:54,287 [John] Yeah 00:28:54,287 --> 00:28:55,848 [Eric] ... future option value, right? That- 00:28:55,848 --> 00:29:06,778 [John] Well, and, and what you said is perfect. Like it's, it depends on the team. So if you're at [laughs] a tech startup and you need a CRM, like yeah, probably like use a startup. I think you'll be fine. But if you're, you're a company that has like 00:29:07,838 --> 00:29:18,978 [John] 50 people in the field that are really non-technical, like it's gonna be a lot of work to train them on whatever you pick. Like, there's way higher switching cost in that, in that case. 00:29:18,978 --> 00:29:26,537 [Eric] Totally. All right. So I'm gonna run through the summary. Build versus buy, give me the quick take. Build versus buy. 00:29:26,538 --> 00:29:43,358 [John] Yeah. Build, build versus buy. Uh, build versus buy, the big thing here is build first, but it's to think about prototype slash requirements even, and then, and then decide do you wanna build for real or, or buy something. 00:29:44,738 --> 00:29:48,648 [Eric] All right. So which actually goes to the headless, uh, side- 00:29:48,648 --> 00:29:48,658 [John] Yeah 00:29:48,658 --> 00:30:00,358 [Eric] ... of things, because you can like sort of prototype on, um, systems that are headless. But headless is sort of mandatory now, would you say, as far as evaluating software? 00:30:01,818 --> 00:30:21,238 [John] Yeah. Yeah. I, I would weight it very heavily, um, that, th- that they have the features already. Um, I would say have a plan for the features. Like, A, it depends on what it is. Like if it's a great fit for your business and like it's coming soon, maybe that still makes sense. If they have no, [laughs] no plans to do it at all, I would, yeah, I would seriously question whether that would be a good decision. 00:30:22,678 --> 00:30:31,787 [Eric] All right. And startup versus enterprise. Maybe for this one, uh, let's say... Let me ask you this question, 'cause this is a great summary. What, 00:30:32,858 --> 00:30:38,598 [Eric] uh, what software have you chosen to purchase as like enterprise software versus startup software at your business? 00:30:38,598 --> 00:30:44,268 [John] Yeah. Um, so we went startup for project management, Linear instead of Jira. 00:30:44,268 --> 00:30:44,438 [Eric] Mm-hmm. 00:30:44,438 --> 00:30:55,228 [John] Um, mainly AI and agentic features. And then we went, I guess you could call it enterprise, like 'cause Slack is kind of an enterprise solution, versus let's say Discord, which is a little more- 00:30:55,228 --> 00:30:55,228 [Eric] Mm-hmm 00:30:55,228 --> 00:30:57,018 [John] ... startup, like techy focused. 00:30:57,018 --> 00:30:57,328 [Eric] Mm-hmm. 00:30:57,328 --> 00:31:08,418 [John] And that was security, stability, integrations, talking with our, you know, a lot of our own companies we work with are on Slack. Um, so those would be the two that come to mind. 00:31:09,538 --> 00:31:15,078 [Eric] Awesome. All righty. Well, thank you for joining Token Intelligence, and we will catch you on the next one. 00:31:22,458 --> 00:31:26,098 [Eric] [upbeat music]