Stop Token Maxxing: Find Where AI Actually Pays Off | Jiaona Zhang

Jiaona Zhang 0:00 Here's where I see things go wrong. You have executives at a company saying, everyone must use AI. And oh, by the way, AI is now in your performance review. Yes, of course, it creates fear because everyone's token maxing because they're like, I've been told to use AI. And I'm like, no, you should use AI efficiently. If you use AI to go like redo the font on this deck when you could have just clicked a button, it is not an efficient use of your tokens. But if you used AI and you took something that took humans five hours per week, an agent did in five minutes, that is Time back.

Conor Bronsdon 0:35 Companies are spending billions of dollars on AI tooling, but tracking ROI is proving challenging for many. We need to solve this gap, and that's what we'll be discussing today on the Chain of Thought podcast. I'm your host, Connor Bronstein. Today's guest is Jay-Z, CPO at Laurel, which she describes as test subject one, a company using its own product to understand exactly where AI is and isn't working in every function. Before Laurel, Jay-Z built products at Airbnb, Dropbox, Webflow, Linktree. She's been teaching product management at Stanford for nearly a decade now and has a clear perspective on how today's organizations need to change, not just what AI tools need to be added on top. Jay-Z, great to see you. Welcome to Chain of Thought.

Jiaona Zhang 1:24 Thanks for having me, Connor. Excited to be here.

Conor Bronsdon 1:26 my distinct pleasure. It's always fantastic to catch up with you and talk to experts like yourself. So I'd love to just dive right in and start getting the benefit of all your knowledge. Why do you think most organizations are struggling to understand where AI is actually working for them? And what does it take for them to actually find out?

Jiaona Zhang 1:47 It ultimately comes down to visibility. So when I think about this, you know, all of us have our cloud accounts, our chat GPT accounts, you know, whatever tool of choice you're using in your organization. And we're all doing our best to figure out AI. And I think that if you really think about AI adoption, a lot of times people are just running at it, thinking, hey, I need to go, I need to do more. I need to figure out how to take the thing that I just did yesterday and I want to like AI-fy it. But what I don't think people are doing enough of, two fronts. One is really understanding where there is efficient gains to be had. So where is your real time that can be brought back? Where is your real revenue that can be driven? And how do you actually go do the workflows that will actually achieve those goals as opposed to, well, it felt really good. It felt really good when I asked AI XYZ things and it came back with something. But very few people are actually looking at the quantification. And this is getting more and more important as you think about You know, tokens today are probably the cheapest that they're ever going to be and the cost is going to go up. And so you're really thinking about this calculus of, you know, where do I actually spend the money on the tokens and where do I not? And you do, therefore, want it to be efficient. And for a lot of people, they don't have a way to really measure that.

Conor Bronsdon 3:14 I do want to push back slightly on this narrative that tokens are certainly the cheapest they're ever going to be. Like, yes, there is subsidizing that's happening. But are we certain that there aren't going to be major advances around efficiency that enable tokens to be much cheaper in the future? I do think we're gonna end up spending more of our time because we'll be using a lot more of them. But I guess I'm not 100% convinced that the per token cost is as cheap [3:39] Jiaona Zhang: [OVERLAP] Yeah, [3:39] Conor Bronsdon: [OVERLAP] as it's ever going to be.

Jiaona Zhang 3:40 I think that's fair. I don't know if the per token cost is the cheapest, but I think that today a lot of organizations are not thinking about their token spend regularly and they're getting hit with their bills. And it's very hard to even understand, well, I basically, you know, equip my team to go run rogue and they can do whatever they want. And really the mandate is very much like use AI everywhere. [4:05] Conor Bronsdon: [OVERLAP] TokenMax. [4:05] Jiaona Zhang: [OVERLAP] I think that's a mandate at, yeah, token max. And the measure of success often is are you just doing it versus are you using it efficiently? And so I think what I'm really trying to get at is you're going to pull into this efficient frontier really quickly.

Conor Bronsdon 4:20 Yeah, totally agree with that. And I think we're seeing some headlines around this, right? Where companies are suddenly saying, oh God, we spent how much last month? We need to adjust. And there's definitely a difference between, let's call it the LinkedInification of AI, where people are like, oh, I rebuilt my workflow with AI and look at how cool this is versus, okay, is this actually delivering ROI for us compared to token spend, compared to time spent? And it can be challenging to do this identification. So I'd love your perspective on how companies should start. I mean, let's say they've just been in this experimentation phase that maybe they're now in production, but they're not necessarily measuring the return on investment for them. Where should they start?

Jiaona Zhang 5:03 I think really around visibility. And so how do you get visibility in terms of your spend versus the outcomes that you're actually trying to drive? And so this is actually taking the time to go and quantify what are those outcomes? Are you trying to drive revenue? Are you trying to drive time allocation efficiency? What is it that you're really trying to drive? And I think one of the big things that people aren't doing enough of is taking the time to even articulate what that is. And so going back to, you know, started this a little bit with this idea of Laurel and being, call it test subject number one. So maybe let's just take a moment and talk about like, what is Laurel and how can we be test subject number one? You know, the way to think about Laurel is, you know, we're this desktop agent. that is able to capture what you're doing digitally. And as a result, when you think about the applications, one really amazing application that a lot of companies use today is this idea of timekeeping and time tracking. So how do you take all the work that you're doing and instead of the lawyer spending the hour at the end of their day trying to really map out, okay, this is for every six minute increment that I've worked, how do I summarize? my work product to a client and into an invoice, we can do that in such a way that is really seamless, right? And so that is one application of how do you take time data, essentially, and use it in a really, really useful way and be a real painkiller for customers. Another application of it is if you know what you're working on, you're able to then take that data and say, OK, well, I know, you know, I know what applications I was in. I know what I was doing in those applications. I also know the general workflow of how I'm going about getting my work done. And when you have the workflow and you have the time against those workflows, and you also have the end outputs, you're able to take all those data points together and get a really good picture of, okay, well, am I actually spending my time in the right ways? Am I spending my time on high leverage work? Or am I actually doing really low leverage work that could be automated? And so that is a little bit of the context here of how can we use our own technology to really understand how how tools are being utilized across our organization and really understand the efficiency of AI adoption and AI usage.

Conor Bronsdon 7:24 It's so interesting that we're I think rapidly having this conversation around adoption efficiency and ROI because I would say it's something that's been a theme of the last call it 25 years or at least the the aughts into the second decade of 2000s where we saw I mean the origins of Salesforce and CRMs really upgrading themselves. And this question in many businesses of why are we doing XYZ? Is it giving us the ROI we wanted? We've seen major focus from businesses working to identify how effective different activities are, whether that's the data and attribution push in marketing and sales, which is a stellar example of this for anyone who's been on the go-to-market side, but also the software engineering layer with software engineering intelligence, DX for engineering teams, something I worked on quite a bit at Linear B. There's plenty of conversations around this. And now we've added this new input of all the tokens we're spending, these new workflows, And in many ways, it is a proxy for our digital employees that we've created, these agents that are going out and running workflows for us. Is it really many of the same problems just now applied to a new technology that is emulating what humans were doing or adding to human workflows? Or do you see this as something where it's a very disparate problem that needs to be addressed in a significantly different way?

Jiaona Zhang 8:53 I do think there are a lot of parallels. I just think that the need to quantify has been greater today than it has been in the past. And so, for example, in many ways I think of agents today are just extensions of the workforce, right? So you used to have 10 humans working on something, and now you have, let's call it, five humans, but each one of them has hundreds of agents. It's just an extension of that person and what it is that they're doing. So I think that the core principles around, it's really interesting, Laura, we talked about this concept of there's a thinker, or there's an owner, or there's a captain. We use the word captain actually quite a bit. And then there's all these other people, and all the other agents that are extensions of that captain to get that particular initiative done or to achieve that goal. And so when you think about it like that, there are a lot of the same parallels. You used to have just a lot of humans working together on one team. And then it was like you had a lot of humans plus a lot of tooling, software working together. And now you have a lot of humans and you have some software and they have agents, which is a type of software, sort of. Right. So I think of it as an extension of some of that. What I do think is really interesting is like, again, what is human versus what is like more of an agent? Right. Humans, I really think about you have to provide them context, clarity and motivation. Now, of those things, what are the same in an agent? You still have to provide them context. You have to provide them clarity. In fact, I think that context and clarity are even more important because, you know, you don't have the human judgment piece to it, but you probably have to, I mean, you do have to think much less about motivation. You know, agents just run all the time versus when you work with real humans, they have real emotions and, you know, interpersonal working dynamics, which are all really, by the way, really important. I'm a big believer of you have to preserve what is human and really amplify what is human and let humans do the best pieces of that versus you actually all of us want to take all the things that were really tedious and full of drudgery and get that off our plates. Right. And so a great example is if you think about the sales process. There's so much of the sales process that is 100% tedium, drudgery. You're doing the same thing over and over again. To your point, I think you brought up a CRM. Nobody is thrilled to be entering data into a CRM. And often you actually can't get the right data because no one has done it because it's so annoying. And as a result, rev ops can't function. There's so many problems. What humans want to be doing and what is uniquely human is, you know, reading a room. physically in person, knowing when a customer relationship is worth the revenue. Or maybe there's a trade off that has to happen that's like intrinsically human and has all these like complex pieces to it. That is what you want humans to be spending their time on. And what you want, you know, AI to be doing is all the other pieces where no human was getting true joy out of doing that. And so to me, it's really just the shift, the shift of work, the shift of time. And really excitingly, I think when you think about it in a very systemic way, you're able to really optimize a system to get your output and get it faster.

Conor Bronsdon 12:18 so many good points in there. One, I love that you brought up this idea of humans and agents working in parallel. It's simply where things are going, it's where things are today. But to your point around motivation, the other side of this is You know, even if we were hiring more headcount that we needed to in certain points in the past, it was a lot harder to scale up that headcount. Like humans are expensive to hire, yes, but they're expensive in time to hire as well. Whereas spinning up an agent today is exceedingly easy and we can create hundreds of agents that are going out and spending the company's money through tokens, even if they're not touching financial transactions or anything else. So it was a lot easier for that spend to just balloon very rapidly without very little oversight, particularly because of how the contracts are set up. So I think that's a great point about why we need to have more focus here. I also love that you brought up kind of the leadership and management skills that are required for humans and for AI agents. Because while, to your point, there is less motivation needed today, though maybe you can do a little prompt motivation, there's so much import to context and clarity and then that is like such a crucial management leadership skill. So there's I think actually a lot of advantage for some leaders today who go out and say, OK, let me go work with a bunch of agents and extend my work and extend my team because they have some of these skills, which I I'm curious if you've been experiencing that with the captain approach that you mentioned.

Jiaona Zhang 13:43 Yeah, I say this often, which is I think sometimes the best users of AI and often, you know, the best builders of these like agent fleets, whatever you want to call it, are previous managers in the sense that, you know, as a manager, you're really forced to think about orchestration, right, delegation. How do you get all of the pieces to come together when you cannot do every single piece? There's a saying that a lot of people have for their reports who transition from being an IC to a manager for the first time. And it's, you know, what got you there won't get you to the next point. And really a lot of it is what excellent ICs are good at is doing the thing. And then when they get to a manager, they actually have to like delegate the thing. Now, what I think is really exciting is you can, actually today, I do believe that you can't be in a world where you're just like, I'm hands off. And in fact, I even think about how I construct my team this way. You know, I'm not hiring, like there are no directors. There are no people who are just like, I'm just like here to manage. But the art of management that you have to build through practice is really helpful. And when you're thinking about, well, how do I go even audit my system? Right. And how do I go build my agents and delegate to my agents? And how do these agents interact with each other? And how do the agents interact with a human? So all of that skill set is really, really valuable. And so that's why I do believe in, you know, the management piece, at least that mindset gives you a really strong leg up. But then ultimately, it's all about full accountability, which goes back to what I was saying earlier on captains. The reason why we love the term captains and we use it very often here at Laurel is because the idea of a captain is you know, that accountability, right? Like, there is one throat to choke, one person responsible. And what's exciting now is you can say, well, okay, if I'm the captain, I want to get from point A to point B as quickly as possible and as efficiently as possible. How might I do that? Should I be shipping code myself and all the All our PMs here, all our designers, ship end-to-end, features end-to-end. They are shipping to production. They're not just prototyping. And because when you think about, well, that's really efficient. I have the customer context. I have the deep understanding of the problem. And I'm able to take that knowledge and I'm well versed enough in design and all the other pieces that I can take this thing end to end. And wherever I get stuck, this is where again, LM's are excellent at answering your questions and filling in the holes, especially when you have kind of the scaffolding. And so that is where you can be so much more productive.

Conor Bronsdon 16:25 Yeah, I think this idea of enabling individuals within the org who have deeper context on a problem to go work with agents to solve that problem more directly is such a fantastic benefit of this current AI-enabled era. And it makes me think about a really great piece I read from James Stanier a few years ago about the importance to executives of staying in touch with the bare metal of the company. So actually talking to customers, actually getting out there and understanding what's happening on the front lines. And he positioned it in this piece, and I think it was very relevant at the time, this was like three, four years ago, of, okay, well, You need to be actually ensuring that you are diving into the details and context so you don't lose touch with that, with what's actually happening at the edges of the company. And at the time, I think for executives, this was much more about like, okay, like, let me understand the sales process. Let me understand what our customers actually need. This is still very important. But there is now an opportunity for decisions to be made at those front lines, I think, because you can delegate more authority and execution capacity to these individuals to say, OK, yes, go run. You don't have to go work with a PM and a whole team of engineers and a designer. One of those people could potentially ship a feature, to your point. And there are pros and cons to some of this. Stakeholders still matter. I'm not trying to say we need to get rid of all stakeholders all the time, though I think there's open discussions around org design that we need to have. And I guess my point is, how are you thinking about these different functions now as we move into this era where a PM, a designer, or an engineer can all ship a feature, or a director or executive can ship a feature if they think it's valuable? How does this new team structure look? Is it a lot of skilled generalists who are seeking context and managing agent fleets? Are you getting rid of these, these management layers? It sounds like what's the approach you're taking.

Jiaona Zhang 18:23 I'll talk about a couple of principles here. And so the first one is this idea of being really lean, right? And I'm a big believer that, you know, constraints will breed creativity. And so when you actually are lean, you push your team to really think through, well, how might I get this done with what I have, as opposed to, okay, well, I could ask for X, I could ask for Y. So I think that's one piece, which is this principle of leanness that is really important. The second principle here is, how can one person, and this goes back to leanness, take something as far as they can go? And the reason why I think that's really important is because that actually helps you understand kind of your efficiency frontier. If you are able to say, look, you know what, my system is clean enough that someone can come in here and ship this feature end to end without being deeply technical and have to go and debug, some of the tech debt that's in here, then you know your system's really clean. I think that's a great litmus test to say, is my system ready? And what you're seeing a lot is you're seeing a lot of places, a lot of products where parts of their tech stack are not ready for agentic development just because of the way it was built over the years. And so you see a lot of companies have to take time to really kind of clean up that tech debt so that agents can actually be really effective on the code base. And, you know, we do this often. I have this analogy I use with my team. It's around this idea of, like, gardening. And so, you know, is there a part of the garden that needs actual work to be done? And then in the future, when someone else comes in and wants to, like, go plant something, they know, okay, well, the daffodils go here, the plants go here, the trees go here. then it doesn't feel at the end of the day like a chaotic mess, right? And so it's really interesting this idea of, okay, well, where can you play? Where can you go from A to B almost end to end? So I think that's the second principle of like, if you take something end to end, that's what you're tasked to do, you will hit up on the constraints. And as a result, you'll have a very systematic way to know what you actually have to burn down in order for this to be a truly functional system. So that's like the second principle here. And then this last principle is this idea of like, what is your expertise and how do you offer that expertise to the organization in a way that's scalable? And again, I think all of these pieces come down to two concepts really, which is like, you know, efficiency. and systems, right? And so when you think about this piece, the best designers, their jobs, they should start thinking about their jobs as how do I create the design system? The same way a lot of people are talking about this, have been talking about this in engineering, it's going from I'm building the feature to being like I'm building the system through which anyone, whether it's a PM or someone on customer success, sales, support, whatever it is, they can be shipping. That's the new mindset. So now they're starting to shift into how can I be an architect versus how am I coding, you know, quote unquote, by hand every single day. And that is the shift where you're thinking about systems as opposed to I got to get that thing done, I got to get it done one time. How do you create the environment in which doing it again is much easier?

Conor Bronsdon 21:48 Let's drill down on this environment and architectural question a bit because I think the idea that even your CS team members or really anyone in the organization can ship code is such an interesting one and one that is evolving for the top companies in the world. I mean, we saw Block make this decision last year with shipping AI agents to 12,000 plus employees. Great episode on that that we put out earlier this year for anyone who wants to check it out. Obviously, your org's doing it where everyone has the opportunity to ship a feature end-to-end. How do you enable team members who aren't highly technical to do this work? What are you doing internally to set this up?

Jiaona Zhang 22:29 Yeah, it's interesting because there is the, how do you get to what I call like a level one, and then how do you actually get to more pro mode? So the level one is literally giving people tools, right? So if you're like, okay, I, you know, it's really difficult obviously for a CSM customer support, you know, specialist who doesn't, sorry, customer success specialist, right? Who doesn't actually know how to code. How do we get someone like that to be able to build something? ship something, ship a feature. The first thing is just tooling. So, okay, what is out there? Are you using, you know, back in the day it was Cursor? We use Devon frequently. Some people just go ahead and use Cloud Code. Whatever the tool of choice is, I think that's step one. That's like level one. But what we saw is, OK, we did enablement. We enabled the team to ship. We taught them how to do it. And there was a spike in usage. And by the way, incentives are also really important here. And so when you say, hey, the first person who ships something is is rewarded, right? Like that motivates behavior. But more than that, I would say that the motivation honestly is quite intrinsic. So being able to understand what is the thing that was frustrating this function for a long time and how do you let them feel freed from that frustration is a big part of how you even do some of the system design. So for example, again, our customer success folks are amazing, excellent. They are on the front lines. They know exactly what our customers are feeling and wanting. And so the core, besides just being like, here's an incentive. The first person who ships gets a prize, which does work in the very short term. The core motivation is I am sitting with a customer. They're, you know, they're confused by something or they really wish this thing was in the product. How might I enable that and create this moment of delight and usability? That's the core motivation that really gets an excellent customer success person going, right? So understanding that you're tapping into the core motivation, giving them the tools is almost step one. But then what we figured out or we learned is We have to then create a system that makes this something that is doable on an ongoing basis. And what we found was, again, we had a spike of people who were shipping. And then very soon we ran into the next step, the next bottleneck in the process. And again, I think a lot of systems design is literally you're like, you design the system, you run stuff through the system. There's a bottleneck, you undo the bottleneck, and then you find the next bottleneck in the system. And that is literally what you were doing all the time. And I think engineers today probably feel that way quite a bit. But going back to the situation where you're having non-engineers ship, what we quickly found was the next bottleneck in the process was, well, what constitutes a good product decision or a good design decision. So you have everyone saying, well, I want to add this and I want to add that. you're gonna have a very kind of like disjointed product. It's not gonna feel cohesive. It's not gonna feel high craft. How do you actually, what do you do in that scenario? And so therefore, that's when we went back to, again, systems thinking, okay, well, the role of the PM is to be able to enable anyone to understand where can they play and where can they not. Again, same garden analogy. So in the parts of, our administration backend where it's really clean, really clear. It's easy to add a toggle here and there. And we believe that there are new toggles to be added, right? Or like it's easy to add integration and there are integrations to be added, go for it. In the parts of the product where we're doing a pretty big overhaul in the core usability of it, if you're adding a little thing here and a little thing there, it's just gonna get more confusing. And so being clear on like the rules of engagement, being clear on, okay, like, let's actually take the time to articulate the system. That is a really key component. And then finally, going back to what we were able to do at Laurel, which we started the conversation on, is we can actually look at the time that is being spent on any given task. And so we can say, well, you know, what was really confusing for the CSM is this step in the process. When we said you have to do this in order to ship the thing, this caused a lot of friction. And we saw that so many people got stuck on the step, or they just kind of stalled out here. Being able to look at that data and then say, OK, well, that data point is what we're going to bring back into our system to go fix it, that is really, really helpful.

Conor Bronsdon 26:59 It's fascinating to see this transition occur where initially I think a lot of AI automation was just in mostly structured work and often highly structured work. And now we're seeing it start to extend to some unstructured work and get used as an extension for humans who are doing that unstructured work. And we're also seeing at the same time the way teams are constructed shift to account for this. I won't pretend to know where the future is going on this, but it's very clear we are in the midst of this massive shift in how work happens and that it is not yet over. We have quite a ways to go, I think, in all this, particularly if some of the predictions that we alluded to at the start of the show around token usage and potentially costs in one direction or other come to fruition. We're going to see people having to continually reimagine and rethink their approach to some of these workflows they've developed, especially if they were built, as you put it, without much consideration for cost. So we've talked throughout this conversation about re-architecting teams for this new era. What would be your advice to leaders who are listening to this and have maybe started the process of re-architecting their teams or are realizing they need to about how they should think through re-imagining their team for an AI-enabled way of working?

Jiaona Zhang 28:30 Yeah, I think there's a lot of, so this idea of how do you change the way you work, I think there's two pieces to it. The first one is whenever you do any kind of change management, you kind of want people to get to the first step as quickly as possible. And with that, I really think about how do you just pick one workflow? One thing where you go and you ask someone, hey, what is the one thing that you spend a lot of time doing that you really wish you didn't have to do? This great analogy of timekeeping. I wish I didn't have to timekeep. That's really tedious and mundane. Or I wish I didn't have to enter things into a CRM. Or I wish I didn't have to move pixels around here or whatever it is. I wish I didn't have to do unit testing. So what is the one thing where you're spending a lot of time And you go to that that org, that person, whatever it is, and you say, well, help me understand what that looks like. And, you know, if you have an ownership mindset, you really can actually sit down with that person if you're AI, you know, PILD, ANative, whatever you want to call it, you can actually take that thing that that person is really struggling with and build them a better workflow. So I really encourage that on the bottoms-up approach, meaning you go out, you understand the one thing, the one workflow that is really time-consuming, and you automate that, you make that better, and you get people to start to experience a mindset shift where they say, oh, I actually should embrace this because why would I Like, why would I do the thing that was taking me forever over and over again? This feels great. It feels great for this work to be taken off my plate. That's how you get the mindset shift to happen, you know, for everyone that you're trying to drive in this direction. Simultaneously, I think you need to go from the top down. So that's the bottoms up approach. And then from the top down, the way I think about it is you have to understand your ontology of work. Or put another way, where are you spending your time? And

Jiaona Zhang 30:23 the way I would think about this is at the end of the day, all of us can say the shape of X function looks something like this. And now that's not static, meaning that's going to change. And in fact, the way the technology is going, that's probably going to accelerate faster than it's ever accelerated before. I would say for many people, the workflows In law, for the longest time, the workflows in law look the same. And the workflows in accounting look the same. The workflows in consulting, private equity, they all look the same for a very long time. Now with AI and just the capabilities of these LLMs, a lot of people are saying, hey, well, I don't need to spend my time writing this deposition anymore. I don't need to spend my time doing this very complex thing for your taxes anymore. This part I can automate. Now, this other piece I can't automate. It really requires my judgment. right. And so there's from the top down it's really important to look at your ontology of work to really understand what do we believe are all the buckets that comprise what I believe this function should be doing to drive the outcomes that we want. And then we can say OK we're going to in mass try to automate this whole chunk. And when you do that bottoms up and the tops down together, that's how you're going to really drive behavior change. So that's the advice I would give when it comes to how do you get an organization to start to move and to really become much more native AI first and just adopters of these AI tools. And then simultaneously, when you think about the makeup of the team, it's really around the shifting of the work, the shifting of the time and energy spent from low leverage to high leverage.

Conor Bronsdon 32:09 One of the problems with this that we're experiencing is going back to the leadership piece we brought up earlier, where, look, as leaders, folks have been spending a lot of time on clarity, on enablement, but also motivation. And now we're moving to an era where we're not having to do as much of that motivation work when it comes to agents. But I feel like there's a challenge where team members, our human team members we work with day to day, are struggling sometimes because not only is there this external pressure of AI is going to take your job and you have to do X, Y, Z to stay ahead. So just, you know, there's this external stress that many folks are feeling. But there is this pincer movement. So I agree with you in your perspective, like change needs to come for a successful org transformation, probably from the bottom up and top down to actually build an AI-enabled team. But it means there's a pincer movement on folks who are not prepared or are feeling uncomfortable about it, where this pressure hits them externally. It hits them from their bosses often. And then they have team members who they're seeing accelerate, and maybe they don't feel ready for that, or they just don't feel comfortable with it yet. And I think that's less so the listeners to the show for the most part. But I do get folks occasionally in the comments who are are saying, ah, you know, like this is going too far or, you know, this idea is out of a field. As A.I. is changing what our jobs are and how they function. What does it mean to keep your professional identity for individuals who are facing this?

Jiaona Zhang 33:47 Yeah that's a great question. If you if you just look at history I would say that there have been many times where our jobs changed because of technology. Right. So back in the day where manufacturing was at a certain point supply chain was a certain point. Our jobs looked very different. People were in factories actually assembling things. And then when we had machinery that did it, people didn't have to do that job anymore. And so I really think of it as a shifting as opposed to a replacement or a loss. And I think that the best way to really embrace this moment in time, which to me, by the way, feels one of the most exciting because it feels like we can empower people in ways that we never could before. How do you get into a mindset of I want to go do more. And now I can. And my job is not to feel worried about the loss. My job is to go run towards an opportunity. Now I want to this is not on just the individual to figure out. It's on an organization to figure out. Right. Because if you are basically here's where I see things go wrong. You have executives at a company saying, everyone must use AI. And oh, by the way, AI is now in your performance review. And, and yes, of course, it creates fear. You're like, well, okay, I just, I have to just use AI. And by the way, it's really tied to the very first topic that we talked about, which is everyone's token maxing, because they're like, I've been told to use AI. And I'm like, no, you should use AI efficiently. If you use AI to go like redo the font on this deck when you could have just clicked a button, that is not an efficient use of your tokens. But if you used AI and you took something that took, you know, humans five hours per week, and you took that and agent did in five minutes, that is time back, right? And so I think a lot of it has to do with the way that the leader shows up to help people embrace change. And so instead, if the leader didn't, if this leader just said, use AI, you're gonna be judged for it, Good luck. And I expect that to happen tomorrow. That is where I think a lot of the fear is coming from. And by the way, that is truly how a lot of leaders have pushed AI onto their organizations. So I understand why there is this sentiment. But instead, if the leader said, hey, look, I have actually done some, you know, deep work here. I understand the time that is being spent in order to achieve, you know, deals getting closed, you know, support tickets being addressed, this, you know, this product line being produced, whatever it is, whatever outcome it is that you're trying to drive. And if the leader came in and said, look, I know how much time it is taking of each one of you to go do that thing. Now, I still need you because I need you to do this other thing that we haven't even had the time or the resources to go do. And, or I need you to go explore this new area that, you know, again, we haven't had staffing. It's just really ironic because a lot of the best companies, they keep hiring because they have so much ambition and things to do that they need more people. It's not like there is a lack of opportunity. And so to me, it's that reframe. The reframe is. Here's what I actually need you to do. This is a new opportunity I need you to go after. And this other thing that you were doing, I've looked at the data and I know that you spent a lot of time doing it. Here's a really easy way for you to go adopt a workflow to make that go away. And so a big proponent, a very easy thing that I think a company can do, and we do this at Laurel, is How do you have a repository of skills? And back in the day, you heard the word playbooks all the time, right? You'd be like, well, what's the playbook for sales? What's the playbook for design? What's the playbook for whatever, whatever function? There was like a playbook. And now the really beautiful thing is you can take these playbooks that used to live in these, you know, PDFs, whatever, documents, office documents that somebody wrote and spent many hours on and then nobody followed because did you remember to open the document? You did not. Instead, you can now get a skill delivered to you, right? And you can actually get that push to you at the moment where it makes sense for you to say, ah, this is what I should be doing. And I'm a new employee. I had no idea that this is what I was supposed to be doing. And instead of, again, me reading an onboarding doc that I never read, reading the playbook that I never read, it is actually in time and is something that the organization can contribute to collectively. So, you know, I would say that for the past couple years, it was all about the foundational models. It was all about the infrastructure. And now I feel like the race, like the thing that we're all trying to figure out is the business context, the context layer. And how is that shared as opposed to locked away in one individual's, you know, like whatever, co-work instance? [38:53] Conor Bronsdon: [OVERLAP] 100% agreed, and I'll say this has been a theme for us on the podcast the first half of this year in 2026. Anyone that's listening for the first time, I recommend if you wanna dive into the context layer, we did a great episode with Jerry Liu from Llama Index, where he talks about moving away from being an agent framework and the success they had there into just focusing on unlocking PDFs and being as good as they can via DocumentOCR and turning documents enterprise context. We also had episodes with Neo4j, Oracle and others, We talked about agent memory and unlocking enterprise data, error byte. So definitely recommend checking those out if you want to dive more into the context layer of a whole series. I should probably create a podcast playlist [39:34] Jiaona Zhang: [OVERLAP] Just, [39:35] Conor Bronsdon: [OVERLAP] when [39:35] Jiaona Zhang: [OVERLAP] just [39:35] Conor Bronsdon: [OVERLAP] I have a chance [39:35] Jiaona Zhang: [OVERLAP] on it. [39:35] Conor Bronsdon: [OVERLAP] about this. Yeah, I think you brought up a really good point, not just on the context layer piece, which I think is absolutely crucial. Like context and memory are where there's so much focus happening today because As a lot of value aggregates to the frontier models and source models, in some cases, for folks who are looking at efficiency, we're all realizing, oh, we need to nail context, we need to nail data, we need to nail memory to make the most of these. And I think we're seeing, you know, agent harnesses evolve very rapidly in a way where it's like, oh, you know, this agent framework is less valuable than it was six months ago. And this is to me, I agree, extremely exciting. There is so much change happening here. People are not that far ahead of you. I think this is the really cool thing for me is like even those of us who may feel a month or two behind at sometimes, that is because these concepts are so new that you can dive in and learn today and be good at them as good as many of the best people in the world at them very rapidly. There is a massive opportunity here. If you lean in and embrace it, which I would say, look, we've been told for decades that growth mindset is important, that lifetime learning is important, that you have to embrace change. I think it's just these concepts are more important than ever because there is more leverage being applied by agents and by AI. I look at this kind of across the board, which is AI is leverage. It doesn't always work for you. You have to enable it. It's not always going to be a magic bullet, but it can help you learn faster. It can help you do more things. It helps me ship more LinkedIn posts and more podcasts and all these different things [41:11] Jiaona Zhang: [OVERLAP] It [41:12] Conor Bronsdon: [OVERLAP] beyond [41:12] Jiaona Zhang: [OVERLAP] has never been [41:12] Conor Bronsdon: [OVERLAP] coding. [41:12] Jiaona Zhang: [OVERLAP] easier to learn, truly,

Conor Bronsdon 41:14 Yes,

Jiaona Zhang 41:14 which

Conor Bronsdon 41:14 yes.

Jiaona Zhang 41:15 is why I think it's so exciting. It'd be one thing if everything was changing and it was really hard for you to get up the curve, but it's actually never been easier to learn.

Conor Bronsdon 41:24 I huge shout out to notebook LM by Google. I have to say one of my favorite products of the last couple of years. And I

Jiaona Zhang 41:31 Yeah.

Conor Bronsdon 41:31 use it for every episode for prep calls. Actually, I'll have my, uh, research agent will build me notebook LM so I can go explore that. And then I'll talk to the guest as well, but it's a nice way to go about it.

Jiaona Zhang 41:42 Yes, I love the prior two. Yeah.

Conor Bronsdon 41:44 So I want to ask you about this learning piece because, uh, learning in this new era is it feels easier than ever for, for so many folks, but it also feels intimidating for others. How, how are you enabling, um, your students at Stanford and then also how are you enabling your team members to, to learn on these tools?

Jiaona Zhang 42:06 Yeah. It's. twofold again, because I think a lot of times problems are not as simple as just do this one thing. And usually actually the pairing of like two things that have tension with each other is where you get the best result. So I'm a big believer of the core fundamentals have not changed. And in fact, they are even more important. And I'll give you just one very specific example. One of the most important things I teach students And anyone who is hoping to do the product role is this idea of you have to start in the solution space before you move to, sorry, you have to start in the problem space before you move to the solution space. And what that basically means is there's so many people who get really excited about a thing, an idea. What if we just built X? Or I thought about this new flow, this new thing, and it's gonna be amazing. And if you just skip to that and build, you're going to be It's going to be very unlikely that you're going to be successful because in many ways when you're doing that, you're just throwing darts in a room and hoping it's going to land on a bullseye. But you don't even know where the dartboard is hung or, you know, the target is hung. You don't even know where it is, how big it is, what the bullseye looks like. But instead, if you start in the problem space and you say, well, this is what success looks like. This is what it means from a business perspective. This is what it means from a user perspective. then you actually have drawn your target. And then the chances that you will throw a dart, which is a very hard thing to do, throwing a dart and hitting a bullseye, hard thing to do, which is why, again, like first products, almost always never succeed. You gotta iterate until it's good, right? And so that is a very, very like one-on-one concept, maybe the most fundamental concept of product of all time. That has never been more important because it has been Today, you can build more things that you could have ever built before. And as a result, you're basically like, I have this machine gun that just sprays darts everywhere, right? And so you're like, okay, should I just spray them? And that is not gonna make your chance of hitting the target better. In fact, you're just gonna use a lot of darts. And again, this is my note at the beginning, Sure, token costs will go down, but if everyone's shooting darts everywhere, like that's just a lot of effort, a lot of energy spent, like literally physical energy spent when you think about like data centers and compute, right? And so that is so important. And so when I think about teaching students, teaching anyone, how do you make sure that the one-on-ones, the most important fundamentals are totally internalized? On the flip side, you know, what is really interesting is your ability to have superpowers, your ability, you know, we mentioned this earlier, to go from zero to one or go to take something from nothing to truly production ready. That has really accelerated. And so I think that the second thing I care a lot about is this idea of curiosity and tinkering and playing and just being out there doing and trying all the time. And when you couple those two things, which is you don't lose sight of the fundamentals, right, you don't get intoxicated by the shiny, and you're really like, this is what I need to do to break a problem down from a first principles perspective, while you are trying things all the time, experiencing, curious, learning. Again, it's never been easier to learn. When you put those two things together, that's where you get magic. Because you're grounded while you're moving at a new velocity. And so essentially, you're able to go much, much faster and your ability to be successful increases.

Conor Bronsdon 45:58 Yeah, I really like the idea and phrasing of outcome engineering, which has begun to make its way into the popular lexicon over the last few months. And it speaks to exactly what you're saying, I think, which is that, you know, I think particularly for software engineers, which is a lot of our audience, agents at coding can feel initially as kind of a threat where it's like, oh, wait, I'm really good at writing code. I

Conor Bronsdon 46:26 don't want this to go away. I enjoy the craft of this. But the job has always been solving problems and it's always been achieving an outcome. And now we're moving to a different abstraction layer of how that work gets done because we're less limited by human bandwidth and execution and we can apply agents to challenges. We're spending more time on architecture and these other tasks, less time hands on keyboard, maybe more time speaking into a mic, prompting an agent to give it enough context. But this is in fact what we've always been meant to be doing. And some of the tool obsession and the how the work gets done obsession that shows up is more about us maybe getting off track. There's been long conversations for years now. about don't be so obsessed with tools, like try to figure out how to solve the problem. It's not always like buying a tool or building a tool or using something. And I think we're just kind of seeing an extension of that now where We have this incredible new tool that can be applied and it's replacing in many ways, a lot of tools from previous eras, just like you would look at, you know, sewing by hand for the sewing machine. Like there's still space for sewing by hand and these handcrafts, but if you're mass producing, you, you probably want to use the mechanized version. Um, and that change can be, feel scary, but it is also very exciting for what it unlocks. Just like how, you know, SpaceX cheap space flights are unlocking all this space technology. AI is unlocking massive amounts of code to be produced, managed. explored for people around the world, whether it's across personal websites or more exciting apps. Just a couple days ago, we had, you know, 82 and 0 go viral for like a vibe coded game where you try to go 82 and 0 in the NBA. And like, there's just so many examples of these little things that are popping up. People are just creating that wouldn't have been probably created otherwise. But it does make technology moats feel less durable because people are now able to throw tokens at a problem and kind of spin up much faster. And so as we come to the last few minutes of our conversation, I'd love to get your take on as our teams are changing, as we're applying these new tools, what does it look like to build a durable moat today?

Jiaona Zhang 48:45 I am a big believer that the future is going to be one where you really have brand moats and data moats and the technology moats are falling away to the point that you made. And so what is a brand moat? What is a data moat? And by the way, I think brand and data moats actually go hand in hand. I'll start with the data moat that's most obvious, which is you have data that nobody else has. But I think that a lot of times having a brand moat coupled with a data moat is actually the holy grail, because to really get truly unique data, you need to have trust. You need to have trust from your users, right? Like, why do you get access to that data that nobody else has access to? Because you have their trust. Or you have a very specific, you're doing something for them from a pain perspective, right? You're solving a problem for them. and that creates a brand mode. And that brand mode is one where I think a lot about why would I use you versus someone else in a world where, to your point, there could be someone that just spins up something that has some resemblance to the thing that you do very, very quickly. And the brand mode to me is one where, you know, it's, why you? Why are you going to safeguard my data? Why are you going to be the ones that continue to innovate? Why are you going to be the ones that actually cares about our relationship, right, as a client to vendor? And those things are really, really important. So we, you know, at Laurel, we think about this a lot. We do not ever want to be just known as a vendor. We want the people that work with us to feel a real difference in everything that we do, including the way that, again, why do we even enable the customer success to ship? We want, you know, the customer interacting with anyone at Laurel to feel like, wow, like they will solve my problem. End-to-end no matter who I call right and so that that that concept of like what is it that you are trying to represent and How does that flow through your entire product and how does it actually result in you know why people would trust you? That is I think the enduring mode in the future. I

Conor Bronsdon 50:57 Is this the rationale to you behind OpenAI acquiring TBTN a couple months ago? Saying, oh, you know, we know it's going to be easy to copy or have equivalent models. But if we're the cultural force and we create a data mode, we can be the winners here.

Jiaona Zhang 51:17 Yeah, that's interesting.

Jiaona Zhang 51:20 I'm not going to necessarily speculate, but I will say this, which is, and this is really much from my time at Linktree, and also when I look at really successful companies, I think of them as having created a movement. And so we talk a lot about this idea of like their X and Y Z company is the category creator or the category owner. And, and yes, there's like some truth in that, but what does that even mean? And when I really think about, you know, the, the companies I've done a really good job and really gotten into the zeitgeist and, and like captures people's imaginations, they've done a good job of like creating a movement and that movement makes them a category creator. Right. So how do you make it such that, I don't know, Clay is a great example where it's like, hey, you know, I'm gonna really make a movement out of the fact that there is this role that is a go-to-market, you know, engineer. And by the way, to be a go-to-market engineer, you don't have to be an actual, you know, engineer by training in the sense that you were a CS and, you know, actually worked as an engineer. You can be like anyone. You can be the person on the sales team. You can be the operations person. You can be anyone. Like, and that creates a movement and a real emotional pull. And so that is what I think is very enduring and really kind of solidifies the brand and reinforces it. And so when I think about, again, like, why did OpenAI do this? I think there's probably two pieces. I think there's one just in general. I think the foundational labs are really thinking a lot about deployment, right? Which is, hey, I have this amazing technology. You know, what's actually hard? Human change management, right? And so [53:00] Conor Bronsdon: [OVERLAP] Yeah. [53:00] Jiaona Zhang: [OVERLAP] it's so interesting. And again, going back to like, what is so interesting about understanding how humans spend their time When you are able to have the best lens on how humans spend your time, that is one of the best ways to drive change management. You literally are able to say, you spend your time doing X right now, I think you should spend your time doing Y, and oh, by the way, let me make that really easy for you. Whoever nails that combination, That is where you, that is the winning formula for deployments, right? And like getting the change management to happen, to get adoption to happen. And now like that, that's just the functional piece. Now, if you can make that a movement where people are like, that is what I want to go do. I want to go make that happen at every company across the world. I wanna be, you know, the same way at Webflow, we had all these amazing freelancers and agencies, service providers, you know, they created a movement around web flow and web development and what does it look like to think in a low-code, no-code way. The future of how do I get a lot of people to say AI transformation is super exciting and I love the human aspect to it and the technology aspect to it together. And hey, let me be equipped with the best tools to go do that. That I think is really exciting. And so I know I talked about a lot of different points besides just why open AI bot. the news outlet, but I think that's the thinking that is probably going on behind the hood. [54:31] Conor Bronsdon: [OVERLAP] No, this is perfect because you teamed me up for my last question here, which is in many ways, Laurel is reinventing the category of work quantification for the AI era of work tracking, [54:43] Jiaona Zhang: [OVERLAP] Yes.

Conor Bronsdon 54:43 whatever you want to define it as. How is Laurel creating a movement?

Jiaona Zhang 54:47 Great question. I want to start here, which is so important to us, and I alluded to this before, we are human optimists, which is we don't think it is, it is not the future to just say, well, all lawyers are going to be gone and it's all going to be lawyers in a box or the AI lawyer and like all accountants, like, in fact, we think the opposite. We think that, you know, what is the one thing we could be doing that would be the greatest way to contribute to you know, humanity and how we all spend our time, which by the way, no matter how much money you have, it really comes down to like, everyone only has so much time, right? And so if we can make it so that everyone is able to do the piece that is most uniquely human.

Jiaona Zhang 55:32 lawyers get to actually think about the spirit of the law. And how does this one case different than this other case? And how are you setting precedent? As opposed to the templates, the redlining, the little tweaks, if those things can be taken off your plate, the, hey, did I actually get my thing out? Did I get my bill out, for goodness sakes? If we can just take all of that back office, all of that really kind of like, why are we doing this as a human, as humankind in today's day and age. That is where I think we fundamentally believe we can add value. And so the movement that Laurel cares so much about is we want to return time. We want to return time to everyone. We want to return time to the individual, the lawyer, to go back to their kids after a really long day at work. We want to return time to the organization at large. We want to find the pockets of inefficiencies, not because, you know, therefore agents to take over the workforce. No, it's because we want to know the right pairing, the right pairing of the human and AI such that we can actually return time to the human and get the efficiency globally.

Conor Bronsdon 56:42 Jay-Z, thank you for a fascinating conversation. I appreciate you melding all these wide-ranging threads together. What's the best place for our listeners to go to check out Laurel and to follow your work?

Jiaona Zhang 56:54 Uh, probably on LinkedIn. Um, and my link tree is still up. So those are the two things.

Conor Bronsdon 57:01 All right, we will have to find Jay-Z's link tree and get that in the show notes. Jay-Z, thank you so much for the conversation. It's been a ton of fun. I hope our listeners enjoyed it. And if they want more conversations like this, as always, they can subscribe to our newsletter at newsletter.chainofthought.show. Thank you to the more than 1,000 of you who have now signed up. So Jay-Z, thanks so much. It's been fantastic.

Jiaona Zhang 57:23 Thank you. It was really fun, Connor.