The Future of AI with Bret Taylor, Winston Weinberg, Garrett Lord and Aaron Holmes | Aug 20, 2025

0:13Welcome, everyone, to the Informations TI TV. My name is Akash Pasricha. It is Wednesday, August 20th, and we have got a powerhouse lineup of guests for you today. Brett Taylor from Sierra and Winston Weinberg from Harvey are coming on the stream in just a few minutes for a very special conversation about what AI agents can actually do. We've then got Garrett Lord, the CEO of Handshake, who is going to come on to talk to us about the future of work in the era of AI. And last but not least, we're going to get to a great story that we just published this morning about how customers of AI models are feeling the squeeze again while companies that sell those models are soaring.

0:52Before we get going, I want to highlight a big story that we published this morning in the information. The information is first to report that Character AI has talked in recent weeks to bankers about a potential outright sale of the company. You might recall Character's founders were hired by Google as part of a$2.7 billion deal about a year ago. There, of course, was a company full of employees that remained. That is what this sale would be about. I should say the company is also talking about possibly raising a few hundred million dollars at a valuation of about$1 billion. And so we will see how this story unfolds.

1:28Okay, on to our first guest. We've talked a lot on this show about AI agents and the hype cycle that space is currently in. Businesses are pouring money into these tools, hunting for ROI, and hoping not to be left on the sidelines. Today, we have a special interview with the founders of two of the most closely watched agent startups. Brett Taylor is the co-founder and CEO of Sierra. You may also know him from his role as chairman of the board of directors of OpenAI. And Winston Weinberg is the co-founder and CEO of Harvey. They're going to be sitting down with our editor-in-chief, Jessica Lesson, to discuss the state of their businesses and the state of the AI agent's market.

2:07Jessica, over to you. Thank you, Akash. Brent and Winston, it's great to be here with you. And I'm going to pose a challenge. I'm going to say, let's try to say agentic the least number of times. It's too early for a drinking game, but we could have a coffee game. But no, I'm just kidding. I'm so excited to be on with both of you because you are two of the leaders of the most closely watched startups for actually having products with traction, a clear focus in the AI space. At the same time, there are so many big questions swirling around about how businesses seize this moment and what's really sustainable.

2:50So, Brett, maybe starting with you, you more than probably anyone I know in the Valley have really straddled the big company and small company world. You know both sides well. In your new perch at Sierra, as you're going out to Fortune 500 companies and convincing them to use your customer service platform to build their own agents, what's working? What are you hearing in the marketplace and how are you getting traction over the last even couple weeks? Yeah, it's a really fun time in the market. I've never seen anything quite like it. I was in college during the dot-com bubble, but I think it sort of has echoes of that to me where everyone knows how impactful this technology can be.

3:34I think where Winston and I are working in customer service and the legal profession, if you talk to an economist and you ask them, where will AI have an impact, that would be at least in the top two, three, along software engineering, probably. And so we're in this area where I would say demand is not an issue. I think everyone knows that this will have an impact on their business. And the question is, why your product? Why your platform? And you talked about big company and small. There's this kind of adage in enterprise software, best of breed or best of platform. You know, best of platform means, hey, I've already got an enterprise license agreement with one of the big enterprise software players.

4:11I'll just buy it from them. You know, it's easy procurement wise. You probably get a discount because you're buying a bunch of other stuff from them. And then there's best of breed and it's saying, hey, I want the best solution in this space. We are definitely in a world where best of breed is winning. And the reason for that is the cost of the software is actually relatively modest relative to the business impact, both cost savings and revenue impact that these agents will drive. If you think about you run a really large call center, if you can remove 75 % of those calls versus 50%, the business impact there is so much greater than the licensing costs.

4:48You really want the product to work. If you're thinking about an agent for sales, which is one of the things we also work on, if you can make more sales that's so valuable, the marginal cost of that software is actually not something you fixate on. And as a consequence, I think you're seeing the surge of startups right now just because our products are better, you know, candidly and they work better. And that's what matters right now. But if you fast forward, you know, the way this works, it's like a pendulum, right? And so you'll start to see market leaders come out. I believe Harvey and Cyr will be two of those.

5:19And then you'll start to see some consolidation. So I think we're sort of in the pendulum swinging towards best of breed because our products work. and right now impact is what matters as the market really matures and products working is no longer differentiated as it is really now i mean there's so much snake oil out there right now to be honest with you then you start to swing back towards platforms and the race for uh winston and uh me is we need to become those standard platforms as the pendulum starts swinging back you know and i think that's just the the evolution of the software industry and so i want to sort of same question coming at you, Winston.

5:54But Brett, do you mean when the platform swings, you want to be the consolidator? Or are you saying you envision that Sierra will be consolidated? Yeah, absolutely. We want to be consolidating. You know, that's the thing is, I think you look back at the birth of the web browser when I mentioned being in college in the dot-com bubble, the new companies then, Amazon, Google, PayPal, are now the incumbents today. And I think that when you saw the mobile phone come out, you know, you had Uber and WhatsApp and DoorDash, all starting around that era. And so I think the opportunity right now for entrepreneurs is this market shift driven by technology and driven by large language models changes the power dynamic between incumbents and insurgents.

6:36And that's really exciting for us. And the question is, can you become the next incumbent, which means driving more customer success? And we're powering hundreds of millions of phone calls and chats this year for everyone from ADT, home security to ramp. and it's very exciting, but we need to earn that, right? And I guess let's put this way, AltaVista didn't get to write the history books, Google did. And so, you know, it's, and being a first mover matters, being great for a sustained period of time is actually what matters here, and it has to start with the quality of the product. So, well, I think both Harvey and Sarah are leaders in their respective markets, where really matters are, you know, pace of innovation over the next five years and 10 years.

7:16And I think that's really what's going to determine who has the privilege of becoming incumbents a decade from now. Yeah. Well, to you, Winston, I mean, I don't think of the legal profession as the most tech-forward industry out there. Come on, Jessica. You know, very risk-averse, very, you know, all of those things. I mean, obviously, Harvey has made a splash. How have you done it? And how are you making, I mean, to Brett's point about the ROI case, which I think may be a little fuzzier still than he lets on for businesses, But how are you making that to the biggest law firms in the country? Yeah, so I would say that the ROI story is definitely different for law firms versus in-house.

7:55So for in-house or kind of like large corporates, it's a little bit more obvious, right? Because a lot of the efficiency gains they can make, you can compare that against things that they might be paying alternative legal service providers to do or kind of like smaller law firms in different countries. So I think the ROI makes a lot more sense there, especially if you're building a workflow that does something specific from start to finish. So like my best example here is it's a little bit harder to do an ROI for like a general purpose or even a vertical purpose assistant in legal. It's a lot easier if you're building a system that does like merger control guidance, right?

8:29Like something that is very specific. You can just see how much this costs versus human labor or versus, you know, a lawyer doing this in the loop. For law firms, I think they're also starting to think about this a lot. Like because of the billable hour, right, like efficiency gains don't make as much sense. But if you think about what kind of a problem, yeah, it's a huge problem. But if you actually think about what a lot of law firms want, they actually want like a better leverage model. Right. And so like a lot of it is like profits per equity partner. Right. And so what they're if you can do more work with actually like less on the bottom of the pyramid, that's good.

9:05And I would argue that that's actually good for the junior associates, too, because the reality is like most tasks in legal are not fun to do. Right. And if you ask junior associates, most of them are in the profession because they want to basically go through 10 years of doing these rote and mundane tasks because they really like what the partners do. Or they really like what the – they want to go to trial. They want to advise clients on deals. They want to be in the boardroom. And so it's actually an easier sell than you think because from the top, it's, okay, we have to do this, and maybe we're going to redesign how we basically manage our services, how we price things from the partners.

9:43And from the bottom, it's, oh, wait, you're telling me that I only have to do the boring part of this job in five years for only one year when it used to be that I had to do that for 10 years and slog it out until I get to do what I want. Yeah, it's a good pitch. Where would you say like the industry is, I mean, in this transition or like what what things are starting to tip now that maybe you weren't seeing six months ago in terms of adoption? Yeah, I would say it changes every three months almost. And I think like in early 2023, it was, should we adopt anything at all? Like, what should we do?

10:18There was a lot of self build as well for law firms. And then in 2024, and again, it's like shifted every like three months, every quarter, basically. In 2024, it was a lot more, okay, we have to do something, but should we self build? Should we buy a, you know, buy a platform, a vendor, and then do a combination? And now it's becoming, okay, wait, I think we have to change our entire business model, right? A lot of the calls that I get on are a combination of products plus change management. In other words, these are the things, like we have tons of market data. We work with a bunch of private equity firms, and we have all of this market data that is incredibly valuable.

10:54Is there a better way for us to actually productionize that? And so I think law firms more and more are starting to think about this as, how do we become tech-enabled services? But that can be all over the place. It can be for the lower end work that they do at a loss to get like the big LBO or the big merger. Or it could be, yeah, we're, you know, a firm in a different country and we're competing against the giant, you know, international firms. Is there a way that we could actually design how we do M &A differently with technology to compete with them at scale? Yeah, that's fascinating. So much.

11:30Really a rethink from the bottom up. Yeah. Yeah. I'm curious for both of you, sort of what's happening to your own costs. I mean, I think this is why, I mean, AI is such a tremendous technology to cover because we're seeing the demand. We're seeing the revenue come far faster than certainly other cycles and tests. But so are the costs. Right now, you guys do not develop your own models, I believe, but maybe to you, Brett, first, and then Winston, how are you thinking about, from a product perspective, how to make sure that costs can kind of, you know, maybe over time stay in line with the revenue opportunity and how do models fit into that for you?

12:09Yeah, I think most applied AI companies, and Winston and I have talked a little bit about this, probably shouldn't be building their own models. You know, the pre-training costs of a model, you know, to actually recoup that, given how quickly they depreciate in value, just really requires just huge scale. And as a consequence, I think we'll have a modest number of very, very high-scale players with very big distribution channels like ChatGPT, where you can sort of essentially amortize the cost of that training. And it's why you've seen, I think, in the intro companies like Character and others, like it's hard to sustain a business if you have such high initial costs.

12:49Within the applied AI market, it's actually better and easier. And I, well, I won't tell you our gross margins. They're quite good. You can. They're the interest. And it's because you're not selling AI. You're selling a mergers and acquisitions product in Winston's case, or for ours, you're saying, hey, we can answer your 1-800 number for 1-100th of the cost with twice the CSAT. And you're providing business value, whether that's originating a mortgage or explaining the benefits for a health insurance company or simple stuff like providing technical support for a Sonos speaker or Abyssal vacuum cleaner.

13:32and that is something that you value independent of token costs and i think this is a little bit the way i think about the market playing out as you have the frontier and foundation model companies and they're going to look like i think the infrastructure as a service market where it's really high scale revenue margins will always be a question differentiation will always be a question and just like in infrastructure service there is differentiation there is in the models as well it's just subtle and constantly changing like oh this model is better for coding and this model's better for that. And, oh, these are really great voice models.

14:03It's a competitive market. I think the agent companies will end up like the software as a service market, where it's primarily bought by business teams, not just technology teams. It's valued according to a business outcome, not according to utilization. You don't buy SaaS based on how many CPU cycles it uses in the database, right? It's just not the way you think about it. And I think that Harvey and Sear are selling a solution to a problem, not selling AI. And it turns out that these problems were unsolvable with digital technology before AI. So that's the reason we can exist. But how we talk about our value problems, you talked about Winston transforming the shape of a law firm.

14:45You know, we're transforming the shape of customer engagement. And just to make it like numerical, typically, if you call like a cable company on the phone, it probably will cost them like$20 to$30, depending on how long you end up on the phone. Most customers aren't worth$20 to$30, and most interactions certainly aren't worth$20 to$30, which is why it's impossible to talk to companies. There's entire websites devoted to finding people's phone numbers. You bring that to$0.20 or$0.30 over time,$0.20 to$0.03. All of a sudden, it changes the game. You can just have more conversations with your customers, and it's going to go from treating these things as a cost center to actually, hey, I can have personalized, multilingual, one-to-one conversations with every single one of my customers.

15:28And you look at the market, like say the mobile market where everyone's fighting for a fixed number of customers, mobile phone plans, the smart companies are going to take advantage of this like step change and say, how can I improve retention, lifetime value, grow subscribers, all these things. And so what we're actually selling is just much deeper than AI. And I like that abstraction layer. And I think it's why I'm so bullish on the applied AI market and the agent market broadly is because I think this is how most companies want to buy technology. Like tokens have no bearing on your business, but your cost of your customer experience, the cost of your legal department, the success of that merger certainly do.

16:08And that's fundamentally what we're selling. I would also even say that, oh God. No, go ahead, Winston. I was going to say, other than efficiency too, if you think about a lot of these tasks, and I think I've seen companies like Brett and us and other companies too, a lot of this pitch is beyond efficiency. So if you think about like a merger, like most reasons deals die is because of speed problems, right? They get like massively delayed. And so my point is that a lot of the value that you're going to start providing is actually not just how much did this cost before and versus, you know, and now it's X divided by three or whatever, but it's actually, oh, there's a higher chance that this actually goes through, right?

16:45Or if you're pre-processing a bunch of documents for like an internal investigation, that makes it so that the outcome of that internal investigation, you probably have a better sense of like what happened at the company. And so my point is, I think that in the next couple of years, we'll see a lot more than just efficiency arguments. We'll say, oh, wait, this actually had like a lot of business impact that had nothing to do with cost savings. You also, though, need, it sounds like you're both banking on the fact there'll be enough competition on the foundation model layer that those prices aren't going to go through the roof?

17:14I mean, is that because all of those companies, as you know, Brett, face incredible pressure to make their lines cross as well. But is it just you think that there'll be enough options out there that that's not something that will become a major challenge for this agent software layer? I do. I also think there's not one foundation model that people should use. I think they're becoming more purpose-built. It's a little bit like almost like the data storage market. You have transactional databases, you have columnar data stores for analytics, you have things like S3 and open source. And almost every one of those has an open source competitor and a proprietary one.

17:52We're seeing very similar things. We probably do on the order of 20 inference calls for every message or sentence you say to one of our agents, which surprises most people. Not all of it is rocket science. It's not all GPT-5 high reasoning mode. It's really simple things like, oh, is that a television in the background or is that you speaking? Or you're on the side of the road for a roadside assistance case and there's a car horn. Was that someone interrupting the agent or should I keep talking? And you don't need AGI to determine that it's a car horn. And I think that's how the model matures.

18:34I really think for an applied AI company, not open AI, for an applied AI company, these are reusable pieces of infrastructure and you'll do a price performance quality analysis on what's right for the task. And I think a great sort of full stack applied AI company will be really good at absorbing these new models as they come out, picking the right purpose built model for the task. And to some degree, part of the reason I don't think training or certainly pre-training models, fine tuning, I think may also weigh in an importance over time. that's a whole different conversation but you know i think that you know you really you should like take advantage of the wave of the market of this investment these foundation models and it won't always just be the frontier it might be sort of this boring important task where you can use a really low parameter account model that's fast and cheap and that's great and that's how you're going to sort of build the supply chain of technology to make these agents shine okay i want to zoom out a bit, but first on pricing, Brett, I think you've talked a little bit about maybe pay per outcome or usage pricing.

19:36And it actually reminded me of Quip if we go way back when. But I feel like before you were at Salesforce in the Quip software land, this was a model you were also thinking about. But talk a bit about how pricing might evolve in the space. So at Cira, we charge our customers when the agent autonomously completes the task, whether that's making the sale, solving the customer's customer service problem, of we have to transfer to a person that's free. And we do that because it really aligns our interests with our customers. It's a great way to consume the software. I'm excited about it broadly. I think this is where the industry will go.

20:08If you look at, you know, 1998 to 02 when the sort of cloud market emerged, you had the emergence of, hey, you can run this software in the browser and you had software as a service where, you know, you subscribed to it rather than paying it for a perpetual license. Those two went hand in hand because they sort of had to, you know, So if you're running, everyone has the same version of the software in the browser, how do you pay for innovation? So you need a different model than buying Windows 98 and then Windows 2000. You had to say, okay, let's just change the model. But it also really aligned software vendors with customers because you could stop subscribing to it.

20:45So it created this sense. That's where customer success, the profession was created by, I think, Salesforce. But, you know, you ended up sort of changing in a really structural way the relationship between vendors and companies. I think with agents, because software isn't just a productivity boost anymore, it actually can complete a task. You know, for tasks like sales, you pay people on commission because when you can actually measure the impact of a task, the best way to align incentives is to pay for that job well done. And I think we're in a position with customer experience where you kind of know that it was a job well done.

21:20Did you solve the problem? Did you make the sale? And so we just love charging for that. And I think it will disrupt the industry, but really, really good for companies. Because just imagine the return. You talked about ROI. It's so easy with Sierra to know you're going to do an ROI because you only pay us when we give you a return on your investment. And I think that is the right way to charge for agents if you can. You know, not all agents have quite so measurable of outcomes as we have, and we're just in an industry where it works really well. So I want to ask you both. I think there's been a lot of talk now, okay, we're going to have all these agents.

21:54They're going to need to interoperate somehow. They might need access to things stored in other places to do that. We've seen companies like Salesforce and others try and, you know, just kind of also for platforms, you can't have agents pinging you all day long either. So there's a real tension, I think, in how, if we fast forward, what this software layer will look like. I'm curious for either of you how you sort of think about that. And are there analogies like the app economy, like something else that can help people kind of grok what a very agent-heavy future looks like? Yeah, I think like you can think about a lot of these agentic systems as what is the type of tasks that they're trying to solve.

22:42And then if you have like a layered task, you're just combining tons of them together. Right. And so my point here is, if you have a task like a merger and acquisition, right, that is going to have all of these different agents that are doing different types of research, different types of diligence. They're looking at, you know, some web browsing, different types of external data, internal data, et cetera. But then you're also going to have probably a bunch of agents that are coming in there that are investment banking related, right? Like if you think about some of these very large tasks. PR agents, imagine that.

23:15Yeah, it's across all these domains, right? Yeah, you're going to be inundated. But my point is, does that end up being a bunch of different companies or do the companies that solve one portion of this or like one vertical, or is it easier for them to tap on agents that solve the other parts of the vertical? And I think that's the question. And that's, to me, like where this ends up is that for the very simple tasks, maybe you'll just have one, but for the really complex things that are more like project-based or like if you take a group of humans to do the task today, right? And they're from different verticals, who owns that?

23:51Is it a bunch of different companies that own each vertical or is it actually one company that somehow figures out how to get into multiple? And doesn't that advantage the big guys? And maybe Brett, back to you. I mean, you've worked at Google. You've worked at Salesforce, right? To be the Google, not the AltaVista, right? To be the platform, not the, I forgot what your other analogy was. But like, how do you, and yes, it's having a great product, but can you unpack that a little bit of like, how are you going up against the Googles and the Salesforce's today? And in a way that, you know, you think the gap will widen five, 10 years from now.

24:28Yeah, it's hard to predict, to be honest with you, even as someone, as you joked, I'm probably as much in the middle of this as anyone. It's really hard to predict the future of the software industry. I mean, broadly, thanks to coding agents, just the cost of making software is going down. And I think if you just project forward, the marginal cost of new software will go down by a lot. we haven't really experienced a world where there hasn't been a scarcity of skilled software developers and we will. And I just don't know what that means to our industry as a whole. Similarly, you know, you've always had these sort of core systems of record like ERP systems at companies and they are both databases, but also a bunch of web-based software that are like workflows on top.

25:12And a lot of people think, okay, agents will be doing a lot of those workflows. So what does that mean for the company? How much of their value is in the database? How much was in the workflow? And you'll end up with a sort of natural tension of people trying to, you know, hold on to the value that they had. Plus, you know, companies, you know, fundamentally, I think it's not any company's data. It's like not any software company's data. It's your client's data, right? So I think that's a non-issue. I don't think we as companies have any rights to this data. It's our customers. But I think how it plays out, like what's more important, sort of the agents running on top or the systems that are interacting with it.