Bret Taylor on AI and the Future of Software | Ep. 42
We unpack the so-called “SaaS-pocalypse” and what AI agents mean for the future of enterprise software.
Bret Taylor is the founder and CEO of Sierra, an AI agent company transforming customer service. Bret’s legendary career includes being CTO of Meta, co-CEO of Salesforce, chairman of the board at OpenAI, co-creating both Google Maps and the Like button, and founding three companies.
We unpacked the so-called “SaaS-pocalypse” and what AI agents mean for the future of enterprise software. We talked through the shift from systems of record to autonomous agents, outcome-based pricing, platform transitions, Codex and the transformation of software engineering, and who is structurally positioned to win in the next era of AI.
Timestamps:
(0:00) Intro
(0:20) The SaaS-pocalypse and systems of record
(12:34) Sierra's competitive landscape
(17:05) Outcomes-based pricing
(24:22) The rapid evolution of AI support technology
(28:21) Young founders vs. experienced founders
(34:12) Beyond support: The full customer lifecycle
(38:47) Codex and the future of software engineering
(51:49) OpenAI and advertising
(54:59) How to run a board
Links:
https://x.com/btaylor
https://x.com/jaltma
Watch on YouTube; listen on Apple Podcasts or Spotify
Clips
AI’s impact on the system of record
I asked Bret a question that a lot of people are thinking about lately since he's as equipped as anyone in the world to answer:
"Does a system of record have a place in the world if nobody logs into it?"
The “Strategy Tax”
Why can a startup with 50 engineers make up ground against an incumbent with 5,000? You'd think big platforms have the data, distribution, money...why can't they build technology faster than startups can earn distribution?
Bret calls it a "strategy tax" and gives a crisp articulation of exactly why startups have the opportunity to win.
Transcript
Disclaimer: Transcript generated with AI assistance and lightly edited for clarity and accuracy.
The SaaS-pocalypse and systems of record
Jack Altman
Bret, thanks a bunch for doing this with me. I’m super excited for it.
Bret Taylor
Thanks for having me.
Jack Altman
You’re one of the best people to ask this following question. What is your view on the SaaS-pocalypse, if we can call it that?
Bret Taylor
SaaS-mageddon.
Jack Altman
SaaS-mageddon. In public markets, all of these companies are trading way down. You go on X and everybody’s talking about how software can now be written in two seconds so there’s no moats anymore in software. It’s leading a lot of people to ask, where does durability come from?
I wanted to start with this topic because you’ve built your own companies, you’ve been the co-CEO at Salesforce, you’re now building one of the fastest growing AI startups, and you’re on the board of OpenAI. How do you see software in this moment, in February 2026?
Bret Taylor
First, I don’t think the market is necessarily reflecting an indictment of individual companies. It’s more of a broad view of these bigger questions you were saying. Every software stock is down, but I don’t think that means every software company is equally disadvantaged. It’s basically just anxiety about the future.
It’s a few things. We can talk about defensibility broadly, it’s a really interesting question. If you look at the history of enterprise software, a lot of the value has gone to the big systems of record: ERP systems, CRM systems, the core databases that Oracle famously powered in the early days of software. Then you end up with all the software-as-a-service companies: SAP, Workday, Salesforce, ServiceNow.
If you look at what a system of record is, it’s essentially a database with a bunch of workflows around it. To date, those workflows are manipulated by people clicking on buttons in a web browser, filling out forms.
Jack Altman
If you had to synthesize pre-AI, why were those businesses so good? Was it the source-of-truth thing? That there had to be some immutable thing, so the database row… Is that what it was? Was it the ecosystem of integrations? What do you attribute the success of systems of record to?
Bret Taylor
The reason why a system of record has always been the most valuable is that it’s the anchor tenant of your technology deployments. If you wanted to create a workflow for quote-to-cash or something like that, you had to integrate with your ERP system and your CRM system.
As a consequence, the companies that owned those databases could either develop that functionality as an add-on, a new SKU. Or if it was a third-party company, they would often be a part of the ecosystem, like Salesforce’s AppExchange or whatever the marketplace equivalent is for SAP.
You ended up with a lot of value in those systems, which meant switching costs were really high because that system plus all the partners that integrated with it created gravity and high switching costs. Similarly you end up accruing a lot of value either by collecting rent from your ecosystem or developing premium add-ons on top.
It became the sun in the solar system for each of the different lines of business that these systems of record were sold into. Then you’d get scale, sales capacity scale. The larger you grow, the more salespeople you have, you can reach more and more people.
Then there’s the proverb: “No one gets fired for buying IBM,” which obviously is a somewhat dated expression. But it was like, “Hey, if you’re going to put in a new ERP system, no one’s going to blame you for choosing SAP because everyone chose SAP.”
Jack Altman
If you choose something new and it doesn’t work perfectly, big trouble.
Bret Taylor
So all those things accrue. But then the question is, now that a lot of those things start getting chipped away with AI agents… First, could you just vibe code it in a weekend? Does it change build versus buy? That’s one risk. Does it change when you come up on that renewal? Are you going to make a different decision?
Secondly, I actually think the more fundamental thing is: what is the role of that system of record if AI agents are doing most of the work? Rather than people clicking around on an ERP system to onboard a vendor, if you just delegate to an AI agent to do it, all of that is invisible to you. All of a sudden it goes from being an application to a database. Similarly, if you imagine a CRM system, and rather than having people staring at it all day to manage their leads, contacts, and opportunities, you just say, “Hey, generate me some leads.”
Jack Altman
In other words, does a system of record have a place in the world if nobody logs into it?
Bret Taylor
It does, but the real question is how valuable is it? How important is it? Going back to my metaphor of a solar system, how important is that gravity versus the gravity of the agents running around it?
It’s just really interesting. If you imagine you’re running a sales team, how much do you value the database of leads versus the agent that generates the leads? In ancient history—three years ago—those are the same thing. But now you’re like, “Gosh, I actually probably care more about the lead generation. How it’s stored and tracked is maybe a more tactical part of it.”
That’s true of every system of record. I just know CRM systems pretty well. If you look at ITSM, which is what ServiceNow plays in, or ERP systems, which is Workday, SAP, Oracle, et cetera, all these questions start coming up.
What’s interesting though is I think every single one of those companies could transform and benefit from AI. I really do believe that. You saw what Microsoft did in the cloud transformation. They went from being dependent on Windows revenue to Active Directory and Azure and all those other things. But it was really awkward. Folks like you and me back in the day used to probably dismiss Microsoft. I certainly did. I didn’t foresee them becoming as powerful as they are today. But it was good leadership, good technology.
But I don’t think the market knows who is Siebel Systems and who is Microsoft in this landscape of software companies. Probably no one knows what Siebel Systems was. That was the company that Salesforce beat to become the cloud CRM.
So can you actually develop this ecosystem of agents around your platform and will it become more valuable than the platform you had? On top of it, you have the existential risk of, is the value of software just going to zero? I don’t necessarily believe that. But you look at all of that, and if you’re just an investor in public markets, you’re like, “I’m going to sit on the sidelines. I’m going to let the market play out a little bit.” I think that’s what’s going on with a lot of these.
Jack Altman
Totally. You can never know for sure who’s going to turn into the next Microsoft, but you can try to think about who has the structural ability to expand. Who’s got the right, with customers, to make the expansions? And then which products will be easier?
In the database question: is it easier for today’s databases to build agents on top? Or is it easier for a modern agent to say, “I’m going to go build a database at some point because I can do that and I’ve got the customer relationship”? How do you think about what creates the rights to expand?
Bret Taylor
I think all the incumbents have a right to win in a lot of ways. In the same way we talked about why a system of record is powerful, you could say the same logic for all the agents running on top. The dynamic that plays out though, not just with AI, is when a new technology comes out—like the web browser or the smartphone—rarely is the expertise on how to do exceptional things with that technology at the incumbents.
There’s this thing in enterprise software, a phrase called “best of breed” and “best of platform.” Best of platform means, “Hey, we’re a Microsoft shop. We just buy Microsoft stuff.” It sounds silly, but actually there’s a lot of logic to it. You get good procurement leverage, everything works together.
Jack Altman
You don’t have to deal with a ton of people.
Bret Taylor
There’s probably some benefits, all sorts of things.
What ends up happening when new technologies come out is the pendulum swings from best of platform to best of breed.
Jack Altman
It’s much easier to get a 10x experience.
Bret Taylor
A hundred percent. Also just think of pre- and post-web browser enterprise software. You’re running client-server Windows software. It’s a completely different skillset to make a web application, as you and I know. So at the time, there’s this window where best-of-breed competitors are light years ahead of the incumbents and it’s a race. Basically, can the best-of-breed upstarts get scale in time before the incumbents figure out the technology.
That’s what’s going on right now. I would argue very few of the incumbents have any credible, decent AI technology, but they will. It’s inevitable they will.
Jack Altman
You know what I don’t understand? Why is that? What’s the real reason for it? These companies have infinite resources, roughly speaking. They ought to be able to hire who they want. They ought to know what the products could look like. They ought to be able to try them.
Why is it so hard for legacy companies to catch up quickly, versus an AI startup with 50 engineers that seems to outperform the teams that are 10x bigger at a big company? Is it cultural? Is it systems?
Bret Taylor
I like the phrase “strategy tax.” I don’t remember who to attribute that to, but we could pull up ChatGPT and ask.
The idea is that in these moments of big platform shifts, what were your strengths can become weaknesses. Let’s take Siebel Systems and the birth of the web browser. They have an on-premises CRM system. When you say, “Okay, let’s compete with this cloud-native CRM system in Salesforce,” you start to say, “I don’t want to just start from scratch. We’ve got all these assets. How do we do it in a way that takes advantage of all of our assets?”
All of a sudden you’re like, “Okay, let’s not just build a great product. Let’s transition from this product to that product. What if someone wants on-premises too? That’s our strength. We should play to that strength.” You start basically making all these decisions that sound really clever because you’re playing your strengths. In practice, if the technology wave is bigger than the category, which I think the web was as an example, you end up basically chipping away at having a pure-play value proposition.
It can also happen with business models. At that time, you’d have perpetual license software. Moving to software as a service, that’s a huge change for a business to make. For your customers, it goes from being CapEx to OpEx. For you as a company, it changes to ratable revenue. Shantanu did this at Adobe. Very few companies can make that transition. You have to sell it differently. You have to compensate salespeople differently. Revenue recognition is different.
So you have the product strategy tax, you have the business model strategy tax. Even with the incentives of salespeople there’s a strategy tax because you don’t want to just have your business collapse overnight. It’s so easy for a clever Silicon Valley person to say, “Just pivot.” If you’re a public company, you have to go in front of your investors every single quarter and be like, “Yeah, hey guys, I know our revenue just went off a cliff, but trust me, it’s going to turn around next quarter.” You don’t survive that.
You compound all those things and all of a sudden you’re like, why does a 50-person company succeed? They have none of those. All of the advantages that you had all of a sudden become anchors that are holding you back from actually doing the right thing.
That’s why I always like to remind our company, Sierra, that the wave that we’re riding, of large language models and this next generation of AI, is greater than any company riding it. Don’t fight AI. It’s going to happen with or without us.
If you go back to the internet, if we were talking in 1995, we’d probably be like, “Search as a category, e-commerce as a category, digital payments… That’s definitely going to happen.” Google hadn’t been founded yet. Amazon probably had around then. PayPal, probably not founded yet. The categories are obvious. Whether or not any of those founders existed, there would be winners.
Jack Altman
And it’s the same now.
Bret Taylor
And it’s the same now. Everyone knows what’s going to happen and you’re competing for the privilege of winning. In a world where the technology is that remarkably powerful, the strengths of the incumbents start to wither in the face of the technical change. That’s why you tend to get new, great companies. The companies that are enduring tend to be created in platform shifts more than any other time.
Sierra’s competitive landscape
Jack Altman
I’m actually curious on this topic, that there are obvious things. Within AI, not to discredit your insight, but support I would count as an obvious thing, in a good way. It looks like it works and you did it early enough to get to a place at the right time. But other people did too.
In some ways you have been playing in a very blue ocean, wide fields. The incumbents are categorically different. It seems inevitable that we’re going to have agents doing support. And then on the other side, a lot of other companies see the same thing. A lot of other people have been building it.
Before getting into the specifics, I’m curious. Experientially, day to day, does your operation of the company feel competitive or wide open?
Bret Taylor
It feels competitive and it feels like a really big market. It doesn’t feel particularly demand-constrained, which is a really great feeling as a fellow entrepreneur.
Jack Altman
So you feel like there’s lots of demand and there’s a contest with each situation?
Bret Taylor
Yeah, that’s right. It feels like there’s too much capital available. Put another way, there’s obviously going to be competition in meaningful markets. It feels like there’s too many competitors that don’t necessarily have strong differentiation. I think it’s probably healthy though. There will be a culling as the market progresses. But it does feel quite competitive.
I’ll give you a quick glimpse of the past couple years. We’ve had remarkable growth at Sierra. We closed $100 million in seven quarters, $150 million in eight quarters, which has exceeded my expectations. But this past year has felt like an inflection point. The first year of our company’s history, we would often go in and explain to clients what an “agent” was. The term was novel and it was part of our marketing, explaining what an agent was.
Number two, people would be talking about, “Hey, AI is maybe non-deterministic.” They wouldn’t necessarily use that word, but that’s what they would be describing: “How can we trust this technology directly engaging with our customers or consumers? What are the risks?”
Now the conversation is, “Clearly we need this yesterday.” Over a quarter of our companies have $10 billion or more in revenue. We’re talking about big companies. We serve most of the Fortune 20, as an example. These are big companies coming in saying, “We’ve evaluated, we know what we want. We’ve heard of you, we’ve done all this evaluation. Here’s an RFP. Let’s go.”
As a consequence, because the market has matured—illustrated by the existence of things like RFPs—you end up in more competitive conversations. And then it’s a question of, why Sierra? I’m happy to talk more about that. Obviously I’d love to as an entrepreneur. I could tell you all the reasons we’re the greatest. But you end up in this world where you’re not explaining what the word “agent” is anymore. You’re saying, “Here’s why we’re the right partner for you”, which is a very different conversation.
Jack Altman
So they’re like, “Yeah, I’m bought in on an agent. So why is it Sierra?” What have you found is the most important thing that makes you win?
Bret Taylor
One thing we really did uniquely at Sierra—the reason why over a quarter of our customers have over $10 billion in revenue—is we’ve tried to serve more complex, more regulated industries. We serve most of the US healthcare insurance market, as an example. We serve US banks, Spanish banks, UK banks. These are companies that, if you know the industry, are regulated by everybody.
It’s easy to make a demo in AI. That’s why you can go on X and see a thousand demos. Demos are cheap. But making an agent industrial-grade is hard. We’ve really uniquely been able to make agents that can have complex conversations.
The other thing we do really uniquely is, in addition to having a really easy-to-use product, we help companies move faster. We went live with Cigna in two months.
Jack Altman
That’s crazy.
Bret Taylor
Which is remarkable.
Jack Altman
How big is Cigna?
Bret Taylor
It’s a Fortune 20 healthcare company. I was on stage with Sachin, who runs their AI practice, at the HLTH conference, and he was talking about this. Part of that is, how can you show up? We’re really great at AI, Cigna’s really great at healthcare. How do you bring those two together to move extremely fast? So for a lot of our clients, the reason they bring us on is, can you help us move quickly? That requires knowledge of AI, knowledge of business. I think we show up with a greater sense of maturity there.
Outcomes-based pricing
Jack Altman
You mentioned that the pricing scheme was one of the difficult things in the past era. We don’t have to belabor it but obviously, going from just buying a license to a cloud subscription… and now usage-based is the future. What are you feeling is important as you’ve created, and probably continue to iterate, on pricing? What are the important levers for agent companies?
Bret Taylor
We do something specific at Sierra that I’m an evangelist for, which is outcomes-based pricing. It turns out in our industry, the outcome is usually well-defined. In a service context, the outcome is: could the agent solve the problem? In a sales context—we do a lot of sales agents as well—could it make the sale? Your companies paid your salespeople commissions, right?
Jack Altman
Yeah.
Bret Taylor
If you can measure the outcome, you want to incentivize the outcome. The interesting thing about agents is they’re autonomous, or can be autonomous. So if the outcome is measurable and trackable, what an interesting opportunity to actually charge for that.
Look at the history of software. Let’s take advertising. We went from impression-based ads to cost-per-click ads to now, for mobile ads, you can do pay-per-install. At least that’s my understanding. Then you had enterprise software where you went from on-premises licenses to subscription-based software. Could outcome-based software be the next?
What’s so neat about that is, for a company, what an interesting and accountable business model. There are some challenges to it because you obviously put some revenue at risk. But I don’t think most advertising tech people would say CPC ads put revenue at risk. It’s the opposite. The closer you get to the outcome, the more valuable it is for the companies. They’re actually willing to invest in it.
My view is that, to the degree agents have a measurable outcome, outcome-based pricing feels like the secular business model for agents. I think it’s both disruptive and a huge step forward.
Jack Altman
Why is it better than token-based? If those are the two reasonable options now, why is an outcome better than token-based, even over the long term?
Bret Taylor
Let’s say you had an AI agent to generate leads for your sales team. What do you care about? You care about the number and quality of the leads. You really don’t care how many tokens the model uses. In fact, it’s not obvious to me that there’s a correlation between used tokens and leads generated.
In the same way, there’s no correlation in a SaaS product between their cost to serve and the quality of the product. You could have a really good engineer write it or a really bad engineer write it. What you really care about is the quality of the product. The reason why I don’t think token-based makes sense is that it’s charging for an input that is uncorrelated with the output that your clients actually care about.
I’m a huge believer in applied AI, but I actually define applied AI as: can you describe your value proposition without mentioning models? Because if you think about, “Hey, we can answer the phone and solve 80% of phone calls without human intervention, with a CSAT score of 4.8 out of five”, you don’t mention models. Models are an input to that, not an output. If you have to mention token utilization, it’s probably a tool. It’s not an application of AI. It’s just a tool around AI.
The closer you get to a business outcome, you should charge for the business outcome, which is uncorrelated with tokens. I also think it’s almost a measure of whether you’re actually an applied AI company, if you don’t have to talk about tokens.
Jack Altman
Do you think there will be markets where things get so competitive that people have to price based on cost rather than value? Or maybe the other format would be where you can’t describe the outcome cleanly. For example with coding, which we both probably think is super important, it’s a little harder to say what the outcome is there versus usage. What are the conditions where tokens do make sense?
Bret Taylor
There’s this old Apple site where they had Apple folklore. There was this one boss at Apple that made people fill out a form saying, “How many lines of code did you write?” This engineer infamously wrote a negative number because he had just refactored a bunch of stuff. It’s the good historical analogy for why tokens don’t matter. It was his way of saying, “Fuck the man. Lines of code have nothing to do with my value”, and he was doing it to piss off a middle manager to make that point.
It’s interesting because in the world of software engineering, the customers of coding agents right now are software engineers who intimately understand these models. So there’s a bit of a customer-product-market fit. So it’s a nuanced point.
But I’ll say where I see it might happen. Right now, if you’re evaluating a coding agent, you’re probably comparing it to the cost of a software engineer. If you fast forward five years, you’ll probably be comparing it to the cost of other coding agents. So I think the second-order effect as AI becomes prevalent is that the reference point for its value will change.
The thing I would say is that’s true where you’re thinking about a cost center. If you’re thinking about top-line revenue growth, that doesn’t necessarily apply. In my example of an AI agent generating leads for your sales team, depending on what you’re selling, a lead is a lead is a lead. You’ll probably value quantity and quality of leads, and there’s a math equation. That probably will remain independent of token costs, is my guess.
A large part of AI is productivity and reducing costs, and there’s a big part of it. But the other side of it is outcomes. Could you imagine a world in four or five years where there’s one coding agent that can actually produce something of greater value for your company? Will you value that? Or will you just look at the token cost? I think you’ll probably start looking for value. Will they all be the same? I don’t know.
I was just reflecting on the past year. There have been all these articles about whether AI progress has slowed down. In our world of software engineering, it’s been the opposite. Every new model comes out and you’re like, “Oh my gosh, it can write increasingly complex software.” My theory of that is that it depends on what you’re testing. If you’re using ChatGPT for trip planning, you probably haven’t seen a material change over the past year and a half because you reached sufficient intelligence for trip planning a long time ago. If you’re using AI to write Rust code, Codex is mind-blowing right now.
So one of the interesting things when I think about second and third-order effects and the progress of AI is where you pass the horizon where every model is sufficient in that task. Then there’ll be some things where the frontier continues to move. It’s hard to imagine, but we’re in a crazy time.
The rapid evolution of AI support technology
Jack Altman
Where are we at with support agents right now? Are there still edge cases, last-mile things that AI can’t do still?
Bret Taylor
I imagine a lot of the technical problems, as opposed to product problems, will become easier. But there’s still a lot of them. We at Sierra support most spoken languages in the world. If you want to support Cantonese and Tagalog, most of the good voice models don’t come from the traditional Western model companies.
One of our clients is Safelite AutoGlass, roadside assistance. It turns out that car noise, background noise, kids talking in the background, are actually all fairly hard problems to solve. Even in some of the advanced voice mode stuff, if you are in a noisy environment, it constantly thinks it’s being interrupted. So you end up having to build proprietary voice activity detection, multiple speaker detection, all these other things.
We develop all this technology because we need to be the best now and I think we are the best now. You’re like, “Okay, that’s probably going to be a commodity two years from now, one year from now.” Who knows? But you have to do it because you need to be the best at every stage of your company’s existence. The way we think about the world is we have a product called Agent Studio, Agent OS. In three years you’ll judge us by our product. Right now, our clients don’t really put it this way, but we’re judged by the technology.
If you go back to 1996, I remember when Netscape had a web server and Apache was new. No one cares how you serve webpages now. It’s a commodity. But at the time, that was what you sold. Now you have increasingly higher-order website building like Shopify. I just think the AI agent market’s going to take that progression. We’re going from a tech-centric sales cycle to a product-centric sales cycle.
Jack Altman
It’s interesting that you’re obviously having to be the best at something that you know is going to get commoditized. I don’t know if you ever had to experience something like that. For that to be true, you just have to be in the middle of an insane rate of change. But that means you have teams who are putting a lot of their life force for two years into something that everybody knows is just for two years. But it still matters nonetheless.
Bret Taylor
It’s crazy. If you look at traditional enterprise software—consumer’s a little different—you think about building up this asset, your intellectual property. There’s a fancy name for it. “Look at this platform that we’re building. It took so many years to build it. It’s got all these features. Now you’re like, “I’m building this and I’m a hundred percent certain we’ll throw it away in the next 48 months.”
Jack Altman
It’s a sand castle.
Bret Taylor
But I have to build it because if I don’t, I can’t serve the bank that has a big business in Hong Kong, or whatever it might be, where they need Cantonese support. That is the reality right now. I’ve been thinking a lot about this.
I think it was Tobi Lütke who said something provocative. When generating the code is easy, it’s almost like the system and the prompts are actually the durable asset. Put it another way: could you terraform your software from scratch? It’s the prompts that led to it. I do think that is the software of the future in a lot of ways. How do you encode the infinite number of little product decisions that you made? So much of that is encoded in code today. If you think about a product requirements document versus the code, what percentage of the emergent product is encoded in the code? Almost 90%. A lot of the little details are in there.
Software companies of the future and the products that they make are just going to take a really different shape. I’m so excited to be a part of it. I think it’s really fascinating. There’s something really interesting about AI impacting the software engineering industry almost first and most. We’re disrupting the craft of making what we’re building in real time. It’s a fascinating time.
Young founders vs. experienced founders
Jack Altman
There’s a prevailing idea in tech that AI is moving so fast that young founders have this massive advantage. I mean this with no offense, you’re not old, but you’re also not young.
Bret Taylor
You’re telling me I’m old. I get it.
Jack Altman
No, but you’re not the youngest founder and you have one of the most successful AI startups. It does seem like you’ve brought a lot of your previous experiences to what you’re doing, but I can tell from talking to you that you’re also just rethinking everything.
I’m curious about your own experience, for yourself and for other founders. Do you think by and large young founders have the advantage? What does it take for more experienced founders to have the advantage?
Bret Taylor
I’m always a big believer… I don’t know if it’s a real quote, but some venture capitalist said, “Why was this founder able to conquer this market where so many others had failed?” And they said, “He was too naive to know it couldn’t be done.”
There’s a certain element of that I love because you end up with this kind of naivete that is actually a form of principled first-principles thinking. A lot of young founders have that. You just don’t know why this messy, bad product dominates the market. You think there’s a better, faster, cheaper way to do it. You don’t have any of the hard-won lessons that can end up as oversimplified analogies keeping you from actually taking that leap. Tony made DoorDash and didn’t care about Webvan’s failure or Kozmo or whatever it was called. I can’t remember all the dot-com bubble companies.
But I do think, especially in enterprise software, the experience that some of our team members bring—including the old men, me and Clay—really does matter. Part of the reason we’re able to serve so much of the Fortune 100 is we can go into a bank or a healthcare payer or healthcare provider, or a revenue cycle management firm, or a big telecommunications company, and understand their business. We’re working with one large medical device company consolidating 40 of their call centers into one, and we can have a discussion about the change management of doing that. That’s not really a tech problem, but it does require understanding business.
We always joke at Sierra that there’s a Venn diagram. There’s a circle of people who understand the next generation of AI and a circle of people who understand business, and we’re the company right in the middle of that, and maybe the only one. That matters. You know that infamous MIT study saying all these AI projects fail? None of ours do. That’s our value proposition. We can actually help you go live. I think the experience has benefited us.
Jack Altman
I’m curious if you can point to what has created the lead you have so far. Obviously I know you’re just getting started, but at the moment you have pulled away in a big way. I’m sure there’s a lot of daily blocking and tackling. But I’m curious if there are any foundational decisions or strategic approaches that over the last couple years you look back at and think, “That was pretty essential to make this happen.”
Bret Taylor
There’s two almost independent areas of investment. They’re not independent, but they’re very different. One is the product and one is our go-to-market and partnership model. They’re both really intentionally built.
On the product side, we’ve tried to balance ease of use and extensibility. When you serve really large companies that have been around for 200 years, you need to work with mainframes, you need to work with a thousand different systems. You’ve done 10 acquisitions. Enterprises are messy. That’s why most enterprise software designed for larger companies tends to be quite extensible. Often that extensibility comes at a cost, which is: is it easy to get up and running?
As a product designer, one of the things I’ve spent a lot of time thinking about is: we’re trying to have our cake and eat it too. Can you go live in two months and still be maximally extensible? I’m really proud of the product that we built. Some of that is born from experience of what extensibility means. We have an opinionated view of what it means and have been able to accommodate some fairly exotic deployment requests and still do it fast. That’s really unique.
The second thing is our go-to-market and partnership model. We knew when we started the company that we wanted to work with the largest companies in the world. Not only, but we wanted to be able to work with the largest companies in the world and we’ve focused on that. As a consequence, we have a really unique partnership model.
There’s a fashionable thing to talk about: forward-deployed engineering in Silicon Valley. We don’t call it that, and it’s a very unique model because it’s not all about technology. Most of our clients build and maintain their agents themselves. It’s pretty easy to do. But we show up and we help you be successful. We’ll just show up. We’re not going to let you fail. I think that is very different. Because we have this outcomes-based pricing model, we don’t get paid unless it works.
Jack Altman
How much of that is technical versus change management?
Bret Taylor
It’s a mix of both. I don’t know if it’s 50-50.
Jack Altman
Do you think of it as two people or one person who does both?
Bret Taylor
We have a mix of roles. We’ve evolved that. We try to hire really technical people in all roles though because part of our secret is we want to be your trusted partner in AI. So you want the person who is working with you every day to be the most knowledgeable AI person.
Jack Altman
It’s like a forward-deployed change management engineer.
Bret Taylor
Yeah, exactly. What’s really neat about it, if you’re a really talented technical person who wants to transform an industry, you can do it at Sierra. We’re working with most of the healthcare insurance companies. If you want to change healthcare costs, what a cool vantage point to do it from. We’ve been able to attract some really remarkable people.
Beyond support: The full customer lifecycle
Jack Altman
You said that it’s not just support agents now. What else are you finding shoots in?
Bret Taylor
I’ll give you one of my favorite relationships: Rocket. Based in Detroit, remarkable story. Their founder’s done more for Detroit than I think any one person’s done for any city. Remarkable company.
They own Redfin, which is a home search site, Rocket Mortgage, which is the number one consumer mortgage originator in the country. They recently bought a mortgage servicing firm as well.
You can go to redfin.com and use an AI agent to search for a house. You can go to rocket.com and finance that house with an AI agent. And then with the acquisition of the mortgage servicing firm, when you’re servicing your mortgage, you’ll talk on the phone with an AI agent as well. Everything from finding a house to originating the mortgage to servicing that mortgage, I think that’s pretty cool.
They have an amazing CTO named Shawn Malhotra, pretty visionary. I love their CEO Varun too. It’s everything from finding a house all the way through servicing. It’s what we believe a lot of businesses will do, look at their entire customer lifecycle from purchase consideration—which is a fancy way of saying browsing. I think homes are probably one of the more considered purchases you could do—through executing the purchase, through having issues with it, all the way through retention.
For a lot of our telecommunications customers, their AI agent is actually doing negotiations. You’ve probably negotiated your cable bill at some point. Our agents are doing billions of dollars of negotiations for everything from satellite radio subscriptions to cable television subscriptions. It’s pretty cool. Over a billion dollars of mortgage folders a month.
Jack Altman
Basically all transactional communications eventually.
Bret Taylor
The way I think about it is: a website is a technology, but your .com, the one with your brand at the top, is your website. We’re doing that for agents. Agents will do a lot of things. The one with your brand at the top that your customers go to, whether it’s buying or servicing, we’d like to help you make that.
As agents go, they often interact with other agents. If you think about a home and auto insurance company, you may have a claim adjudication agent that’s quite complicated. Our agent that’s having the phone conversation when you’re in a fender bender will interact with that. But it is almost the intersection of all of that technology because it’s your front door.
Our whole hypothesis is that every company needed a website in 1997. Every company needs an agent in 2027. We want to be that company.
Jack Altman
What’s the nuance about agent builders though? I know you have a view that just being a generic agent builder is not the right thing.
Bret Taylor
I’ve been surprised by how many large incumbent enterprise software companies’ first foray into AI was an agent-building tool. It just feels like an inevitable commodity in my mind. Maybe making a website was hard in 1995, but today there’s a million ways to make a website. Most of them are open source. You have cool companies like Vercel, which I love, but it’s not like there’s a huge market for this stuff.
In practice, I think the same will happen with agent building. OpenAI will have a great tool. Probably all the foundation model companies will. There’ll be open source packages like LangChain and LangGraph. The idea that you have the right to win there… I don’t know if anyone has the right to win there because it’s just a horizontal technology and I believe in open source. It’s going to become a commodity.
My belief is that value is really going to be in agents that do things, hiring those agents and purchasing those agents for what they do. I believe in companies like Sierra, I believe in companies like Harvey. I really admire what they do. They have an agent that will do an antitrust review. I think there’ll be a finance agent that audits your financials. There’ll be one that helps you onboard a supply chain vendor.
Just think about onboarding a new vendor. There’s a procurement process, a legal process, a contract review process. Whether or not it’s completely autonomous or human-in-the-loop, all of that could be augmented by AI. That’s a product. Agent building’s not a product. Agent building is a technology.
Codex and the future of software engineering
Jack Altman
Speaking of the platforms, aside from being the founder of Sierra, you’re also on the board of OpenAI. You’re the chairman there. I wanted to ask you specifically about Codex. Over the last couple weeks, it’s been unbelievable. It’s like a curtain just came down. Did you expect this? Did you think that what has happened here was going to happen? When did you start to have an inkling that code was going to go vertical like this?
Bret Taylor
I’ll say yes, I expected it just because being on the board of OpenAI we talk a lot about it. All the labs—Anthropic and OpenAI in particular—talk a lot about using coding agents to help build AI. Certainly building an AI researcher is an important part of building an AGI lab.
The weird part about it, for me as someone who is a software engineer, is I didn’t feel it until I used it. You can talk about it all the time, and then the first time you one-shot something and it turns out really good—not slop, but really good—it’s an emotional experience. For me it was. It was like, “Holy shit. This is real.” As you said, it’s really over the past three months that it has felt really materially different to me. I’ve been thinking about it a lot.
I’ve been thinking about the past 20 years of software engineering. I remember the first time I worked on an engineering team that had real CI/CD, where you’d check in code and it would just automatically end up in production. I remember how it felt. If you’ve ever worked on an engineering team that did that versus one that did manual releases, it’s completely different.
Because to have something that can safely go from commit to production, there are so many things that have to happen to make that work. You end up relying a lot on testing—unit testing, integration testing, and canary testing—because the last thing you want is someone clicking a button and taking down the service. It’s almost impossible for a team that is doing manual releases to convert into true continuous delivery because there are so many implied processes that are incompatible with that. It’s easy to start that way and very hard to convert.
Clearly in three years, we could talk about what the best practices are to set up a software team that’s optimized for this technology and we’ll know what those best practices are. Right now we’re just figuring them out in real time. My hypothesis is that the companies that figure it out first will move the fastest. The other part of that is the companies that don’t will move much more slowly. Andrej Karpathy had a really interesting post about this too. A lot of folks are deep in here and have been thinking about it and it’s fun to see the industry you love flipped on its head in real time.
Jack Altman
It’s interesting because you have software engineers on one end and then somebody who’s in some part of the country where AI has not yet gotten its tentacles fully extended into. There’s a wide gap in people’s current comprehension of what AI’s going to do. It’s a little bit unknown. There are a lot of blog posts going on right now that are breathlessly saying it’s all over. I’m probably more in the camp of, maybe software is solved? I don’t know if it’s that.
But I’m curious if you have a view on whether Codex and Claude Code, the latest in coding, is going to change the way companies are built? Here’s one easy straw man question there. People, like my brother, have been claiming there’s going to be these 10-person billion-dollar companies. Are we at the precipice of that? Does that make sense? Are there other changes? What’s going to happen now?
Bret Taylor
There probably will be a 10-person billion-dollar company, but I don’t necessarily think it’ll be the norm. The reason for that is competition. If you imagine the mobile phone market in the United States, there are three main competitors: Verizon, AT&T, and T-Mobile. They’re all competing for a fixed pie of mobile subscribers. It’s why it’s extremely competitive. There’s promotions, there’s ads.
Jack Altman
They can’t make more of us.
Bret Taylor
They can’t make more of us. They can build up their network, they can do pricing and packaging. It’s a really complex business to run. All of them have access to AI, every single one. So the idea that you could deploy AI and not have to do things that you’re currently doing because of AI is probably true. But if any one of them figures out a way to use a person to gain market share against the other one, they’re going to do it. As a response, their competitors will do it too.
We spoke about this earlier. It’s the reason why when automated teller machines were introduced to banks, the teller job went away, but there are no fewer bank branches and no fewer people in those bank branches. Because JPMC or someone figured out, “Hey, if we put financial advisors in there and other things, we can actually make more revenue per branch.”
My personal take is that in a competitive market—and that’s the key, you need competition so people can’t just pass the cost savings onto shareholders or dividends—the second-order effect of the efficiencies of AI will be investment to compete, lower prices or customer acquisition or whatever it might be.
Jack Altman
We won’t have fewer engineers per company. They’ll just be way more productive, so you end up with way better software.
Bret Taylor
Or you might have fewer engineers and more of something else. Or you might have more engineers, I’m not sure. But the idea that it’ll be what it is today but just more efficient, I think that’s a lack of imagination, in my opinion. The interesting thing though is that software engineering does feel special. I think people extrapolating too much from software engineering are being a bit simplistic.
Jack Altman
You’re saying the same thing might not happen to every other function.
Bret Taylor
I’ll be really simple about it. Finance and software engineering might be limited by intelligence, meaning they’re largely digital. They’re largely manipulating digital things. You could imagine AI automating that.
Most of the economy isn’t exclusively digital. If you need to ship a t-shirt from Vietnam to here, you could automate some of that stuff, but at the end of the day, that cargo ship still needs to be in the water. Imagine you run a pharmaceutical company. You can think about how to make a therapy, but you probably need a wet lab. That intersects the real world. Maybe you could do robotics. But then you need a clinical trial. A lot of the economy is real.
It definitely will change the way companies are built, but I think when people say everything will be 10 people, billion dollars—
Jack Altman
It’s maybe just the stuff that lives in bits.
Bret Taylor
Yeah, that’s right. Which is a lot of the economy, but not the economy.
Jack Altman
It’s easy to talk about this, but you’re right. If you just move around the physical world and you get off of this podcast and this computer and into the world, there’s trucks moving dirt around and people who need a building that has lights in it. There’s a lot of physical things.
I tend to think the value of that stuff’s all going to go up until maybe robots happen. But in general, I think the value of bits goes down, the value of stuff goes up.
Bret Taylor
I think you’re probably right. Robotics will have a big impact as well. But I think people are thinking about this a bit simplistically. Intelligence is clearly on the cusp of going up exponentially, but it doesn’t mean adoption of that can be absorbed by the economy perfectly exponentially. I just think people are being a little bit simplistic.
Jack Altman
Do you think there are any cognitive things that are immune from intelligence? Dylan Field, when he was on this podcast, gave the example of “Brat Summer“ as something where he was just like, “That would’ve been such an insanely hard call for an AI to make. You needed so much context and taste and opinion.”
Where my head was going is, with coding, whatever’s happening there is happening. But what about brand or storytelling? I’m asking you this both as an operator and as somebody who’s very deep with OpenAI. Do you think that these other parts of intelligence also go the way of AI?
Bret Taylor
I don’t know if taste is necessarily related to intelligence. It might be. I’ve got three kids, including a 16-year-old and a 15-year-old. When they decide what they’re going to wear to school, I don’t think they would consider ChatGPT’s opinion. They care more about what the person in class next to them is wearing.
Similarly, if you go to the most elite, competitive college preparatory school, or the worst school in the world, there’s always going to be the smart kid in class and the dumb kid in class, the strong kid and the fast kid, and all these other things. It’s all relative and it’s all very local and it’s all very human.
The idea that because AI is smart, it takes something away from us as humans, I don’t necessarily subscribe to that. You see these things that go around online where people are lamenting older technology like the bicycle. We’ve been weaker than machines for my entire life. I don’t think it makes me feel weak as a person.
This is the first time we have computers that are going to be more intelligent than us. The emotions I had about Codex writing code that was high quality was an experience, because I might have some of my identity tied up in that task. The next day I woke up and I’m using it as a tool and now I can make better software. I’m like, “This is great.”
Jack Altman
Probably actually a good self-actualization anyway to go through that and be like, “Oh, I’m not my ability to code.”
Bret Taylor
People’s vocations and their identities are often very intertwined. But once you absorb the technology, I don’t think it’s actually your identity. I’m actually quite optimistic that we will be human. We’ll all be status-seeking animals. We’ll all compete for the real estate here in San Francisco. Even though our standard of living will go way up, we will all be jealous of people still. We will all compete. As a consequence, I think humanity will be just fine. That’s my view on it. It’s hard to imagine, but it doesn’t mean it’s going to be catastrophically bad. I think it’ll be largely good for humanity.
Jack Altman
Everybody’s already completely addicted to their phones and it’s a disaster. Now you have all this AI happening. A friend of mine was saying that he basically thinks it’ll actually become a status signal to become increasingly offline.
That might be an interesting call. I do think people will hit a tipping point with a lot of this stuff where all of it will happen. Intelligence will get so good and then people will just be like, “Enough of all of this.” Hopefully there’s a big screen time reduction.
You saw parents revolting about social media for their kids. A bunch of schools now, all the parents are like, “Nobody takes a phone. Everybody agrees.” So that’ll be an interesting thing. Is there an essential humanity that gets sharpened?
Bret Taylor
I hope so. I love the iPhone. It’s one of the greatest inventions of this century. I hope we’re not staring at a glowing rectangle in 10 years.
Jack Altman
It can’t be the right way to do it.
Bret Taylor
Now that AI can talk to you and you have human-computer interfaces, this is my hope. Hopefully humanity can become more self-actualized as a consequence of this. That is the purpose of technology.
Just like the Industrial Revolution had Luddites, and globalization led to job loss in the Rust Belt of the United States but certain goods got less expensive in other parts, there’s not going to be no issues. It would be callous and insincere to imply otherwise. But I think it will largely just really accelerate humanity in a really positive way. For me—and for anyone thinking about how this impacts them—have a more flexible view of your own identity. How you do your job every day doesn’t define you.
I love this metaphor, because it was so obvious before and after. Imagine being an accountant before Microsoft Excel and after Microsoft Excel. So much of the act of being an accountant was adding up numbers. Now it’s building a model. What you did, the value you provided, didn’t change, but the actual act of doing it is completely different. The skillset is completely different. A lot of us are just going to go through that in a very compressed period of time. It’s okay. It’s just a little anxiety-ridden.
OpenAI and advertising
Jack Altman
My last question about AI. There was a shot from Anthropic at OpenAI around the Super Bowl commercial about the ads. There were good ads, they were funny, but I think it sparked a debate around the whole topic of the role of these foundation labs, how they should bring AI to the masses, the appropriate business model, the trade-offs.
You’ve obviously had experience with social networks and a lot of different pricing models. You know OpenAI well. You know how to consume AI. I’m curious how you think about this. What is the right thing when you consider a lot of these dimensions?
Bret Taylor
I’m very optimistic about ads done in a tasteful way. I started my career at Google. I think I arrived the day AdWords came out. It was interesting because when I started there—you’ll laugh at this—everyone in my family when they found out I was working there was like, “How do they even make money?” and laughed. I think I listened to the Acquired podcast and it’s literally the most profitable business ever created. But as a consequence, Google is widely available for free for people who want to use it and has created an economy around it for demand-fulfillment advertising.
There are reasonable criticisms of advertising if it starts to get in the way of the sanctity of what the AI is recommending you, which was the backhanded implication. But I just think it’s not true. I actually think if ads are clearly labeled and not changing the experience, it’s really aligned with the OpenAI mission. Our mission is to ensure artificial general intelligence benefits humanity. Obviously the most important part of that mission is safety. But after you get past the Hippocratic Oath—first, do no harm—the job of a doctor is to cure you.
So after you say, “okay, it’s safe”, how do we widely distribute it? We have an obligation, being mission-driven. I’m the chair of the foundation and on the PBC board. Our mission matters. Being able to offer it for free widely is a huge part of that and we need to be able to afford that.
I just find it inauthentic. This is an incredible opportunity to provide this at scale to society. The idea that it will somehow taint the experience is too strong.
Jack Altman
It’s funny. I grew up in the suburb of St. Louis, so it’s a whole different world than what we’re in now. When I think about people that I grew up with or from other parts of the country, $20 a month is a lot. It’s easy to forget in our ecosystem that not everybody wants to or can spend $20 a month on stuff, but they really want these services. If the whole world had to pay for Google, that’d be a worse world. It’s really good that everybody has access.
Bret Taylor
Absolutely. I just think it’s important we do it well and we will.
Jack Altman
People want good ads. I like good ads. If people bring me the right product, I’m like, “That’s really nice.”
Bret Taylor
This is the other part of it. You want businesses to be able to grow from scratch. There’s such a purpose to it. It just needs to be done in the right way. I find the discussion not particularly authentic.
How to run a board
Jack Altman
The last thing I wanted to ask you about was how you’ve chosen to finance the company. I’m curious about three parts: 1) how you got started and working with Peter Fenton, then 2) what you’ve done since then to date and what’s been important for you, and then 3) as you think about the future, what’s important to you as you think about other partners or capitalizing?
I’m asking because this is a podcast that has a lot of VC in it, so I gotta have a little flourish.
Bret Taylor
We have three members of our board, which represent our three rounds of investment. Peter Fenton from Benchmark, Ravi Gupta who just left Sequoia though he’s still a venture partner there, and Neil Mehta from Greenoaks. Just a fantastic group of people. I chose them all for both the firm and the person.
Notably, with Peter, I’ve worked with him at both my previous companies. For our first round of financing, I didn’t talk to anyone else. I introduced him to Clay, my co-founder, who hadn’t spent time with him. We talked once, he sent me a term sheet, I signed it, no edits. It was very much a trust relationship.
One of the things I’ve really appreciated about Silicon Valley… There are some downsides to how insular the community is, but one of the great parts is the relationships you can forge over years. For me, it meant Peter and I could start on third base just because we’ve worked together a lot before. You don’t end up with a lot of the… There’s no funny business in the fundraising process. No funny business in the boardroom. It’s just, “Let’s get to work.” It was fun to get the band back together there.
But the fun part for me is I had never worked with Ravi nor Neil before, and Clay and I just… It’s just a great board, people we seek out advice from as opposed to people we report to every quarter. It’s amazing.
Jack Altman
We won’t go back through the story, but when OpenAI had its “Oh my god” moment, Sam was like, “Bret, you’re the board member”, and then you’ve also got a board. You’re in both roles at once. How do you make the most out of a board? Obviously you’ve got these particular relationships, but what do you expect that relationship to look like?
Bret Taylor
First, I really like written documents for boards over presentations, both as a board member and as a founder of a company. You end up letting people synthesize information ahead of the board meeting, so you end up with more substantive discussions in the boardroom.
I’ve done this for the last two companies I’ve started. It’s just been great to send out a board document. Sometimes people will comment ahead of the meeting, but the main thing is it has been read, ahead of time. Then you end up with a meeting about the actual meat and potatoes of the topics. You’re not staring at a bunch of sales numbers for the first time.
Jack Altman
You’re not running through slides.
Bret Taylor
You’re not running through slides. I find it to be incredibly… I think most companies should be run this way. The other thing that is really interesting: don’t write it with AI. It’s so funny to have to say that now, but I find that—
Jack Altman
The process of the writing.
Bret Taylor
The process of writing is a process of clarifying your thoughts. For Clay and me, this is a process by which we synthesize what’s been happening. You know it and talk about it, but to actually write it and write it eloquently and concisely is incredibly important because it’s essentially a way of… What’s that famous line? “If I had more time, I would’ve written a shorter letter.” Spend the time, because that’s actually how you can show respect to your stakeholders, that you’re thinking about the strategic issues going on in your business.
The last thing I’d say is that board members aren’t single-issue voters, but everyone has their strengths. At OpenAI we’ve recruited a pretty diverse set of skills. Zico Kolter is a professor at CMU and specializes in, among other things, jailbreaking. He’s one of the experts on some of the more subtle safety aspects. Nicole Seligman was a great attorney and she’s an expert in a lot of legal issues. What’s really nice is that when you grow out a board beyond your initial investors, find people that your management team will want to go to for advice. Obviously the audit committee chair and your CFO have a really unique relationship.
Who’s your head of sales going to go talk to? Do you have someone who’s been there, done that? Because you want them to have that kind of relationship. I always think of it as, who are the advisors you want to surround your management team with? A functional board really has those relationships. Then when you’re in a board discussion, you have all these board members who have had lots of engagement with the company, but in a really valuable, targeted way.
I like to think of the board as a collection of people. Don’t look at the individuals. The whole should be greater than the sum of its parts.
Jack Altman
That’s awesome. Anything this year you’re particularly excited about that you can share?
Bret Taylor
The real exciting part is going to be adoption in regulated industries. We are moving beyond the early adopters to everyone. If we talk a year from now—
Jack Altman
You’re going to be doing the hard stuff.
Bret Taylor
It’s going to be the really hard stuff.
Jack Altman
That’s awesome.
Bret Taylor
If you want a hot take, my intuition is that regulators will start asking for agents. The idea that you have a human set of controls over a regulated process will start to feel like a risk, rather than the risk being AI. I don’t know if that will happen this year, but I think that will happen.
Jack Altman
Alright, I’ll call you in a year and we’ll do take two of this and see.
Bret Taylor
That sounds great.
Jack Altman
Thanks so much for doing this, Bret. This was great.
Bret Taylor
Thanks for having me.