On the other side, everyone's like, 'Oh no, all jobs will be fine and technology always creates more jobs.' And honestly, nobody knows. Probably both those extremes are wrong, but nobody knows how it's going to work, or how it's going to work over time. Maybe jobs are lost and new ones are gained, but it takes a while for the generations that were in between—they could have lost career opportunity. When you talk to college students and recent grads, they're not as rosy. Take a look at all the recent speeches by AI people at graduations—it's not all roses. My rising ninth grader is going to a private school, and there was an open house for the new students. They had artwork from all the existing students displayed—beautiful art. This one student had an artwork that just said 'F AI' over and over again in every part of it. Then they had a talkback where all the artists were on stage explaining their work, and somebody said to that student, 'Hey, explain what you meant by your work.'
Right. 'F AI and anybody that works on AI is evil.' I never felt more unsafe in my life. I was like, I got to leave this. Oh, they better not do any LinkedIn searching on me. But the point is, yes, it's not as rosy—at least emotionally—for younger people as it might be portrayed for a lot of people in Silicon Valley. Nobody knows what the truth is, but we need to have the conversations, and that's basically what New York Foundation is about.
It's interesting. Yesterday, Megan and I did a live show, and this woman came up to us afterwards. She's like, 'How do I get Gen Z folks in my company to start embracing AI? Because they're so anti.' And then she also had kids in college who were rejecting it. So I have a bit of a contrarian position on it, which is that I actually think it's really good for kids to be anti-AI. The reason is because I think there's some sort of bodily reaction—they need to build competence before using AI. If they were to just jump straight to AI and use it for all of their work and their thinking, it would be very easy to just offload every single bit of learning to AI. And there's just something in them that's saying that would be really bad for me, so I'm just going to say AI sucks altogether. Like, I just decided that McDonald's smelled disgusting—and that was good for my body to believe that. Obviously, there needs to be some sort of reckoning when they meet the work world and there's expectations and they need to compete. So bridging that gap between, okay, build confidence and competence and stay away from AI, but then you've got to leap into it—I love this concept that's kind of what it sounds like you're doing. I like your interesting counterconventional thinking there. I do get it—from a fear perspective and from what you're saying about there being some element of developing your own self. It's a weird thing where we can't generalize to everyone in the world based on what entrepreneurs do. Yes, entrepreneurs all use AI, but we're also people that have already gone through all the learning. That's a common Silicon Valley mistake—you take what people in San Francisco think in the tech world and assume all humans need exactly that same thing. In fact, Ben Thompson on his podcast said just yesterday that one mistake people make is assuming everyone is obsessed with productivity. The truth is, most people are not. They like wasting time. They want to watch TV, be on social media, chill and have fun and have a good life. There's a big mistake we make when we assume everyone is like us. I had that realization—this is turning into mom therapy, but I had that realization maybe about five years ago. I have this big job, I have no time, and I carve out a day to spend with my mom. We were doing something together, and I realized there was a very significant tension—because her intentionality was to waste as much time as possible from every possible task in every possible way. And I'm thinking, 'Oh my god, I have exactly one day off. I have this massive job, I'm always traveling, and I'm trying to spend time with you. And the last thing I want to do is waste every possible moment being the least efficient in our day as possible.'
We need efficient bonding time.
Exactly. We were just driving from here to there and then back again, as opposed to planning our day to make any sense. I was really struggling—do I just take over the reins and say we're doing it this way, or do I relax and let us do it the dumbest possible way? I realized I was completely missing the point. The point is we're just spending time together, and so what? We're spending time in the car going back and forth. Fine. Just relax about it. That entire thing about presuming that other people have your orientation is always dangerous, and it's so hard to remember somebody else's lived context is different than yours. It's not malicious—it's just really hard.
It's so true. I fully endorse everything you said, except when my mom spends like 10 minutes trying to order food at the restaurant. That's where I draw the line. There we need more productivity and efficiency. I will just do the ordering. The server is making money on tips—we need them to get to the next table.
So, I want to go back to customer success a little bit, since we have the father of customer success here. Maybe that could be your identity.
I'm glad you said that. Some people said grandfather. I don't like that.
Oh, yeah. No, no, no. You're not old enough for that. Sorry. Not cool. So there's been a lot of narratives about forward-deployed engineers, and you've written extensively about it. Meg has a great concept that she'll probably get into—we've just started back at the EDS model again, so we'll have her talk about that too. But I just wanted to pick your brain a little bit—is this something new? Is this something old? Is this the new face of customer success? What are we really talking about here, Nick?
Yeah, just for people's background—the term 'forward-deployed engineer' was kind of coined by Palantir, because they were selling software to the government and to the defense department. The term 'deployed' comes from the military. The software defense space is such a complex area that they needed to customize it per organization. So they would have a team of two people, from my understanding: one that really understood the military processes, one that was an engineer, and they would actually modify the code per customer in the early days, and then eventually merge it back into the main branch. That's heretical for traditional software companies, where it was like, 'We have one product, everyone uses the same thing.' In SaaS, that was the whole deal. You might have professional services to configure it, but you wouldn't change the code. The issue with that is software became pretty good for everyone but not great for anyone—that's the general experience most people have with enterprise software. Now you start selling AI, where it's infinitely configurable, but it also requires you to reconfigure not just the product but actually your business. To really use AI well, you've got to rethink your business processes, how you work, and your team. If you just give somebody cloud code or tokens from OpenAI, they can get somewhat far, but often they build things that don't get used, or they don't really change results and revenue doesn't go up. What you need to do is get them to change their business—to really operate differently. You see this starkly when you compare an AI-native startup versus a big company in the same space. The native startup will literally be generating 10x more revenue per employee, because everything from day one was designed around AI, and the bigger company has a hard time changing. So people said, 'Palantir did this thing, maybe we can apply it to all AI companies,' where we can have forward-deployed engineers—which the term 'engineer' is a bit of a misnomer. It could be a business consultant helping with the process, like a McKinsey type, or an actual engineer, or a little of both. They go to the client and say, 'You've bought our AI software. We're going to help you redesign your processes and configure the product to really take advantage of it.' Every agentic software company has adopted this mindset. Is it customer success? In the most generic sense, yes. The original definition of customer success was, 'We're going to make sure the product helps you get to your business outcomes, and we'll do whatever it takes.' Now, CS people historically weren't always super technical, and they were often assigned at a high ratio—one CSM might have 20, 50, or 100 customers. No matter how good they were, they couldn't go deep. FDE is much more intense—smaller number of clients, more technical. Also, because many AI companies are paid based on token spend and consumption, it's in their interest to get maximum deployment and value. In a SaaS model, the challenge was the contract might not renew for three years, so you'd give the bare minimum to maximize profit margin. Now the vendor has a financial interest. AI creates a compelling event because you have to get them to really change, and you're bringing in more technically-oriented people. Those are the changes that make it different.
Yeah. And to some degree, it's also just maturity, right? Everything's so new—there's a lot more to be educated on, a lot more to learn. One thing I hadn't really heard but makes complete sense is that it is changing business models rapidly—meaning how work processes are done. Whereas with SaaS, it was really about fitting existing processes into new technology. With AI, you really need to rethink the business processes. So you do need a lot more handholding potentially to do that.
Exactly. Before, the actual handholding and business process work was just on the customer. They also had to do it, but back then they would change through ops people and change management—internal stuff. And by the way, in some cases they just wouldn't change, and they wouldn't use the software well, and then the users hated the software. That's how we got to where we're at. It's a new relationship between a vendor and a customer in the AI world.
That's exactly right. And the other thing tied to that is these are such new paradigms. People say AI is non-deterministic, meaning it doesn't always work the same way every time. There's all these new things—conventions, agents, models—and there's so much that people are like, 'I don't know where to start.' Most people have this experience as consumers now—where do you start with AI? What's the best way to do this? And so somebody that comes in and says, 'Yeah, I've done this for three other companies,' gives you the confidence. You're totally right—confidence is a big part of it. Another thing you just mentioned was the economic incentives, and that gets back to budgeting for tokens. I know that's also another company you're deeply invested in, and something we've been talking about on the show. So I was wondering if you wanted to talk a little bit about that.
The company you're referring to is called Laridin. L-A-R-R-I-D-I-N. I joined the board in February of 2026, so just a few months ago. It's backed by Andreessen Horowitz, and the two founders are people I've known for a long time—great entrepreneurs. They decided to work in this problem space of companies needing help to get the productivity benefits of AI and also manage the costs. What's been interesting watching the journey is we've gone through about three eras. One era was when agentic coding just started, but a lot of the pricing was still in a subscription model. There's a model called Opus 4.5 from Anthropic that really opened up agent coding. Somebody's using Claude Code or Codex, and they're like, 'Awesome, use it as much as you want—200 bucks a month, that's affordable.' Then the model providers were like, 'We need to switch to charge per API call because it's just exploding.' People assume token costs would just plummet over time, and they haven't. Then we get to the second phase—what people affectionately call 'token maxing,' using the Gen Z 'maxing' term. The idea is, how do companies really adopt AI? Let's get everyone to build stuff and use as many tokens as possible. Create leaderboards of how many tokens each engineer used—that's the measure of success. Pay bonuses based on that. The first phase was about December 2025 to February 2026, and the token maxing period was February to about May 2026. Then CFOs started seeing the bills. They're used to budgeting with fixed budgets, and this wasn't budgeted. So the third phase is a more rational balance—you want adoption, you want productivity benefits, and you want the right cost profile. It's a business. Laridin does both sides. They measure AI productivity by looking at what you're doing with AI versus without AI, and what processes could be AI-enabled—that was their first product. Their second is: how many tokens are you spending, by model provider, how can you reduce the cost? You can send simpler workloads to cheaper models or open-source models. The economics of tokens is what this company is about, but it underlies a bigger story: AI in enterprise has to be about business, not just technology. Fortune 500 CEOs can get away for a year of 'Hey, we're investing in AI,' but then shareholders are going to ask about earnings per share and revenue growth. CEOs are going to start getting those questions very soon, and we need to take this much more seriously from a business perspective.
