Latest / Cheeky Pint / Bret Taylor of Sierra on AI agents, outcome-based pricing, and the OpenAI board

Transcript

Bret Taylor of Sierra on AI agents, outcome-based pricing, and the OpenAI board · Cheeky Pint

• [00:00:01.01] John Collison Bret Taylor is the ultimate Silicon Valley veteran. He was one of the creators of Google Maps, invented the Like button, was co-CEO of Salesforce. He pushed through Elon's acquisition of Twitter when he was on the Twitter board. He's now the chairman of the OpenAI board. And his day job is founder and CEO of Sierra, which is bringing AI to customer service. He's one of the smartest people I know on the topic of how AI is changing established companies. Cheers. [00:00:20.11] Bret Taylor Cheers. [00:00:26.08] John Collison Most important question, have you installed OpenClaw on your work laptop? [00:00:30.11] Bret Taylor I have not. [00:00:31.05] John Collison Or have you played with OpenClaw? [00:00:33.12] Bret Taylor I have played with OpenClaw. I haven't bought a Mac mini. You can put these things in virtual sandboxes for less money. It's really interesting. I mean, it's very compelling. It's probably the first... I wouldn't have predicted the first broad... I don't know if consumer is exactly accurate, but maybe hobbyist use of AI would have been this semi-rogue open-source project that goes through three name changes in three days. And I love it. I love everything about the chaos of it just because people in our circles have been talking about AI agents for consumer use and all these fancy computer-using agents. Instead, you're chatting over WhatsApp with a thing on a Mac mini that is mildly unhinged and insecure. It's just fascinating. The whole thing is fascinating. [00:01:24.12] John Collison The thing that seems to me is funny is if you look at the landscape, still in 2026, if you open a new Gemini chat or if you open a new ChatGPT chat, it's basically a blank slate. There's no memory. Then, I mean, Claw, people talk about the WhatsApp and Telegram integrations and things like that. But it feels to me a big part of the value is not only can it do stuff proactively, but it has memory. But the way it has memory is this super janky... It's like the movie Memento. It writes things to a markdown file, and it's just writing the things to remember, and the compaction is kind of buggy. It didn't always write down the exact right things to remember and stuff like that. But isn't it funny that you can get super polished mainstream consumer apps that have no memory at all, or this wildly insecure three-name-changes project that almost remembers things by scribbling notes in the margin, and that is the state of consumer AI. [00:02:21.22] Bret Taylor I have a, probably not very thoughtful, but technical theory on this. Coding agents to have gone through transformation over the past four months. The difference between if we were here in October versus now, our conversation about the future of software engineering would be materially different. How often can you say that about a technology? People always, in my circles anyway, you look at a coding agent and you extrapolate to other domains. You're like, "Could all digital tasks be like this?" The answer is obviously yes, over some period of time. But it's really interesting because sometimes, I think the hard part of engineering is in the details, and code repos have very specific qualities. One is all the context is in one place, in files that are largely textual, not binary. For most broad information tasks, that's not true. When you're writing your annual letter, my guess is the sources of information were in so many different systems, data warehouses. It's not impossible for an agent to use those things, but the idea that you can straight-line from coding agents to writing the Stripe annual letter, I don't totally buy. Then similarly, when the agent's actually performing work on a code base, there's feedback, there's compile errors, there's often unit tests, there's integration tests. [00:03:46.05] Bret Taylor There's the history of every change that we made in a really formal format, along with code reviews. It's almost designed for a robot, and you can self-reflect. Maybe we as engineers have always modeled ourselves after robots, and now we can actually fully realize that vision. What's interesting about it is the idea that it wrote a markdown file for memory, I think is maybe more significant than a hack. I actually think to some degree— [00:04:15.00] John Collison Turning your life into code? [00:04:16.05] Bret Taylor Yeah, it's like you almost want to put everything in a file system that looks like source code, not because that's the only way these agents can work, but actually it's quite an efficient way to get a mix of context and random access memory. If you think about a vector database, it's more random access. You have to know what to look for. But actually, that's not how real memory works. There's a mix of it. You're loading a markdown file, and as you said, compaction, all these things matter. But the messiness of it actually probably produces a more useful agent than a lot of the fancier things. I use memory in ChatGPT, and I love it. But I actually think this idea that there's a directory of just everything you've ever done is actually maybe more useful to an AI than people think. Actually, if you follow over the past couple of months, just this emergence of harness engineering, where you're building the harness around an agent to do work, I wonder if in the short term, it might just be one of those idiosyncrasies of history, like mimicking a code base is actually the best way to make a general-purpose agent work. [00:05:23.16] Bret Taylor Maybe over time, we'll get fancier than that, but it's actually a relatively efficient harness for an agent. Anyway, maybe that's why it's— [00:05:31.06] John Collison It's very terminal-centric. It's backwards compatible with— [00:05:36.01] Bret Taylor You can use grep. You don't need to make some vector database. [00:05:39.10] John Collison The AIs really know how to use all the Unix tools. You got a lot of lift from that. [00:05:45.11] Bret Taylor That's exactly right. Software engineers were notorious for making tools for ourselves first, so then we just almost bend every other domain in digital towards that domain. But the reason I brought up things like unit tests, integration tests. OpenAI did this blog post—I can't remember the engineer who did the post—but on harness engineering. One of the more interesting parts of it was documentation. Rather than just having a single agent's markdown file, they had a directory of essentially the entire product, the architecture, and they're filling this out over time, and agent's markdown became pointers to it. My hypothesis, having used Codex a lot, I wonder if the output of a session where you make a change to Stripe's products should be a documentation artifact in addition to code, where the documentation artifact is actually what the product manager version of John, and the code was the engineer version of John, where there's a lot in the code that is more transient. You might be fine to delete that. What was the intention? What was the PRD? What was the customer problem? Is actually the more durable asset. I wrote this on X, and one of the funniest comments, it would be like, "It would be the greatest irony if software engineering agents made all of us just write documentation the whole time just because notoriously every good engineer hates writing documentation, and now that's our job." [00:07:09.04] Bret Taylor But I don't know, it resonated with me. [00:07:11.15] John Collison How much are you AI code... You're a very prolific engineer in the old-fashioned, hand-spun way of writing code. How has that changed? [00:07:21.10] Bret Taylor Bespoke artisanal code. [00:07:21.19] John Collison Exactly, yeah. Pour-over. [00:07:24.00] Bret Taylor That's funny, I would use that. I am trying to get to a world where I'm not writing code. It's hard, emotionally, if that makes any sense. I have a hard time not caring. I don't care about the assembly language produced by the compiler. [00:07:45.06] John Collison Why should you care about the code? [00:07:46.12] Bret Taylor Why should I care about the code? I care about correctness, I care about robustness, and I think I know intellectually, I don't need to look at how the compiler unrolled this loop to verify its elegance and correctness. Yet somehow I feel that way about code. I'm not saying the code doesn't matter, but I've been trying to force myself to not care because I feel like I won't be a self-actualized software engineer in the future if I'm too precious about that artifact which used to be so central to me. Right now, writing markdown files, maybe that's fine. It feels somewhat like a local maximum, and maybe we'll just be like, "Oh, of course, markdown is how we work with machines." If you think about what a compiler does, there's this interesting mix of formality and informality, and if you've used Python versus Rust, different ends of the spectrum. [00:08:40.16] Bret Taylor Now that you're not writing the code, I really wonder what that programming system should feel like and look like. I don't mind chatting with Codex, that's fine. But I also think, as you imagine, all the tests that you care about, it showing you demos and mockups. I wonder what the future integrated development environment, for lack of a better term, will be in that world. What I'm trying to do is force myself to not be emotionally attached to the code, which is very hard for me because that was my entire life before. I was proud of the elegance of the code that I wrote. But if I still care about the craftsmanship, what do I want? I haven't quite visualized it yet. [00:09:22.10] John Collison It feels to me like a very interesting time in agentic engineering because you were talking about this domain of harness engineering and people having skills and MCP and everything like that. It's always interesting when not only is the leading product in a category changing, we're actually just figuring out what the categories are that we need things for MCP or skills or stuff like that, and it's all very fast-moving. That just feels to me like a very interesting time where clearly a new way of engineering is shaking out. 2026 is clearly not the final word. [00:09:57.14] Bret Taylor Absolutely. In fact, I'm growing more skeptical of MCP as a meaningful part of the future. It's fine as a protocol, but it's interesting. Going back to your joke around OpenClaw just writing a big markdown file. I think it works better than a bunch of MCP servers. But going back to the point of every AI agent knows how to use grep and knows how to use all these things, I feel like this view of a multi-agent world was you have all these agents that do tasks that are fraud detection, another one over here for personalization. Then you make a super agent, it does all these things, and it looks really good on a whiteboard like most elegant-looking but completely nonsensical architectures. Then you realize if you just imagine you anthropomorphize the Stripe experience and you're the checkout concierge. What information do you need to have no a priori to actually make that a humane experience? What ends up happening in these multi-agent systems is you stuff all the context in the subagents, and the one on top has no ability to actually not sound robotic. Then in contrast, you look at something like OpenClaw, it's just a bunch of markdown files, and the memory feels right, even though it's a little bit kludgey. [00:11:17.16] Bret Taylor Similarly, if you go back to my arguments about a source control, like a repo, it has so much context. It's not like you just have the myopic view of the file you're editing. It really has some expansiveness. My sense is we're making true agents over time. The way we think about context and how that context is shared so that the agent that's orchestrating it actually understands what's behind all these APIs and why and the history will maybe look a little bit more like OpenClaw and less like MCP over time. I think these agents need a lot more context than what MCP affords. [00:11:55.11] John Collison One thing we've noticed... There's a bit of a what's old is new again phenomenon where with this agentic commerce stuff that's happening. We actually built the APIs for this 10 years ago as part of... Do you remember that move of social shopping that was in for a while? Buying on Instagram, buying on Twitter. It didn't quite happen for a few different reasons at the time. But the concepts are very similar that you want some action at a distance, you want to be able to go manipulate stuff off-site. Similarly, I think Patrick has wanted for the longest time in Stripe is the ability to just SSH into your Stripe account. What do you mean? It's a very ergonomic way for developers to work is you should just be able to log into your Stripe account and you have a command line there, and then you can list out all your... Or you can tail the payments log. [00:12:45.23] Bret Taylor He wants tail and pipe— [00:12:47.13] John Collison Exactly, all these things. [00:12:49.06] John Collison Of course, now we're building that because it's much more relevant in the agentic world. But I find somehow, all the agentic stuff is also bringing back... I don't know if you have this experience as well. It's bringing back a lot of ideas that you might have had before. [00:13:02.00] Bret Taylor Well, it is because to some degree, the elegance of Unix, which has been the basis of why everyone wants SSH and the curl command that was famously on the Stripe homepage. For people who got it, it was remarkable because you could have all these tools that did something well that was small and useful, and you could chain them together to make something great. I wonder in the future, we've talked a lot about this, if you look at the canonical software-as-a-service application like Stripe's console, and obviously you have what consumers see, but the configuration that a Stripe customer will log into, you would have a web app, and that's the forms and fields and buttons and graphs, and then you have the API. It was typically like a REST API or GraphQL API, and you could do stuff with it. This is how computers talk and this is how humans used it. I wonder if the web application of the future will actually be... Certainly, you want a web app for the rare human who wants to sign in, but will you have an agent harness? What I mean by that is something more than the APIs, but just like if you think about the harness that you provide in a code base, the skills, the documentation, the rules. Imagine the person who's the greatest Stripe expert who knows how to extract the most value from their Stripe account. [00:14:22.12] Bret Taylor That's the harness. Not the API. That's the button you click. Will that be an endpoint on Stripe.com, and so that your agent knows how to just get the most value from Stripe? I imagine you don't care if your merchants log in. What you want them to do is drive value for themselves, drive GMV, drive payments. I'm really excited about that because an API is great. APIs are awesome. But a harness is basically like, here's the instruction manual for all the Unix commands that power Stripe. That's very interesting. [00:14:58.01] John Collison Yes, and I think if you look at the shape of a lot of APIs that services have, and I think Stripe's API coverage is probably more complete than most, but ultimately, it is a way to manipulate some of the highest-value business objects in the thing. Whereas actually what you want is, one, all of the data to be browsable in some agentically accessible or textual way, and then all of the actions to be able to be taken by agents. It turns out there's a lot of switches in the dashboard that there's no API for, and we are all as an industry collectively discovering it. [00:15:35.01] Bret Taylor It might be easier. Imagine being a product manager in the future. You just need to add the switch to the dashboard. You're like, "Yeah, it looks like a Russian submarine to switch this, but who cares?" Agents can handle it. As long as the harness describes when to use that switch, it's easier than UI design in some ways. That's fascinating to me. [00:15:55.04] John Collison One funny point Dario made is it's not clear... There's a race between people getting their stuff accessible via agents and just desktop computer use getting better. It's actually not clear, will the approach be, Stripe builds way more APIs, and that's how your agents manipulate the Stripe account, or you just give your agent access to Chrome and a login. [00:16:23.07] Bret Taylor Well, so actually, I'll give a funny story here. Sierra, my company— [00:16:27.01] John Collison Sorry, we'll get to Sierra. [00:16:28.08] Bret Taylor No, it's fine. But it's a real funny story here. Sierra powers a lot of healthcare companies. On the healthcare payer side, health insurance. [00:16:36.05] John Collison Well known for their API quality. [00:16:38.11] Bret Taylor First of all, they're actually pretty sophisticated engineers at these companies. I really enjoy working with them. You end up with Cigna, Blue Cross Blue Shield on the healthcare payer side of insurance. Then you have healthcare providers like Sutter Health who we work with. Then you have revenue cycle management. R1 and revenue cycle management basically help providers get paid by the insurance companies. Then you have a lot of other people in the middle. Pharmacies, PBMs, and they all call each other. So a healthcare provider has to call a payer because a procedure happened and they have to get paid. We have payers with AI agents that pick up the phone. And we have providers that have AI agents that pick up the phone and make phone calls. We have revenue cycle management companies that work to make outbound calls to do it. We've already had— [00:17:24.22] John Collison Do they switch to the agent language? [00:17:26.07] Bret Taylor They don't. [00:17:27.04] Bret Taylor We've done English over the publicly-switched telephone network. You have TCP/IP and English over PSTN. It reinforces, I guess, Dario's point, which is you can engineer all these fancy protocols, but the rails that are already there already exist. English is spoken by all AI agents, and the public switch telephone network has been around for 100 years, and it all works, which is fascinating. You have all these fancy MCP things, and we're doing English over PSTN. On one hand, I think I actually agree with the principle that one of the powerful parts about AI, with its ability to do text, do audio, and I'll say, I'm not sure how you qualify computer use, but you can call it a form of image recognition and manipulation. Certainly that's useful because you get to the point where you don't need to fully finish the last mile to get value. The thing I'd say, though, going back to talk about all the actions, and they're not all in the product, all the APIs don't exist. These visual interfaces were designed for us. Think of... Stripe is, I think, famously, it was one of the few enterprise software companies with good design for a long time. [00:18:37.00] Bret Taylor The Stripe dashboard is really elegant, right? And most enterprise software, you can't say that about their dashboard. I don't think the ideal agent harness will be that elegant because it's optimized for something else. It's optimized for the context that you need to perform complex multi-step procedures on behalf of a person. And my guess is it's just very different. I think seeing the way you write a harness for a software or a coding agent is just so different than the way you do UI design. I'm certain that it's great that you can click around a green screen or whatever—click around a green screen because it's an oxymoron—but type around a green screen or click around a legacy on-premises enterprise software system. I think these harnesses will be really good. I wonder if, is there a world two years from now where Stripe's ability to work with the agent that manages commerce for a direct consumer e-commerce company, that will be one of the ways you're evaluated. In fact, if your... For lack of a better word, harness is not compatible with the way their agents work, that's actually like you're not compatible with them. I'm not sure that's right. [00:19:55.02] Bret Taylor I think it's great that these things are backwards compatible. It's great that our agents have spoken over the telephone already in English. But I don't think it's the long-term future because there's so much value that you can provide. Put another way, the agents using a sophisticated application harness can just do a lot more and do a lot more with higher fidelity. [00:20:15.00] John Collison Yes. Well, we should get to Sierra because you see a lot of real-world AI adoption. Maybe start by grounding us. What is Sierra? The business has scaled very quickly. One of the latest metrics that you can share because they keep changing from month to month as you guys grow. [00:20:30.01] Bret Taylor So at Sierra, we help companies make AI agents for their customer experience. If you have a big phone line, these AI agents can replace your IVR system and just pick up the phone. If you have a digital chat system, an AI agent can pick it up. You don't need to wait on hold. These agents can not only answer questions, but take action on your behalf. We work with healthcare companies like Cigna. We just did a great case study with SoFi, and I'm really proud that we raised their net promoter score by 33 points just because it's just so delightful. It's really fun to see all these different brands across a wide range of industries get so much value from their agent. We're the leader in the space. You talked about the metrics. We reached $100 million in ARR in seven quarters, $150 in eight quarters. We're, I think, around $165 now, one month into our next quarter. So growing really rapidly and really proud of the momentum that we have. [00:21:27.17] John Collison That's super cool. What is the typical adoption? Are people using it for email chat support because that's the easiest modality? Do they adopt it for everything, including phone and stuff like that? [00:21:39.18] Bret Taylor It's changed a lot over the past two years, but I'll say the median customer, and then I'll describe some interesting outliers, which I hope are glimpses of the future. Most will start with one channel and a few use cases. At a lot of healthcare companies, phone remains the dominant channel. So say, "Hey, for a few types of phone calls, let's have the AI agent take them and see how it does." Do people like it? Are people comfortable with it? Does it lower our cost? Does it raise whatever metrics? Usually, it's customer satisfaction. Does it work more effectively? For example, for a car insurance company, it'll be a first notice of loss. I got in a fender bender, and that'll be the typical way you start. For a lot of more digitally-native companies, they'll start with chat, and similar. But almost all of our clients will do both. SiriusXM, if you call them on the phone, their AI agent, Harmony, which I love that name for SiriusXM, will pick up the phone. If you go to their homepage and you see the chat, that's also the same agent. The neat part is, I think it's pretty neat because you have literally all of your, I'll say, customer experience team or whatever you might call it at your company. They can spend all their time on one thing, and it actually works over WhatsApp, it can work online, it can work in your website, it can work in mobile app, it can pick up the phone. [00:22:58.14] Bret Taylor That's a pretty big change. A lot of our clients, when we start working with them, they'll have a digital team and a call center team and all these different teams. We've gotten to the point because we've digitized the last remaining analog channel, which is the telephone, those are all unified. When I look at a glimpse of the future, we have a few really ambitious customers like Rocket Mortgage, great Detroit company. They own Redfin. They bought a mortgage services company called Mr. Cooper. If you go to Redfin, you can search for a home using an AI agent. If you go to Rocket.com, you can originate a mortgage with an AI agent, and you can service that mortgage. [00:23:33.08] John Collison It becomes product usage rather than just customer service. [00:23:35.20] Bret Taylor Really end-to-end. Sales, service. I think that's really exciting. Our whole view is that if we were in 1994 and you were doing Cheeky Pint about this internet phenomenon. [00:23:47.13] John Collison That would have been fun. I was a bit young, but yeah. [00:23:49.12] Bret Taylor Yeah. I would have my Nirvana shirt on. We would be talking about, like look, this is your digital front door, or maybe we wouldn't have the prescience to say that. [00:24:01.00] John Collison On the information superhighway. [00:24:01.22] Bret Taylor On the information superhighway. I think the same is true of most companies, AI agent singular. There are lots of agents, but the one with your brand at the top that your customers interact with is special, and that's the one we're trying to power for companies. [00:24:17.18] John Collison You think this, what customer is built on Sierra, your aspiration is that it just becomes sometimes the primary way people deal with the company? [00:24:27.04] Bret Taylor I think a company's AI agent will be the vast majority of their digital interactions. I think digital has come to include the telephone. That's a big shift because we think of that differently. That's a huge change just because the bigger shift... Customer service, which is one big part of what we do, but not the only thing we do, has traditionally been thought of as a cost center because it's really expensive. I'm sure you have people answering the phone for your clients, and depending on where they're located and how well-trained they have to be, like how simple is the case? It can cost $10, $20. It can be much less if it's a more simple case. You have some customers who pay you millions of dollars, and you'll answer the phone anytime they call, and you might have one that has not even started monetizing yet. You might want to call them, but there's a limit to literally how much you can afford to talk to that person and still have a profitable business. I always joke, it's probably easier for you and me to call Sundar than to get Google customer service on the phone. [00:25:28.17] John Collison It's very hard to get Google customer service on the phone. [00:25:31.01] Bret Taylor But it's not because they don't like you. It's just if you think about the average revenue per user of Google, they literally just can't afford to do it. Now, if you take that $10 or $20 phone call and you make it 10 or 20 cents, and over from one to two cents. All of a sudden, not only can you afford to provide a great customer experience to more people, even less profitable customers or in lower-margin businesses, which I think is very exciting. It's not just doing what you did before but new. [00:26:01.05] John Collison You can provide better customer service. [00:26:02.12] Bret Taylor You really can. Then just think about running a subscription business where you care as much about customer acquisition, you care a lot about churn because that's how your lifetime value equation works. You think about, "Okay, if I had a budget of how much I spend on service, and now I can do 100 conversations more than I could before, can I actually reduce my churn rate? Can I improve lifetime value?" Then the interesting thing is then you realize that, "Wow, all of my competitors have access to the same technology." And then you're saying, "Okay, what are my competitors going to do to actually take my customers away from me?" [00:26:36.05] Bret Taylor Then that's where you start to get things like, the ATM machine didn't actually reduce bank branches because some bank had the great idea of, I'm going to put different people in this branch. It'll generate revenue. All of a sudden, it wasn't job displacement, but something completely different. I think the exciting part in our world is you're taking something that was just so expensive that people literally hid their phone numbers so people couldn't call them, and you're making it inexpensive and delightful. The thing I'm excited about is the second and third-order effects are going to be really interesting and very hard to predict, and that's pretty exciting. [00:27:14.19] John Collison I want to come back to this idea of the agent as the UI because I find it really interesting. We talked about this a bit in our annual letter in the context of agentic commerce, where, again, I think people are trying to pitch too much of the end state of fully autonomous, the robots just choosing a few. The point we always make is, let's just start with not having to fill out the web form. No one likes filling out forms on the internet. [00:27:34.11] Bret Taylor Speak for yourself. [00:27:35.21] John Collison I wonder just, will using websites have been actually a bit of a... It's like the fax machine. We used to hum emails over the telephone lines as a way of transmitting information. I wasn't working for this, but there was an era of voicemail memos. Were you ever in the working world for that? [00:27:56.09] Bret Taylor Some people still do this where they do the voice. [00:27:59.10] John Collison Companies will blast a voicemail memo to employees at the company. That's a way of distributing information. All these things are very moment in time. Maybe navigating websites and filling out forms was a bit of a moment in time. Is that how you see things playing out? [00:28:13.23] Bret Taylor I don't know. It's interesting because if you look at the past few iterations of technology, you had the PC revolution, then you had the internet and the browser, then the smartphone came out and the tablet. I was more optimistic about tablets than the way the world turned out. I see more tablets on airplanes, but I'm guessing if I walked around Stripe, I would see very few tablets out. Similarly, there's more smartphones than people, but there's still about 2 billion PCs in the world. I think it peaked some number of years ago, but it hasn't gone down as far as I know, and I haven't tracked this. That's interesting. We added to our digital world. But I think perhaps the more interesting metric is for you and me, what percentage of emails were sent through each device? Certainly from 2010 to 2020, most of the world might have transitioned from percentage of email on desktop to smartphone significantly. It's almost like market share of digital interactions, which I think is a really interesting way to think about it. Certainly, as you think of Stripe's business, where does commerce originate? You saw that move to mobile. [00:29:25.22] Bret Taylor But it doesn't mean that people… It's actually a very big… You wouldn't want to eliminate the PC commerce business. That would actually be catastrophically bad. Then you look at AI agents, and I believe most businesses, it will be their primary digital interface. It's because it works over WhatsApp and it works over the phone. If smart speakers make a comeback, they'll work over smart speakers. [00:29:48.23] John Collison Which they may now. Maybe smart speakers were just too early. [00:29:51.18] Bret Taylor Yeah, well, it's like ask for the weather just turns out to be not the biggest market in the world. But now— [00:29:56.04] John Collison Set a timer. [00:29:58.18] Bret Taylor Set a timer. I mean, it's amazing they made I'm not much money off a timer-setting speaker. It is very future-proofed because it's fundamentally conversational. But maybe it's going from punch cards, mice and keyboards, touch screens, now voice and chat and probably 3D-immersive at some point. Does it just add and make the other ones less important is probably the way I think about it. I do wonder if we'll see the end of the smartphone at some point. It doesn't seem anywhere close to right now, but it is interesting. I think most people don't love how much we're addicted to staring at this glowing screen. On the other hand, you can't talk to TikTok. It's fundamentally visual. But I wonder if there's a world where you could actually be really productive without such an invasive device on your body. If that's the case, can it offer an opportunity to unwedge some of the addictive properties of these technologies and get a lot of the benefits from it. Because at least for me, I think all of us are so connected, you end up like, "I'm going to check my email," and then you're like, "Where have I been for the past hour?" [00:31:08.04] Bret Taylor The fact that we actually have technology that affords that kind of innovation now, I think that's quite interesting. I don't know what the future is, but I'm very excited for it. I know that sounds really cheesy, but we've now changed the ingredients available, and we have a lot more recipes we can cook, and I think that's very exciting. [00:31:26.04] John Collison I agree with you. I'm excited for not having to look at the screen for as many things for a variety of reasons. When a customer installs Sierra, I know there's a significant customer satisfaction component as well as cost, but I'm curious what cost difference does it make and maybe relatedly, when a customer is fully deployed, what kind of mix do they see between queries fully resolved agentically, things that end up having a human who is presumably somewhat AI-assisted, but just what does a normal equilibrium look like there? [00:32:01.17] Bret Taylor It turns out most of our clients have pretty different priorities. Some are very focused on cost savings, and you can automate very high percentages of your cases. There's a company called Ramp that's a really impressive tech firm. [00:32:17.06] John Collison We just had Eric here. [00:32:18.23] Bret Taylor Oh, that's great. Well, they're automating 90% of their cases. They're really sophisticated, though, because they're basically getting in front of cases before they escalate. But I think it's an example of just a really fantastic company implementing really well. You can see anywhere between 70-90%, which is really incredible. The interesting thing, though, is there's counterintuitive effects to it. The cases that do make its way to your customer service team can end up more complex, by definition. What's called average handle time will actually go up. We heard from one of our clients that actually their satisfaction of their call center agents went way up, too, because it turns out it's way more fulfilling to solve a hard problem. [00:33:05.04] John Collison Totally, than like "Have you tried plugging it out and plugging it back in again?" [00:33:07.09] Bret Taylor Exactly. The other interesting one, we had one retailer whose total volume went up almost as much as they saved from the AI agent. [00:33:19.01] John Collison Jevons Paradox. [00:33:20.01] Bret Taylor It was a form of that. If you've used a chatbot from three years ago, they were really annoying. Three years ago, if you said, "Do you like chatbots?" Zero people would say yes. [00:33:32.12] John Collison It's so funny that there was a Silicon Valley wave of hype around chatbots. It was even earlier than that. It was like 2018. It was like pre-LLMs, pre-transformers. [00:33:39.15] Bret Taylor They were just multiple choice machines or something. It was just the worst products of all time. Replacing it with something that was delightful. People are like, "I'm going to talk to this thing a lot more." They ended up keeping... Their cost didn't really go down, but their volume of customer conversations went up 2 or 3X. The CEO was incredibly happy about it. They were like, "We're now actually listening to our clients." It sounds funny, but it's a little bit of a choice how much you want to drive cost savings with AI versus other metrics. Most of our clients are interested in the top-line metrics. Given if you could save $10 or save $1 and improve your net promoter score and competitive positioning by a meaningful amount, everyone in the world would choose the latter. That's the interesting thing going on right now because, again, going back to, it was in 1994, and we're talking websites on this podcast. I think if you were to go to a major bank and say, "If you launch a website, you're going to have a competitive advantage against every other bank." With the benefit of hindsight, that would have been over-promising, the correct thing to say was, "If you don't launch a website…" [00:34:54.20] John Collison Totally, you have to have a website. [00:34:56.06] Bret Taylor This technology is broadly available. As a consequence, you can't just launch… In all parts of AI, not just our business, you can't just launch AI, absorb the cost savings, pass it on to shareholders, unless you have a monopoly. [00:35:12.05] John Collison In most businesses, it's just a consumer surplus. [00:35:14.17] Bret Taylor Exactly. You're either going to lower prices. But I think that's why it's an overused analogy. But the ATM bank branch thing is really interesting because if every single company in an industry has access to technology, I would say it's an imperative, not a competitive advantage. The more interesting, I would say, board discussion is when everyone adopts the obvious things, customer experience, customer service, software engineering, legal. Just pick the ones where there's off-the-shelf solutions available now. What will the industry look like? My guess is you could ask ChatGPT, think. My guess is there's some really interesting second-order effects. When you have competitive markets, you're going to end up investing, lowering prices, whatever it may be. And that's the thing I don't think gets talked about enough. I actually think that it happens with every technology change. You project it through the lens of what you're doing today, and you don't take into effect. It's like a multiplayer game that we're all in right now. That's fascinating to me. The change is disruptive, but I think it's going to be like... I'm very excited for the next few years as the world absorbs the technology, we start getting to some of those second and maybe even third-order effects. [00:36:29.13] John Collison What's the most impressive AI adoption or AI native behavior you've seen from a client? [00:36:36.07] Bret Taylor Oh, that's a really good question. I'll probably say Rocket, where we have a really great relationship. Varun is their CEO, Shawn Malhotra is their CTO. Two people who really are, I would say, not only just curious about AI, but very interested in transforming the home ownership experience with AI. I don't know when you got your first mortgage, but it's super... It's very intimidating. [00:37:05.09] John Collison It's not a modern process. [00:37:07.03] Bret Taylor It's not a modern process. They literally call it mortgage folders for a reason. It used to be a folder. For me, it's an example of a company trying to transform an industry. The reason I brought it up in the context of our last question is it's not just saying, "How can we take AI to do this?" But it's like, if you were to think about the homeowner experience from searching for a home on Redfin all the way through servicing it, and you had AI available, what would the ideal experience be like? It's just really interesting to see Rocket with their acquisition strategy to integrate that experience. That's what I'm excited about. I think there's an opportunity for CEOs and industries like that to have a bold vision of what the future could be. Going back to my point, imperative and not competitive advantage, it is a competitive advantage right now. If you imagine, I haven't tracked the market share of all the big US telcos, but if you look at T-Mobile, Verizon, AT&T, and you tracked it over the past 10 years, you end up with surges in market share growth. The iPhone came out, you ended up with 5G, and you end up with these things. [00:38:18.17] Bret Taylor But my impression of the industry is you end up with these moments that drive market share, and then it ends up at an equilibrium. What's interesting about it as the iPhone moment for telecommunications companies like SoftBank in Japan. This is the moment where perhaps if you have a competitive equilibrium, you can absorb this technology, use it, and you'll have this window where you can actually shuffle the deck a little bit. [00:38:42.18] John Collison This is a technology that shakes the competitive equilibrium. [00:38:44.13] Bret Taylor Yes, exactly right. We definitely notice that. [00:38:47.20] John Collison You talked about how coding is such a domain that is suitable to AI because all of the context you're working with exists in the repo. It is in text. It's neatly organized to be executed and read by humans, and so there's a good balance there. The problem is that customer service agents are not of that character. How do you actually smush everything into a format where your AI agent can answer it? [00:39:28.15] Bret Taylor We spend a lot of time thinking about that. To some degree, one of our engineers called it almost like we're creating a domain-specific language for specifying customer experience. What is the mechanism of specifying it? We use this metaphor we call journeys, which is what is a customer journey end-to-end, and what does the agent need to be successful in that journey? What tools does it need to access? What information does it need to access? If you think about the capabilities of an agent, like skills in a coding agent, you'll add different capabilities over time as the customer is talking to you. The key thing that's been a breakthrough that is probably not surprising to the technologist listening to this, but has been a huge difference between those crappy chatbots of four years ago is the reasoning capabilities. We had one client who had acquired three companies, and they had three identity systems, three CRM systems, three of everything. They had this big IT project where they were going to unify all those systems. But I was like, "Why don't you just have the AI agent go in all three of them and just think?" [00:40:40.09] Bret Taylor They're like, "Well, what if there's duplicate data? What if the data conflicts?" They're like, "That's going to fool—" [00:40:44.17] John Collison Sensor fusion. [00:40:45.04] Bret Taylor I was like, "Well, what does a person do?" They're like, "Well, they think about it." I was like, "Let's just do that." That's the interesting thing about these AI agents is they actually... The basic humane, basic reasoning, not superhuman ASI, turns out to be the huge breakthrough in customer experience. The other interesting thing is the innate knowledge of the LLM. You don't want an AI agent to hallucinate, obviously, but Sonos is one of our clients. Do you have a Sonos speaker at your house? You probably have a Sonos. [00:41:18.06] John Collison I have had, yeah. [00:41:19.16] Bret Taylor If a Sonos speaker ever breaks, it's never the speaker. It's always Wi-Fi. That's what I've learned. It's always true of me, too. There's always some Wi-Fi issue. If you wanted to make an AI agent to help you with your Sonos speaker, you obviously can give it all the manuals for the speakers, all the technical stuff. You can give it the device telemetry, all the stuff you need. Do you really need to give it the history of Wi-Fi? Well, now it turns out large language models have encountered every possible Wi-Fi problem. Why does the Sonos AI work so effectively? Well, it knows a lot about Wi-Fi in addition to all the Sonos things. If you look for any given AI agent, it turns out being trained on all of human knowledge is useful as a starting point for a lot of tasks. I think that's been the big breakthrough. How do you give it all of its knowledge? Well, first, we've built, I think, the best platform in the market to do so, where you can really narrow the guardrails for regulated conversations, widen them for less-regulated conversations. But the fact it starts with knowledge of obscure Wi-Fi idiosyncrasies turns out to be the greatest breakthrough of all that. [00:42:21.18] John Collison Have you had the opposite problem where there's a customer whose problem domains mostly don't exist in the public internet? It's like, "We provide the drill bits used in deep sea oil drilling." It just turns out there's nothing on Reddit about that. [00:42:35.14] Bret Taylor Yes, 100%. We work with this medical device company, and it's a deep cut of human knowledge. You can train it all on that. In fact, we do a lot. One of the things you want to be really careful about is you have a really well-known brand. We work with... I want to say a third of our clients have over 10 billion in revenue, over half have over a billion in revenue. Most of our clients are actually quite well-known. One of the challenges when you're offering either sales or service or customer experience to a really well-known brand is it's harder to ground it. It's actually easier when the internet has never heard of you and you want to make a well-grounded agent, it's actually pretty easy because there's no temptation from the LLMs to go off-script. Actually, I would say— [00:43:20.09] John Collison They know what they don't know. [00:43:21.05] Bret Taylor Ironically, the harder challenge is when it's a very well-known brand, it's like, "No, I got this." I'm like, "No, you don't. You got to go look it up." That's actually a harder problem. [00:43:28.23] John Collison How do you force the LLMs? Mechanically, how do you force them to not answer off the top of their head, but actually look it up? [00:43:39.05] Bret Taylor We call it a constellation of models. Our platform, we call it Agent Studio, you essentially configure the goals and guardrails of a process. Goals and guardrails, not the sequence of steps because you want agency but you want guardrails around it. Within that, we'll use reasoning, but we use supervisor models to actually inspect that reasoning. If you were an AI agent in Sierra and you decided to go off-script, "I got this." What would end up happening is a supervisor agent would observe your reasoning, say, "I think John should have actually looked up the policy here" and send it back with notes and say, "Actually, you're not allowed to make that decision. Here's the reasons why. Go redo that decision." It's a really effective technique. The way I think about it, which is a little simplistic, but I think basically right. If you imagine a reasoning system is right 90% of the time but has some either guardrail, malfunction, or hallucination 10% of the time, it's obviously better than that. Then you have a supervisor that's right 90% of the time. If you chain them together, you get 99% effectiveness. That methodology of layering reasoning and intelligence has been really effective. [00:44:58.04] Bret Taylor In general, it makes sense. You're basically layering, compute, you're layering reasoning on top of it. What's the neat about it, though, is we can abstract that complexity from our clients. They're expressing the goals and guardrails, and we have all these evals and tests and all these other things. We can find ways to make it more and more robust over time, but it doesn't require you to prompt engineer, write in all caps or whatever the hacks that people use to get these things to be conformant. [00:45:27.12] John Collison You started in '22, '23? [00:45:31.15] Bret Taylor We launched the company on February 13th, two years ago. I guess a little more like our— [00:45:38.21] John Collison Oh, '24. [00:45:39.01] John Collison Yeah. Our two-year birthday was a couple of weeks ago. [00:45:42.11] John Collison What I was thinking as you were saying that is, did you co-evolve chain of thought and RL and some of these things that are now in the models, but did you have to build your own janky version of them before they were in the models? [00:45:55.08] Bret Taylor Yes. I'll also talk about just the weird part about building a product right now in a company right now, because so much of what we write, we plan to throw out later. It's just a very weird way to build a company. Google's chain of thought paper, which preceded o1 and doing reinforcement learning on chains of thought, was out roughly when we started the company. It was an earlier paper and effectively provided a substantive basis of why asking a model to explain its reasoning step-by-step produced more robustness. We use chain of thought all the time, and it was a methodology we used. Then OpenAI very innovatively came up with the idea of we could do reinforcement learning on those chains of thought, which is where o1 came from, and then most labs are doing that now. We'd throw out things all the time. You do it and you're like, "Okay, the model just does this for us now." We work with a lot of financial services firms. We work with one bank that is a large Hong Kong business, and they speak Cantonese. Okay, well, we need really good Cantonese voice support. It turns out that that's really hard, and there's not an obvious model that does that. [00:47:06.02] Bret Taylor We spend all this time evaluating all these models. What certainty would you ascribe to every voice model supporting Cantonese well in three years? 100%? 99%? [00:47:16.21] John Collison Pretty close. [00:47:17.17] Bret Taylor We did all this work. In fact, we, I think, have the best Cantonese support on the market. Great for us, and it's a huge selling point. It's a technology that will certainly be commoditized in three years. A lot of what we think about is going from essentially technology innovation now. I think a large part of why we work with the largest companies in the world is because our technology works. In three years, those same clients will work with us because we have the best product. If you look at the early marketing for early software-as-a-service companies, they'll explain why having multiple tenants in the same database is safe. That was a huge part of their marketing. Nowadays, if you came and you marketed your product that way, people are like, "What are you talking about? I don't care what database Stripe uses." I think we're just at this period where the technology is so immature. It's a very technology-forward conversation just because people are figuring it out. [00:48:19.10] Bret Taylor Just like when Netscape's business was monetized through a web server back 100 years ago. It will evolve from being a technology-forward conversation to a product-forward conversation. The interesting part about building an applied AI company is you can't have the luxury of waiting for all the models to catch up with your aspirations. But you know they will. You have to have the best technology and have to be comfortable with throwing it out. It's a real momentum and pace of innovation game rather than thinking of this as precious intellectual property, if that makes any sense. [00:48:52.16] John Collison It absolutely does. But isn't this organizationally hard where, if I'm the head of Cantonese language at Sierra, my incentive, and not disingenuously so, I'll notice all the corner cases where the models aren't that good at Cantonese yet. Obviously, we saw this in prior tech waves, where the cloud adoption laggards were companies that had their own on-prem stuff, and they had a million reasons—half real, half fake—as to why cloud did not suit their business purposes. But how do you avoid getting stuck in this mode of thinking where, "Oh, well, their chain of thought doesn't do what we need," is the classic thing you would hear from someone within the organization. [00:49:39.12] Bret Taylor It's a huge shift. Going back to the first thing we were talking about, it's hard for me to not care about the elegance of the source code, which I think is an impediment to my fully realizing being a software engineer in this new world. I think teams that start to treat the code that they wrote as precious, that has been obviated by a general-purpose AI model, will fundamentally fall behind. [00:50:08.11] John Collison Public markets deem the software industry 20-30% less valuable than they did maybe three months back. [00:50:18.10] Bret Taylor A day ago. [00:50:18.19] John Collison Yeah, exactly. Very recently. The two sides of the debate are, one, the valuations were based on what the businesses will do in in 2030 or 2035, like far in the future, and just there's much more uncertainty there, and so this is deserved. The counterargument is that it's still not the case that agentic software production is really going to build you a Workday. Indeed, Anthropic just installed Workday, very famously. Where do you net out on, is this a rational response or not? [00:50:53.06] Bret Taylor I think it's rational, but I think it's a bit overblown at the same time. I think it's rational just in the sense that there hasn't been more uncertainty in this market ever. Unless you have a strong thesis about an individual company, my guess is, will these companies be less valuable 10 years from now? I think the answer is probably yes. Will that be true for every individual company? I don't think that's true. If you're just thinking about a portfolio of investments, I think it's an indictment of the sector more than it is an indictment of an individual company. I don't know if the value of these platforms was who could vibe code it in a weekend ever. Not that we knew what vibe coding was. My point is everyone who's ever built a software-as-a-service application has had a Hacker News comment of, "I could have coded this in a weekend." Every single one. Famously, Dropbox, I'm sure you have as well. Every single product I've ever made. It just happens. It's like a write-up. In fact, if no one says it about your product, I'm sorry. It's not relevant. [00:51:57.01] John Collison It's not relevant. [00:51:59.02] Bret Taylor Obviously, I see most of those comments were incorrect, but if you think about all the work you've done in compliance or the relationships you have with large financial services institutions, the work you do on fraud, the things under the surface that aren't the forms and fields in the web browser are actually incredibly valuable. If you think about a large software company, they'll have thousands of quota-carrying account executives that represent sales capacity, which is basically a channel, and distribution turns out to be a very important part of software. There's social proof. There's the old saying, "No one gets fired for buying IBM," which few people say right now, though IBM is actually doing really well under Arvind. You want to be maybe the first healthcare insurance company to adopt something. There's another healthcare insurer who says, "I want to be the fifth. I want other people to prove it." There's all these network effects around these businesses and scale and moats and Silicon Valley speak around them. I think the big risk is where is value in the software industry n years from now? One risk is that more people will build than they do now versus buy because the marginal cost of writing software goes down. [00:53:15.13] Bret Taylor I think that'll be true for some software, particularly developer platforms and things like that that are already being consumed and purchased by other engineers. Little libraries or— [00:53:29.02] John Collison Things that already were part of a build-versus-buy calculus, it shifts the balance of power. [00:53:32.19] Bret Taylor Absolutely. The other part of it is systems of record. I think these systems of record have always been the gravitational center of their relative solar systems, and it roughly breaks down by department. ERP systems are associated with the finance department, and SAP and Oracle and Workday have ERP systems. You have Adobe in the marketing department, and you had Salesforce in the sales department, and you had ServiceNow in the IT department. Everything rotated around them. Why? Well, first, their database was truly the system of records. Every application that wanted to interact with the data in that had to... You essentially collect taxes from your ecosystem. And then similarly allowed each of those systems of record company to essentially have revenue expansion opportunities to go to adjacent areas where they're all sold to the same buyer and all of that. The thing that's really interesting is AI agents are actually performing valuable labor. Is the database in the system a record? [00:54:41.06] Bret Taylor Does that continue to be the gravitational center of each of those workflows? I'll just take marketing as an example. The database of your customers that you use to drive sending out an email blast on Black Friday has some value. But if you had an AI agent that drove way higher, more leads for your sales team from that marketing blast, you probably... That's worth more to you than the system of record itself. Similarly, if you imagine, I'll just take a CRM system, and you think about the AI agent that's carving your territories if no one ever logs in to actually do it manually, all of those things have a lot of value. And a lot more value, relatively speaking, than they did because they're actually performing the action. The real question to me is, does it upend this? I would say something that's been true for 30 years, which is all the values in these systems of record. The way I think about it is agents are to some degree a system of record of a process of generating a lead or auditing your financials or reviewing a contract or whatever it might be. I don't think we've ever had a piece of software like that. [00:55:58.20] Bret Taylor Will those encoded, well-optimized processes start to have more value than the databases? I don't know that's the case. For example, if your ERP system is your company's ledger, that'll have a lot of value. But I wonder for all these others, and my theory is, the closer you get to, literally the database is the value, i.e. a ledger, the more durable it is. The closer you get to be in a system of engagement, the less durable it is. [00:56:27.12] John Collison That's a very interesting framing. It gets back to the point you were making about the company that was looking to standardize and not have three different ERPs and stuff like that. You're like, "Why? Just try not doing that." I think maybe the consumer example of this is I think people have probably had the experience of you paste data into an LLM to do something with it, and the formatting is all messed up and the tabs and the spaces don't come through and everything like that. It doesn't matter. LLM doesn't care. You can just paste it in whatever and it'll work with it. This idea that, as you say, if the system of record is important because it's your general ledger and it matters to the auditors, that's one thing. But if it was a system of record in this all your data in one place way and because it was easier to build incremental software atop it, maybe that advantage is going away because the agents are fine plucking data from 10 different places. [00:57:26.11] Bret Taylor That's roughly my view. But the bull case, I always mix up bear and bull. I need to spend more time on Wall Street. The bull case, though, is I think all these companies have a right to win. They're all big, they still have sales capacity, they have all these advantages, but it's a race. How fast will smaller companies build differentiated, scaled businesses before the incumbents go into this new world? But for a wide variety of well-documented reasons, like disrupting their own business model, it's harder. But I think your ask is, "Is it irrational?" I don't think it's irrational. I think there's just more uncertainty now than there's ever been. I think markets are telling you there's a lot of uncertainty, and that's why you see people recede from the whole category, basically. [00:58:16.16] John Collison I feel like there's also a totally separate thing playing out here, where for a long time, certain companies were criticized for not taking profitability that seriously. And at some level, there's just a return to normal valuation levels on a fully loaded, stock-based, comp baked in and everything basis. That's independent of the AI thesis, but maybe just some return to, again, on a fully loaded GAAP basis, more normal valuations. [00:58:48.03] Bret Taylor Well, essentially, if you look at a traditional software-as-a-service company, the way most people model it is you have annual recurring revenue, which is basically an annuity, and it should throw off that much cash every year. Then you have attrition, which is subtracting from the annuity. Then you have net new ARR, which is adding to the annuity. Your salespeople sell software to add to the ARR. You typically have an account management team or customer success team to keep churn down. You grow that annuity and you grow your headcount, often just a little bit ahead of that annuity because you need to grow a new business. If that annuity is not an annuity… Yeah, that makes sense. [00:59:30.11] John Collison Then that math really changes. [00:59:31.13] Bret Taylor It really changes. Because the whole idea of software-as-a-service is you can just slow down hiring and you become very profitable because the annuity starts throwing off cash. That's been the thesis of every private equity firm who acquires slow growth software-as-a-service companies. If you don't assume that that revenue is going to be there two years or three years from now, your discounted cash flow analysis looks pretty different. I don't think it's actually It's quite so dire in the time frames that people think. But again, if you're asking markets are, there's all those great quotes about weighing and I don't know. [01:00:08.16] John Collison Voting/weighing machines. [01:00:09.16] Bret Taylor Voting/weighing machines. I get it. There's probably more safer sectors to invest in, but I don't think it's an indictment of individual companies. That's my point. I actually think... When we first met, I doubt either of us had an extremely positive view of the future of Microsoft. At the time, it felt a previous generation company. Now you look at Azure, their OpenAI relationship, all these things, what an impressive turnaround. I think any one of these companies could do it. I think it's just... [01:00:41.06] Bret Taylor But it's more of an indictment of the market. [01:00:43.20] John Collison I have a lot more questions. Would you like another Guinness? [01:00:46.05] Bret Taylor Sure. [01:00:50.16] John Collison Brett has been through a few platform shifts, and one thing he's been pretty consistent about is being mindful of the external forces that are shaping the ecosystem you're in. He talks a lot about building with the broader wave of AI agents in mind. Stripe Sessions is our way of helping builders see that wave up close. What's changing in the internet economy, what's actually working in production, and what the next era of software looks like when agents are running real commerce workflows. It's not the usual conference fluff. It's insights into what the fastest moving companies are actually up to. If you want to experience the next chapter of the internet economy firsthand, join us at Stripe Sessions this April. Use the code Cheeky Pint for 50% off a conference pass at sessions.stripe.com. [01:01:35.00] John Collison You talked about business models. Are you guys usage-based? How are you innovating on the business model front, or are you? [01:01:39.19] Bret Taylor We are trying to. We do outcomes-based pricing. For a customer service context, that means if the AI agent resolves the case, no human intervention, there's a pre-negotiated rate for that. If we do have to escalate to a person, that's free. For sales, it would be a sales commission. And wherever possible, there's a way to align our interests with our clients', we choose it. I'm a huge believer in this. I think the analogy of going from impression-based ads to CPC ads is apt. I don't think any ad platform thinks like, "Man, think of all the impressions we're giving away for free." Because when you charge for something closer to a business value, it's actually more valuable. [01:02:21.12] John Collison It's more efficient. [01:02:21.23] Bret Taylor It's a lot more efficient. I think the idea... If an agent's outcome is measurable, it's a really a compelling way both for clients, obviously, because it's aligned with their business. But it's also quite disruptive because most, I'll say, legacy software companies are not necessarily equipped to do it for a variety of reasons I'm happy to go into, but it's just a very disruptive model. [01:02:46.21] John Collison Yeah, there's a few lens you can have on it. One is that you get more alignment, like usage-based is more aligned than other ways of charging. As you say, it's more efficient because you're incentivized to drive the right outcomes. People also make the analogies to it's almost more correct for the labor substitution dynamics that you get, or just because you have real inference costs, you kind of have to do a usage-based model. I mean, do those factor in at all, or it would not be possible, almost, to do a fixed-price contract? [01:03:19.15] Bret Taylor I would actually argue outcomes-based is pretty different than usage-based. Just because think of it this way. If you have an AI agent that is making sales for Stripe to small businesses, and I told you I will sell one-tenth the number of new Stripe GMV, however you value that, but I'll use one one-hundredth of the tokens, you probably wouldn't care. You care about the value to your top line of your business. I would argue there's not a strong correlation between token usage or utilization and value. There may be, but there's not always. There was that infamous, I think it was called Folklore, but it's this website where that Apple engineer used to put just all this Apple folklore. [01:04:07.12] John Collison Folklore.org. [01:04:08.17] Bret Taylor Folklore.org, yeah, I love it. It's like, if you're an engineer, it's a fun site to go to. But there was a story about some new bozo manager asking for lines of code every day, and one of the engineers wrote a negative number as a way of saying, "F you" to the man because he refactored a code base or whatever. I think that is the essence of why tokens are not correlated with value. [01:04:30.06] Bret Taylor They may be, but the idea that they definitely are, I don't think stands to reason. I think usage-based is like charging for storage or something. Outcomes-based is what business outcome is this agent designed to produce and did it produce it effectively? That is really aligning because it creates this whole vertical alignment. As a company, reducing your token utilization for the same outcomes is your problem, not your customer's. That's a great incentive to drive more efficiencies over time. It means that to grow your relationship with the client, you actually have to make your product better and not just theoretically better over a steak dinner. Like better, better. [01:05:13.00] John Collison How do you have usage-based... Or sorry, outcome-based when you move beyond customer service where there's a clear, was this resolved or not, to product usage where people are shopping and yeah, they didn't buy a house there, but they mostly don't buy a house on most website visits, but it was a successful visit. [01:05:34.00] Bret Taylor It's the right question, and there's not a great way to do it for every type of agent right now. You can always fall back to usage-based, which is fine. But in that, over time, it's like, wouldn't it be interesting? I think AI agents should have memory. I think AI agents should drive relationships, not conversations. It would be really interesting to say, could we make an AI agent that actually drives home ownership over time. I think that's actually... It's hard, but it's not— [01:06:04.03] John Collison As an AI agent, you have a territory, kind of. [01:06:06.01] Bret Taylor I think so. I mean, even because it's hard today and we're a pragmatic company, I think it's the right thing to ask, though, because that's fundamentally the value the software is designed to produce. Actually it's a really values-aligning thing. It also changes the dynamics of a software company's relationship to its partners, to its clients, because if you go back ancient history, four years ago, there was a really stark separation between software and implementation and usage. It was the client's accountability to use the product well. It was either your IT team or a systems integrator's responsibility to implement the software. The job of the software company was just to make it and throw it over the wall. [01:06:56.03] Bret Taylor Obviously, it's not exactly that, but that was the market we were in. Everyone had good intentions, but what's the saying, "Success has a thousand fathers, failure is an orphan." When the software didn't go well, everyone was blaming everyone else. The client was like, "I'm using it just fine. It was implemented poorly." The person to implement it was like, "No, the platform's broken." The platform people would say... It was like everyone's pointing at everyone else. What's nice about outcomes-based, whether or not the client sets it up, you become more accountable to help them be successful because until they do, they can't use it. If there is some long last mile of implementation, creates a strong incentive for the software company to have skin in the game to just help you navigate that last mile. I think so many of the problems in the software industry are due to that lack of accountability. If you talk to any company who's ever implemented an ERP system, it's like a multi-year process. [01:07:55.01] John Collison It's invading Russia. [01:07:56.22] Bret Taylor You don't even remember why you're doing it midway through. You've gone through two CFOs and three CIOs by the time it's done. We're like okay with that. That's just the way software works. My view is just like AdWords changed the advertising industry on the internet, just drove it. I think you can even pay for mobile app install now directly and truly pay for outcomes. I think it's a really positive step forward. It's not going to be possible for everything. You have to have a pragmatism. But I think it's the right way to actually have a partnership. You should share in the outcomes. [01:08:31.03] John Collison You want to wire the company to be thinking in this outcome-based way. In your main customer service stuff, you can do that. In other ways, you might not be able to yet, but you want people to be spring-loaded to be thinking that way. [01:08:40.04] Bret Taylor That's right. If the whole company is incentivized towards outcomes, it's... We're a way better partner to work with because of it. [01:08:48.03] John Collison We find this at Stripe where, again, we have outcome-based pricing. Exactly. It's transactional. But we find there's a lot of uplift we can get on just getting people more revenue and finding ways... We're sometimes hammering customers where it's like, "You should be accepting local payment methods for internationalization," or "You're crazy not to be turning on this feature." But we really feel it because we have the same incentive as the customer. It's like this will be revenue maximizing for both of us. I'm going to ask a very AGI-brained question. I just can't resist. [01:09:18.04] Bret Taylor I'm glad we're on our second Guinness. [01:09:19.14] John Collison Exactly, yeah. Now we get to it, which is you described building stuff that you know you're going to throw away because the model capabilities will get there and you're occasionally... They are developing capabilities that you developed yourself. Isn't Sierra itself "short AGI?" Sorry, I said I couldn't resist. [01:09:38.06] Bret Taylor No, it's the right question. The short answer is I don't know. I mean, the fog of war in the software industry is pretty thick right now. I really believe in the applied AI market, though. I think most companies don't want to buy models or buy software. They want to buy solutions to their problem. If you just go back to the cloud industry, why doesn't Amazon and Microsoft do everything for everyone? There's not really a... By somewhat similar logic, why should any software-as-a-service company exist when you have bigger scale, all this technology? In theory, they could just develop all the software. Actually, many of them have tried. There's actually competitors to Salesforce and almost all the above. I think there's so much nuance in how these companies align themselves with different departments at these companies, solve their very unique problems in very specific ways. That is a mix of product, not technology, but product, go-to-market. It's an ecosystem around it. I think a lot of that still exists because I'm not sure coding the software was necessarily the hard part. Then similarly, I actually think, especially in enterprise software, how you engage with your clients really matters. [01:11:00.00] Bret Taylor I think it turns out that GPT-5 and Claude, whatever version it's on right now, or Opus, excuse me, is sold to a different buyer than the CFO or the Chief Customer Officer, the Chief Digital Officer. It seems small, but it's actually big. I think you tend to see software companies orient around individual buyers within companies. You tend to see consolidation around departments and around buyers. It's possible that you can go beyond those lines, but it hasn't happened traditionally. I think the reason for it is most business users want actual solutions to their problems, and they want a company that serves their unique problems in a very specific and bespoke way. I actually am extremely bullish on applied AI. I actually think we could accelerate. I'll make one statement, which is I think if we paused model development, we'd still have trillions of dollars of economic value. [01:11:58.07] John Collison I totally agree. [01:11:59.04] Bret Taylor That have yet to be realized. I think if we had a mature applied AI market where the CFO could go buy that agent to onboard new supply chain vendors that just worked, we could actually accelerate that trillions of dollars of economic value. [01:12:16.00] Bret Taylor I think not only am I somewhat skeptical that there will only be two companies in the world, I actually think one of the main things impeding adoption of AI is the lack of existence of all those other companies. So many of the startups, particularly around here in San Francisco, are basically doing relatively rote tools around the AI rather than actually building agents for business processes that are boring but important and valuable. I'm really bullish on it. [01:12:43.16] John Collison I guess you help companies ensure that they can always have access to the latest models, which sounds like a minor thing, but the leading model is always changing, and so that's not a trivial... [01:12:53.21] Bret Taylor I agree. I don't know. I'm not sure how much of a long-term value it is. I think it is. I think your customer— [01:12:58.12] John Collison Up to this point, the race is led by a matter of months, right? [01:13:05.03] Bret Taylor Well, every single month, there's a new frontier model, and your customer experience doesn't change that frequently, so you're absolutely right. But I also think there's just a big product, like our clients use it to optimize their sales. That is a product, not a technology. It's very particular to the workflows of people building customer experience teams, building sales teams, and that's really what we're focused on. I think those departments deserve purpose-built software, and I think there will be enduring value there. But it's interesting. It's the right question to ask. I don't think we've ever lived in a world where production of software was easy. Software engineering was the most scarce asset in a company, and now it's the most plentiful, and I don't think we've ever lived in that world. [01:13:50.02] John Collison Well, that gets to one of the biggest conundrums in Silicon Valley right now is what will the shape of AI productivity be? I think there's this strong sense that the AI has gotten really good, and it should change the composition of companies and it should change the hiring plans somewhat. You've seen this in some corners. Block announced their 45%, 50% AI layoff yesterday, and you have some companies not growing as quickly. At the same time in coding, you see a lot of AI benefits. You can argue that either way. You can say, "Engineers have gotten much more productive, therefore, we should hire fewer engineers." Or you could say, "Engineers have gotten much more productive. The ROI on a single engineer is way higher. We now have super engineers that we can hire, therefore, we should hire way more of them because there isn't a fixed amount of stuff for Stripe or any other company to do." Then the AI productivity story in other roles is just a bit less clear because, as we've discussed, AI is uniquely well-suited to coding. What do you make of just... How does the AI productivity show up? [01:15:12.11] John Collison I feel like every company in Silicon Valley is trying to figure this out right now. [01:15:15.18] Bret Taylor Well, first, I think I'll go back to my why I believe in applied AI. I think the atomic unit of productivity in AI is a process, not a person. I don't think AI... I don't know if you have an assistant, but if you do, he or she might help you prepare for a podcast, might help you prepare for a meeting, he or she might also get you a cup of coffee. AI will be really good at the first two, but quite poor at the last one. No matter of AGI, short of robotics, will get you a cup of coffee. I think it's wrong to think about AI as replacing people. In addition to being inhumane, it's just nonsensical because AI operates in the world of digital technologies. I think if you go to an example of even a mundane process in your business, like onboarding a new supplier, think about all the departments and people involved in that. There's a legal department to do a contract. There's some finance department procurement to negotiate the relationship. You probably have IT that's involved to onboard them into your core systems. Then there's usually a business that's sponsoring it. Fairly mundane, happens all the time. [01:16:31.12] Bret Taylor Let's just say you tracked what is the median amount of time it takes to onboard a new supplier, and it was 17 days, just for argument's sake. I bet you could say as a CEO of a company, I want to use AI to optimize that process and make it 17 hours or one day. You could go through, and if you had a product manager on that and optimize every part of it, I bet you could achieve that. But the hard part isn't a person's job. It's actually all the systems and people in between it. I think part of the reason why I think it's been slow to get the productivity enhancement is we ship our org charts as companies naturally. That's the natural state. There's not usually a person responsible for that process. There's the legal team responsible for the contract, there's the procurement team. I think actually we will end up reimagining our companies with the benefit of AI, will we actually think of our companies as a collection of processes, have people responsible them with KPIs who can apply AI. I think I bring it up just because that's my theory of the world. [01:17:35.23] Bret Taylor I might be wrong, I might be right, but I'm not sure companies are set up to essentially absorb the benefits of AI efficiently right now, and we need to do that to really do so. But the bigger point, I think, is that there's the paradox of, well, you want more software engineers. But on top of that, most of the world isn't just digital technology. I think a lot of the people in the AGI community have only ever worked at a research lab or a software company. You look around, you're like, "Wow, AI is going to do all of this." As they walk by the flower shop and get their coffee at the coffee shop, and you think about the local flower shop, if you took all the AI in the world and gave it to that, you gave it super intelligence, how much would it impact the flower shop's operations? Maybe a little. I mean, I'm sure it would help, don't get me wrong, but someone's still clipping the ends of the stems of the flowers, arranging the bouquets, and thanking you on your way out the door, and congratulating you for your daughter's wedding, or whatever it is. [01:18:38.00] Bret Taylor I think if you think about what parts of the economy can absorb intelligence really efficiently, it's certainly software, and we're seeing that already. Finance seems particularly meaningful here because so much of finance today is just digital information. Everything's in digital systems now, not even just crypto. I mean, just everything's in digital ledgers everywhere. It still doesn't touch a wet lab. It still can't do a clinical trial. You still need to get a crate from this country to that country on a ship. As a consequence, I think I'm not sure we'll see the productivity enhancement we see in software in every sector as quickly. Then on top of that, I think companies need to stop just giving Copilot to every employee and be like, "We're AI now," and start to think about from first principles, what are the parts of your business that have a lot of digital workflows? Where can AI have a real big impact? How do you actually set up your company to actually have someone accountable to drive that? That feels like a real big change management opportunity that most companies haven't done. [01:19:42.12] John Collison Just to push on that. Software engineering, I think we clearly are seeing a lot of AI productivity gains. Software engineers have always loved tools and the latest tools and are just headlong diving into it. Then you have stuff like you see in the flower shop, and stuff that requires really good robotics that we're far away from. That will take a while. What I'm talking about is like— [01:20:06.16] Bret Taylor And by the way, I might prefer a flower shop with the florist. [01:20:12.01] John Collison Totally. Yeah. [01:20:12.22] Bret Taylor Just to say it. I'm not sure it solves a problem I have in my flower shop. I might be wrong. I might be unique in that. [01:20:20.03] John Collison I think a lot of the economy is actually white-collar knowledge work, not coding. [01:20:27.18] John Collison Think of finance departments, legal departments, things like that, where you should be able to see a lot of AI uplift and a lot of AI productivity improvements. It just feels like a current course in speed. We're not on track to get those productivity— [01:20:47.04] Bret Taylor Well, I'm not sure I'm right, but I would argue thinking about it by department rather than by process is where it's off. [01:20:54.14] John Collison We can talk about the processes as well. [01:20:55.23] Bret Taylor Hear me out, though, in this, because if you said, "I want to make the legal department more productive, so I want to make it easier to do red lines," and you optimize that. But why is the contract there? What is it for? If you're, for example, onboarding a supply chain vendor and you have hundreds of them, you might actually say actually making an abstract technology for your legal department to red line contracts more efficient is actually a harder, more general problem than for your supply chain vendors, because you might actually have very rigid rules around your supply chain. Let's say you're a CPG company, and you might actually have very specific saying, "Look, if you want to work with us, here's our core legal terms, here's the axis of independence." If you want to make an AI agent to automate that contract, that's actually a much more narrow problem domain that doesn't require general purpose redlining technology. In fact, if you reduce it, you could say, "Well, there are 10% of our suppliers where we let them negotiate their contract, but only for this spend." [01:22:02.21] Bret Taylor Let's have them go through our legal department. The rest, let's do it all with AI. My point on it is, if you look at it through the lens of an end-to-end business process, you can turn science into engineering. I think solving legal through AI, that's a science problem. This is my point, though, which is I think people are going through department by department. Similarly, there's not a person accountable for that end-to-end process. The more you can narrow the domain that you're solving with AI, the more you can build a harness or a scaffolding with existing technology to actually fully automate it. My hypothesis is most companies just aren't set up that way. That's just not how we're organized. As a consequence, we're all optimizing our silo. We're all just installing Copilot. Copilot is great, by the way. I didn't mean to insult it. But it's not actually like... [01:22:52.07] John Collison But it's incremental. [01:22:54.09] John Collison To be fair, that's the thing we're doing. Obviously, good companies did this before AI, continuous process improvement, and it feels like that is the best thing to do. I think what you're saying is there's no such thing as an AI lawyer. Instead, there's improving your commercial contracting. That is a thing that you can— [01:23:15.01] Bret Taylor Even more narrowly, pick one domain of commercial contracting and solve that. [01:23:21.07] Bret Taylor I actually think those are truly solvable. I think the companies that really think about their business that way, I think they can see the value. Again, I'll go back to the immaturity of the applied AI market is probably one of the bigger barriers right now. My hope is that as the applied AI market matures over the next few years, we'll see a step change in productivity. [01:23:47.21] John Collison There is a canonical way to build a Silicon Valley company. You have engineering and product and design, you have this number of ratios of engineers to product managers and engineering managers. Then you have your go-to-market organization, and you have these pipeline coverage ratios, and you have the product marketers, and all this stuff. But I find it interesting how similar so many Silicon Valley companies are to each other because they've all learned from each other. There's a shared recipe and a shared playbook as to how to build a company. It gets tweaked, but ultimately, I think it's a pretty good IP. Certainly, companies are much better off with it than without. How is that canonical template for building a company different post-AI than before? [01:24:32.08] Bret Taylor Yeah, it's a really interesting question. One is, I've always believed in the primacy of tech leads over engineering managers. Both Google and Facebook, where I spent some of my early career both did this well, where in a product review, you weren't just talking to a manager. You were talking to the tech lead and PM who were product manager who were building the product. Whereas if you went to companies that produced worse software, I'd notice you move up the chain of the command, like the military. I think that we will end up with individual tech leads who, because of the existence of AI agents, will become even more important, where if you are a, I'll say, product engineer, I'm trying to find the right word for it—we might invent one—who has taste but didn't necessarily know CSS, who has infrastructure ability, meaning that you understand the basics of distributed systems and debugging, and you understand your customer very deeply. With the presence of Codex, you can produce amazing results. Those people are truly worth a thousand X other people because it's relatively easy to find someone who's a great infrastructure engineer. Not easy, easy, but relatively. [01:26:03.22] Bret Taylor Finding someone with good taste, that's relatively easy. Finding someone who also understands your customers extremely well, the nuances of the problem they're solving. Those people who can combine that will, I think, end up being able to actually produce products like, capital P, valuable products with relative autonomy. I wonder if it will change our view on generalists broadly. I've always identified myself as a generalist just because I've been both a software engineer in a suit, basically, and I've gone back and forth in that world. As companies grow, you tend towards more specialization just because the person who's the Jack or Jill of all trades ends up not fitting in. There's not really a place for them because, okay, well, you're not really the deepest engineer, you're not really the best designer, you're not really product manager. If you've been at the company for a while, we'll give you an honorary something to do, and you have to lead through influence, and da, da, da, da, da. Could that person actually endure as one of the most valuable people at these companies? I don't know whether it's naive optimism or true, but I actually think those people who often exist in early-stage startups are often the people who get sidelined, but in a way that actually really harms the company. [01:27:31.00] Bret Taylor I'm hopeful that in a world of AI agents, those generalists who, again, I think the most important part is understanding the customer need with agency, no pun intended, and empowerment can end up more powerful in the Silicon Valley company in the future. [01:27:43.13] John Collison I've noticed the exact same thing at Stripe. The exact same thing, which is high agency, really caring about customers, just really caring generally. High work ethic people who maybe weren't the best engineers previously or now, those people are massively ascendant, as far as I can tell, because they suddenly got the exoskeleton. They always have the ideas as to what we should be doing, and this is the better way to serve the customers and everything like that. But now they have the way to make all their schemes real. I've really noticed that at Stripe. [01:28:23.12] Bret Taylor Well, it's interesting. You talked about work ethic. It's addictive right now because you can do much with the technology. Everyone I know who's really used it works harder because it's like, "Wow, I could do so much." You think you're about to go to bed. You're like, "Should I get an AI agent to do something? Am I wasting the next eight hours of my life?" That might be a novelty that wears off, but I think it's really exciting. I'm hopeful on the product engineering design side, you end up with these hyper high-agency people who really deeply care. I really like the way you said it, actually. It's right. It's not just customer problems. It's care, period. Just care can end up more empowered. I'm curious what that means for organizational structures. [01:29:13.18] John Collison I think we have a new job role we need to invent. It's like, what role do these people in? [01:29:18.00] Bret Taylor Hyper-generalist. [01:29:18.00] John Collison Yeah, like product managers, but sometimes maybe without a product, like minister without a portfolio. They're just doing stuff, but now they can do much more. [01:29:27.17] Bret Taylor It's almost like product designer, product manager, engineer. That's why I said product engineer, but that means something different. But it's interesting because we've talked about this. You end up where the grass is always greener with org structure. You go functional organization. Okay, we're going to have engineering, product, design. Let's go to business units. They're like, "Wow, that led to silos and worrying factors." [01:29:46.22] John Collison "We're one reorg away." [01:29:47.23] Bret Taylor Yeah, you sway back again. That's welcome to— [01:29:51.10] John Collison Just one more reorg. [01:29:52.08] Bret Taylor Yeah, exactly. "I'm a middle manager now." I think that it is interesting if these people become extremely important. What does it mean to organize around them? I think it does feel like something that will end up flatter just because of the amount of impact an individual can have. That feels really exciting to me, but it feels like a blurry picture right now. Enhance. [01:30:20.21] John Collison I agree. It's very blurry. It's so interesting. You were on the Twitter board during the super interesting takeover battle with Elon Musk. What are your reflections on that experience a few years later? [01:30:39.21] Bret Taylor It was really interesting to be in the public spotlight. I hadn't really experienced that in my career before. I joke that no one really cares about enterprise software. I worked for Salesforce for six and a half years. [01:30:55.07] John Collison Sorry, what's the joke? [01:30:56.07] Bret Taylor I worked for Salesforce for six and a half years, and I don't think my mom knows what Salesforce does. To have something that was not really just a business issue or a technology issue, but in the mainstream. I realized I didn't love that very much. [01:31:14.05] John Collison You actually prefer enterprise software. [01:31:16.13] Bret Taylor Yeah, exactly. I'm like a builder. I like to build things and have people use them. I know it sounds funny and reductive. That's what gives me joy. The one thing I realized is the conflict of it all. However it turned out, victory, defeat, whatever it was. It just didn't... It wasn't something that filled my bucket very much. [01:31:38.09] John Collison What do you make of the fact that in all these head count debates, Elon is now running Twitter with 80-85% fewer people. I think Nikita Bier tweeted recently that all of eng, product, and design at Twitter is 50 people. [01:31:58.01] Bret Taylor It's really impressive. [01:31:59.07] John Collison Yeah, it's been a little flaky in pockets or just at times, but mostly the service works, and they have shipped new features. I think those two statements are undeniable. But just what's your takeaway from that? [01:32:10.23] Bret Taylor I don't know. I haven't followed as much as the… I didn't see that tweet as an example. Do you call it tweet still? [01:32:17.16] John Collison Sorry, I'm retro. I'm old-fashioned. [01:32:19.11] Bret Taylor I don't know about that, but it is interesting right now because obviously a lot of that predated AI. But I mean, any person has been an individual contributor, engineer, knows that the size of the team does not produce linearly greater outcomes. Everyone in the world has experienced that. I think the idea of can you actually give individuals with good taste more agency, no pun intended. I think it's always been an enduring thing. What was the Jeff Bezos—"It's a two pizza box thing." [01:32:55.15] John Collison But then do large tech companies underrate this phenomenon? Do they pay lip service to small empowered teams and two pizza teams, but X shows that maybe they should be doing more? [01:33:07.02] Bret Taylor I think their companies largely act somewhat rationally. I can't remember who the CEO was, but it might have been the Rippling CEO just talking about there's this idea of being lean and agile, and then there's you want to capture market share and grow your product and grow your platform. At the end of the day, you can be clever but not smart. You might be so clever to think, "I'm not going to have anything more than two people on these features." If you have a competitor who maybe does something a little less elegantly but wins, who cares that you are clever with your two pizza box team or two-person team or one AI agent team or whatever it is. When someone said, "We're going to have a X billion dollar company with one person," I think that might have been right, but it's not— [01:33:58.00] John Collison Maybe you could have had a $10 billion company if you'd hired a bit more. [01:34:00.22] Bret Taylor That's right, and I would actually argue the more specific thing is if all of a sudden, for some clever reason, you want to prove you can, the idea that a competitor might have 10 people and beat you is probably more likely than even having a $10 billion company. [01:34:16.03] Bret Taylor I think at the end of the day, when you're building a business, especially one that's in hyper growth, which successful businesses in tech tend to be, if you are too clever and austere, and going back to your point about Silicon Valley culture is all being the same, there are examples of companies that really innovated in culture. You wouldn't think of it this way, but HP, a lot of the traditional open office floor plan came from them. Facebook. [01:34:45.16] John Collison Woz worked at HP. [01:34:46.19] Bret Taylor Oh, I didn't know that. That's interesting. Then Google offered free food to their employees, which a lot of people did. Then Facebook, a lot of the layouts of offices all looked like Facebook for a long time. But then you have other companies like, "We're going to innovate in HR, and they spend all this time and energy on it." In fact, the smart thing to do is just be like, "It's not what we do. Let's just do the same old thing as everyone else because everything is push button. I don't need to worry about it." I do think it's the right question to ask for every technology company. [01:35:22.12] John Collison After being on the Twitter board during the Elon takeover, you were then on the OpenAI board when Sam got fired. Have you considered that you are the problem? You are bringing the drama. [01:35:34.21] Bret Taylor I came in after the drama there. [01:35:37.05] John Collison Oh, right. You joined after. Sorry. [01:35:38.21] Bret Taylor I was brought in as the mediator. [01:35:40.22] John Collison I see. Post the... Okay, your hands are clean. [01:35:43.23] Bret Taylor Maligning my reputation here. I wasn't actually on the other side of it, but I got a phone call, was it Saturday or Friday after? Basically, my understanding was I was the person that both the existing board and Sam agreed upon to help mediate the situation. [01:36:02.16] John Collison What have you learned on the OpenAI board? [01:36:05.08] Bret Taylor A lot. I mean, certainly the most interesting part is the AI research. I've never been affiliated with a true research lab before, and that's fascinating to me. It is very inspiring. I mean, it's very easy to grow, not cynical, but you can look at OpenAI, Google, Anthropic, and say, "Whose model scores better on this leaderboard?" To actually go in and see this company where every single researcher is trying to make safe AGI and not come out of those board meetings inspired is impossible. It's amazing. The other thing is it's the first not-for-profit board I've been affiliated with. That's really interesting as well just because— [01:36:53.04] John Collison It's a different thing. [01:36:53.22] Bret Taylor Well, when I mentioned the fiduciary duty, is you have a duty is you have a duty to commission. That is really clarifying and interesting as well, because when you're making decisions and you realize your sole duty is to ensure that artificial general intelligence benefits humanity, that's really different. It's really interesting. I've never had a fiduciary duty to a mission before. That's really interesting to me because I take those duties really seriously and reflecting in a board meeting and you're making a decision, you think about it very differently through that context. [01:37:26.12] Bret Taylor Then the other thing was because I was brought in after that crisis, there was three people on the board on the other side of that when I agreed to temporarily be the chairman. It's still there. We had to grow the board essentially from scratch. That was really interesting, too. Just to think about… Normally, you add one board member at a time. This one was like, "Do you have a bulk rate?" [01:37:51.15] John Collison "We're going to build a board, put together a team." [01:37:54.03] Bret Taylor You really think about spend time with the other two board members just really thinking about what does the composition an OpenAI board look like? How do you represent the not-for-profit part of it? How do you represent safety? How do you represent the economic impact of AI? We're doing lots of infrastructure investments. How do we find someone with that specific type of financial expertise? That was really rewarding as well, just building a board, not from scratch, but effectively from scratch. [01:38:24.02] John Collison Last question. What are your AI predictions for 2026? [01:38:27.14] Bret Taylor I think we will have some scientific breakthroughs with AI that positively break through into the mainstream press and awareness. We've already had some interesting math proofs, but I joked with one of my friends, "Until I can understand what the title means, I'm not sure it's going to make..." [01:38:49.10] John Collison You're not excited about the interventional manifold space. [01:38:51.23] Bret Taylor Yeah, exactly. It won't quite be like the Apollo landing, but I remember the Kasparov chess match, and certainly things like AlphaGo were really meaningful. Given the progress in math, I'm hopeful we have at least one moment of discovery that is inspiring, because I think a lot of the dialog around AI right now is economic opportunities, but also what could go wrong. I actually think one of the main things that can go right is actually discovery in science that actually can improve the human condition. I'm really excited for it because I think it will contextualize why so many of us are excited about this technology in a way that captures attention. As you said, something beyond an end-dimensional manifold blah, blah, blah. I feel not confident in that, but it certainly feels like the ingredients are there for that. I think we'll continue to see mainstream adoption of AI by both consumers and companies. That doesn't really feel like a prediction, but I think this will be really a year of adoption of agents. We're certainly seeing that in Sierra's customer base, but I think we're going to see it more writ large. Then you've already seen ChatGPT growth, really unprecedented levels. [01:40:17.12] Bret Taylor Things like OpenClaw, you can see that translate over to agents and the more long-running autonomous tasks. It does feel like by the time we exit this year, can that go from a niche community to something more mainstream? It feels probable to me. Then the other thing is, I think most companies in Silicon Valley won't write code by hand. That might seem... It's funny that it feels obvious right now. You're like, "Oh yeah, of course." You're just nodding like, "Yeah, why not?" But if I had said that four months ago, that would have been a bold prediction. But I think that's really interesting just because that's such a fundamental state change. I say in Silicon Valley because I do think it takes a while for these tools to diffuse through society. Silicon Valley is insular enough that I think it will here, but I'm not sure it will happen through every company in the world yet. [01:41:15.10] John Collison The year of agents across businesses and just people finally getting their Claw-style agents, and then all code written by AI as well. [01:41:24.23] John Collison It's a good set of predictions. Great. Thank you. [01:41:25.20] Bret Taylor Thanks for having me.