What this covers

Patrick interviews Bret Taylor, co-founder of Sierra and chair of OpenAI, in a conversation that starts with his career and widens into the organization and economics of AI. Early on he takes on the Mythical Man-Month, arguing that adding people often slows software down because small teams hold the whole system in their heads. He also corrects the legend of the Google Maps weekend rewrite: it applied roughly six months of hard-won browser lessons rather than starting from a blank slate. He describes struggling as Facebook CTO until he began asking each morning what the most impactful thing he could do that day was.

The middle of the episode concerns AI agents. Taylor compares free-form conversation to the move from Yahoo's directory to Google's search box, since customers reveal needs a menu never offered. He describes software shifting from fixed rules to goals and guardrails, and he works through the agency-control tradeoff: more capable models with tool access can be given more autonomy, but they demand more sophisticated safeguards. He also covers the three inputs behind model progress (algorithms, data, compute), AI's real marginal costs, and an internal bet with his cofounder Clay on when personal agents will outnumber people talking to company agents.

The wider turns are more contested. Taylor explains the Jobs to Be Done milkshake story as advice for incumbents, and he describes OpenAI's nonprofit structure and its for-profit subsidiary. He argues AI may be equalizing by removing creative gatekeepers, and he expects adoption fast enough to reshape work within years. He also warns that agents will have second- and third-order effects no one can predict.

Sharpest takeaway

Taylor argues AI agents will become branded, customer-facing digital assets, and that the binding constraints are the non-determinism and cost of LLM-based software plus organizational willingness to adapt, so companies should adopt AI now and redefine the job their customers hire them to do.

  • Agents are branded, customer-facing software like websites were in 1995, and their value depends on free-form conversational expressiveness
  • LLM systems are slower, more expensive, and non-deterministic, which turns software building from a methodological problem into a statistical one
  • Companies should adopt AI internally and externally now, and first ask what job their customers hire them to do

The claims · ranked75 claims · weighted by value

This asset isn't compiled yet

You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.

0.80

The Mythical Man-Month is the myth, common among business-school-trained managers, that adding people speeds up a software project; often the inverse is true because small high-functioning teams hold complex systems in their heads, and adding people adds process and bureaucracy that disempower the most advanced technologists and slow things down.

causalhigh valueestablishednovelty 2/4durability 4/4· Bret

“to make a software development project go faster, you should add more people to it, and often the inverse is true”

0.80

Small empowered teams understand the customer problem precisely and hold an accurate model of the whole system, which enables them to take shortcuts safely and change direction quickly, whereas large teams narrow each person's view and require heavy upfront planning that reduces agility when new information arrives.

causalhigh valueestablishednovelty 2/4durability 4/4· Bret

“most people understand why they're doing what they're doing, not just what they're doing”

0.80

When a new technology emerges, such as the Internet, the web browser, or the smartphone, it correlates with startups that become very large companies because natively built companies adopt different business models and move with more agility than incumbents, as Google did with ad-based consumer software and Amazon did to booksellers.

causalhigh valueestablishednovelty 2/4durability 4/4· Bret

“when there's -- a new technology emerges, like the Internet and the web browser or the smartphone, it tends to correlate with some start-ups that become very large companies.”

0.80

Companies should identify the job customers hire them to do rather than optimizing product features, as the milkshake example shows: morning buyers wanted portable sustenance for a commute while afternoon buyers wanted to make a child happy, so changing the recipe was the wrong fix.

normativehigh valueestablishednovelty 2/4durability 4/4· Bret

“what is the job your customers are hiring them to do, which is to provide sustenance on a long commute and to provide a way to make your child happy in the afternoon.”

0.80

Consumer products are well crafted because sales correlate with craftsmanship, while enterprise software is often clunky because the buyer is distinct from the user, though product-led growth companies like Slack break this pattern.

causalhigh valueestablishednovelty 2/4durability 4/4· Bret

“typically, the buyer of enterprise software is distinct from the user of enterprise software”

0.80

The societal impact of technologies is hard to predict: when Steve Jobs introduced the iPhone in 2007, few anticipated that social networks, smartphones, and push notifications combining would produce the social issues parents and society grapple with now, so AI agents will likely have counterintuitive second- and third-order effects.

forecasthigh valueestablishednovelty 2/4durability 4/4· Bret

“you have both the first-order impact of these technologies and the second and third order, which are very hard to predict.”

0.75

As AI models gain tool access and internet access, the surface area for things to go wrong expands, and the sophistication of required safety measures grows, creating a tension in which better models permit more agency but also demand more complex guardrails.

causalhigh valueestablishednovelty 2/4durability 3/4· Bret Taylor

“once AI can access tools or access the Internet, the surface area of opportunities that things go wrong increase as well”

0.74

Traditional software platforms have near-zero marginal cost per page view and mostly pay for headcount, whereas AI platforms face real margin pressure from training and inference on high-parameter models.

factualhigh valueestablishednovelty 3/4durability 3/4· Bret Taylor

“Your cost as a software platform is almost exclusively in your head count. With AI, you can see the margin pressure”

0.68

Because hallucinations and jailbreaks are more frequent in current models, responsible deployment gives agents smaller bite-sized tasks, like writing a function rather than a whole program; as reasoning and tool use improve, developers can turn up agency responsibly, but each increase in autonomy widens the attack surface and requires more advanced guardrails.

normativehigh valuecontestednovelty 2/4durability 3/4· Bret

“you can turn off the knob on agency responsibly”

0.68

AI tools remove gatekeeping in creative and professional fields, letting people with good taste and judgment, regardless of budget or permission from financiers, produce high-quality films, art, or products, and thereby possibly making the technology broadly equalizing.

normativehigh valuecontestednovelty 2/4durability 3/4· Bret

“it enables people with good taste and good judgment to accomplish a lot more than they could individually.”

0.68

Progress in large language models comes from three continuously improving inputs: algorithms such as the Transformer from the paper Attention Is All You Need, data including multimodal and synthetic data, and infrastructure in compute for pre-training and post-training, so if one input plateaus progress can continue on the others.

causalhigh valuecontestednovelty 2/4durability 3/4· Bret

“And the combination of all that is even if one of them hits a plateau, you can see progress on the other.”

0.68

OpenAI is a Delaware nonprofit whose board serves a mission rather than shareholders, with the mission of ensuring AGI benefits all of humanity; because building AGI requires vast capital for computers and training, a for-profit subsidiary was created to raise funds, since a nonprofit steak dinner cannot raise billions.

factualhigh valueestablishednovelty 2/4durability 3/4· Bret

“rather than being a fiduciary to shareholders, folks like me, we're fiduciaries to the mission.”

0.68

Balance is achieved through intentionality: people let inboxes and others' requests define their time, but time allocation is a choice, so a person can decline dinner meetings, quit jobs, or say no, and treating choices as unavailable is a way of letting the world happen to you.

normativehigh valuecontestednovelty 2/4durability 3/4· Bret

“everything is a choice, so I tried to not let the world happen to me.”

0.60

When Bret struggled as Facebook CTO by trying to be heroic rather than growing his team, he shifted from conforming the role to his technologist identity to asking each morning what the most impactful thing he could do that day was, which made him better at the job and more fulfilled, and taught him that impact matters more than identity.

causalhigh valuespeaker onlynovelty 3/4durability 4/4· Bret

“I was trying to conform the new job to me rather than conform myself to the new job”

0.60

Free-form conversational agents are more expressive than menus, so customers reveal what they actually want, including unexpected needs, and companies learn things about their customers they did not anticipate; this is analogous to the shift from Yahoo Directory taxonomies to Google's search box.

causalhigh valuespeaker onlynovelty 3/4durability 4/4· Bret

“we've had multiple customers learn things about their customers that they didn't expect by providing this really free-form experience”

0.60

Companies organizationally and psychologically conflate what they do and the value they provide with how they do it, which constrains their response to technology disruption through their existing delivery model; companies with a crystallized view of the value they provide can apply new technology more easily because they do not treat existing practices as precious.

causalhigh valuespeaker onlynovelty 3/4durability 4/4· Bret

“so many companies end up organizationally and psychologically conflating what they do and the value they provide with how they do it”

0.57

The famous Google Maps weekend rewrite was not a blank-slate miracle: the team's code base had become a mess of hard-won browser hacks after roughly six months of research-driven work, and Bret rewrote it by applying everything the team had learned, which cut the JavaScript bundle to about 20,000 uncompressed bytes and made it work reliably across browsers.

factualhigh valuespeaker onlynovelty 3/4durability 3/4· Bret

“I think the story is a little bit overblown. I did rewrite it, but it was based on all the lessons I learned.”

0.57

AI allows brands to scale their best human service experiences, operationalizing magical moments and delivering personalization and empathy that were impossible before, because a discovered technique can be deployed instantly rather than retrained across a slow, expensive workforce.

causalhigh valuespeaker onlynovelty 3/4durability 3/4· Bret

“if you discover a technique that works either intuitively or through some measured automation, you can actually deploy it instantly.”

0.57

Software is moving from rule-based systems, where the same input reliably yields the same output, to goals-and-guardrails systems, which requires a very different mental model for building software because LLM components are slower, more expensive, and nondeterministic.

forecasthigh valuespeaker onlynovelty 3/4durability 3/4· Bret

“software systems are moving from rule-based to goals- and guardrails-based”

0.57

Unlike the PC, broadband, and electricity, which needed years of infrastructure investment, AI is a largely software-defined revolution whose adoption can be instantaneous once the technology works, so the change to the nature of work could happen within a handful of years, creating an urgent social problem without simple solutions.

forecasthigh valuespeaker onlynovelty 3/4durability 3/4· Bret

“This is largely a software-defined revolution, notwithstanding the investment in data centers”

0.56

Conversation is the most accessible computing interface because everyone already knows how to talk, which explains why iPads let older users skip the PC era while mice and keyboards intimidated them.

causalhigh valuespeaker onlynovelty 2/4durability 4/4· Bret Taylor

“no one needs an instruction manual on how to have a conversation”

0.56

Companies in the age of AI should map the job customers hire them to do, asking what will change and what will stay the same, and then map a path from current to future delivery, with beginner's mind about roles that may need reskilling.

normativehigh valuespeaker onlynovelty 2/4durability 4/4· Bret Taylor

“how do we map a path from point A to point B?”

0.54

Over 30 to 50 years society will conform the economy around the technologies it produces, as it shifted from agrarian to industrial to services, so many of the jobs on the other side of AI will be unrecognizable today.

forecastestablishednovelty 2/4durability 4/4· Bret

“society will conform the economy around the technologies that we produce”

0.53

Personal agents will interact with company agents, and Bret and his cofounder Clay have an internal bet about when conversational traffic to company agents will be dominated by personal agents rather than people.

forecasthigh valuespeaker onlynovelty 3/4durability 2/4· Bret

“when the conversational traffic to our company agents will be more personal agents than people”

0.52

Three ingredients make a robust customer agent: factual knowledge about the company, provided via retrieval augmented generation so it doesn't hallucinate; procedural knowledge, meaning the processes that the best employees follow; and access to underlying systems so the agent can take action rather than only answer questions.

normativehigh valuespeaker onlynovelty 2/4durability 3/4· Bret

“So three ingredients, factual knowledge, procedural knowledge and systems integrations, I think, are the three key ingredients.”

0.52

Generative AI is very easy to demo and very hard to make into an industrial-grade system, and whether 90 percent accuracy is acceptable depends on the process: fine for homework help, but unacceptable for revenue-impacting or compliance-sensitive operations.

normativehigh valuespeaker onlynovelty 2/4durability 3/4· Bret

“generative AI broadly is a technology with which it's very easy to make a demo and very hard to make an industrial-grade system”

0.52

Just as a company needed a website in 1995 to exist online, companies will engage customers through their own branded AI agent, which becomes a digital asset and the digital manifestation of the brand, deserving as much care as a website or mobile app.

forecasthigh valuespeaker onlynovelty 2/4durability 3/4· Bret

“Just like in 1995, the way you existed online was to have a website, we think in 2025, the way you will engage with your customers will be your AI agent”

0.52

AI progress will be fairly iterative, a gradient rather than discrete steps, with occasional large jumps such as instruction tuning in ChatGPT and GPT-4, and between jumps many incremental model improvements will occur.

forecasthigh valuespeaker onlynovelty 2/4durability 3/4· Bret

“I think of it more as a gradient than I do steps.”

0.52

Companies face a design choice between control and agency: maximal guardrails make a conversational experience robotic, while too much agency risks hallucination or failing to protect the brand, so the right balance is both a technology problem and a philosophical branding question each company must answer.

normativehigh valuespeaker onlynovelty 2/4durability 3/4· Bret

“there's a deeper question here, too, that's actually a philosophical branding and design question as well”

0.52

Traditional direct customer conversation requires contact centers, overstaffing for peaks, and heavy retraining for each product change, so it is economically prohibitive for most brands; AI reduces the marginal cost of a conversation to near zero, making personal conversation with millions of customers feasible and turning customer service from a cost center into a channel for impactful interaction.

causalhigh valuespeaker onlynovelty 2/4durability 3/4· Bret

“With AI, we have this opportunity to bring down the cost of a conversation down to something that's fairly marginal.”

0.52

Conversation is the most natural interface to computers, a progression from punch cards to keyboards to touch screens to speech, and GPT-4o's voice and multimodal experience was the first time for many people where it did not feel like talking to a robot, making conversation the interface that needs no instruction manual.

forecasthigh valuespeaker onlynovelty 2/4durability 3/4· Bret

“It didn't have that uncanny valley feeling of I'm talking to a robot.”

0.52

The most important thing incumbent companies can do is adopt AI both internally, by encouraging use of sanctioned ChatGPT and setting automation milestones in back-office processes, and externally, by building customer-facing agents, so they are not on their heels when startups exploit the technology; history includes Blockbuster and Netflix as examples.

normativehigh valuespeaker onlynovelty 2/4durability 3/4· Bret

“the most important thing companies need to do is adopt AI internally and externally.”

0.52

The most complicated second-order effect of AI automation is that a new competitor may emerge with drastically different profit margins and cost basis, allowing it to compete unequally on pricing and packaging against incumbents.

causalhigh valuespeaker onlynovelty 2/4durability 3/4· Bret

“can a new competitor emerge with drastically different profit margins than yours?”

0.52

Two practical starting moves apply across industries: build a customer-facing AI agent so customers meet an AI version of the company, analogous to having a website in 1996, and open the door to AI use inside the company, since frontline employees will generate the best applications, as spreadsheets did in finance.

normativehigh valuespeaker onlynovelty 2/4durability 3/4· Bret

“number one, build a customer-facing AI agent, so that your customers are experiencing an AI version of your company.”

0.52

Product-led B2B companies can become arrogant and patronize customers by assuming their product vision is correct, whereas Salesforce, under Marc Benioff, was deeply listening to its customers, which is more effective when the company's world view is wrong.

normativehigh valuespeaker onlynovelty 2/4durability 3/4· Bret

“that can lead to either patronizing customers. They don't really want what they say they want.”

0.52

The AI boom will rhyme with the dot-com bubble, producing both hype and snake oil alongside generational firms, because conceptually obvious ideas succeed through execution, as Amazon and Google outlasted Buy.com, Webvan, Pets.com, and AltaVista.

forecasthigh valuespeaker onlynovelty 2/4durability 3/4· Bret

“Amazon got to write the history books, not Buy.com. Google got to write the history books, not AltaVista or Inktomi because of product technology, go-to-market execution.”

0.52

Sierra's value of intensity responds to an AI boom that will rhyme with the dot-com bubble, producing both hype and generational firms, where the winners are determined by execution rather than conceptual novelty, so the company cannot afford patience.

normativehigh valuespeaker onlynovelty 2/4durability 3/4· Bret Taylor

“Amazon got to write the history books, not Buy.com. Google got to write the history books, not AltaVista or Inktomi”

0.52

Agents fall into three categories: customer-facing company agents, internal persona-based agents that perform job functions such as paralegal, software engineering, or operations analyst work, and personal agents that act on an individual's behalf; technically they are similar but differ enough in job and integrations that different companies will likely build each.

factualhigh valuespeaker onlynovelty 2/4durability 3/4· Bret

“the third one -- category is probably personal agents”

0.52

The job of a software engineer is shifting from emitting code by keyboard to operating a code-generating machine, and engineers who learn to do this well will operate at very large scale; this is analogous to Excel displacing manual accounting work without eliminating accountants, and the true job is producing software, not typing.

forecasthigh valuespeaker onlynovelty 2/4durability 3/4· Bret

“I don't think the job of a software engineer is to type into a terminal. The job of a software engineer is to produce software.”

0.49

Since 2007 many startups and big tech companies have tried to build devices that augment or replace the smartphone, but the smartphone is so powerful and universal that new devices usually only do a slice of its functions slightly better, so the smartphone with a voice agent will likely remain the dominant interface because it exists, not because it is best.

forecasthigh valuespeaker onlynovelty 2/4durability 2/4· Bret

“If I had to guess right now for the foreseeable future, that will be the dominant interface. Not because it's the best, but because it exists.”

0.47

In markets, conceptually obvious ideas like selling books online or AI customer agents are not differentiators; winners are determined by product, technology, and go-to-market execution.

normativeestablishednovelty 2/4durability 4/4· Bret Taylor

“They're conceptually obvious, but it's all about execution”

0.43

Because productivity gains from AI tools are amplified for people who use them effectively, the digital divide grows more important, so ensuring universal access to these technologies becomes even more important than before.

normativeestablishednovelty 1/4durability 3/4· Bret

“the proverbial digital divide is an increasing concern because if the productivity gains from effectively using these tools is amplified”

0.40

In 2004-2005 web pages were mostly static, so making a Windows-style interactive map such as dragging the map work in a browser was mind-blowingly hard, and the team repeatedly crashed Firefox and Internet Explorer and relied on Firefox engineers at Google to debug browser behavior.

factualestablishednovelty 1/4durability 4/4· Bret

“you could actually drag the map, click your mouse and drag left and see the map move, which seems so simple now, but it's the first time you really had anything that interactive inside of a web browser”

0.39

Retrieval augmented generation grounds a model by presenting it with the company's knowledge base at answer time rather than relying on its training, which avoids the cost of fine-tuning on proprietary data and prevents hallucination by giving the model only the knowledge it is allowed to use.

causalestablishednovelty 1/4durability 3/4· Bret

“you present the AI with the knowledge that it's allowed to know”

0.38

Sheryl Sandberg told Bret that his project was going poorly because he was trying to do the job himself as a heroic individual contributor rather than growing the team underneath him, a hard truth that he credits with changing his leadership approach.

factualspeaker onlynovelty 2/4durability 4/4· Bret

“the reason why this project was going poorly was basically because I was trying to be heroic and do the job myself rather than grow this team underneath me.”

0.35

The book Unreasonable Hospitality illustrates delightful one-off service moments at a three-star Michelin restaurant, and its central question is how to operationalize and repeat those moments.

definitionestablishednovelty 1/4durability 3/4· Bret

“his vision of the book is, how do you operationalize that and repeat it?”

0.34

Netflix defined its job as evening entertainment rather than streaming movies, so its competitor was any evening entertainment such as gaming platforms, not only HBO, a framing Bret attributes to Reed Hastings.

factualspeaker onlynovelty 2/4durability 3/4· Bret

“the job of Netflix wasn't to stream a movie. The job was entertainment in the evening”

0.34

Designing conversational customer experiences is a new discipline with no settled design idioms yet; websites took years to develop conventions like hamburger menus, and conversational interfaces will similarly need years of accumulated practice before they are standardized.

factualspeaker onlynovelty 2/4durability 3/4· Bret

“the act of actually designing conversational customer experiences is a relatively new discipline”

0.34

At large firms the real problems and best applications of AI are found by employees on the ground, so leaders should empower them with structures rather than rely primarily on top-down direction.

normativespeaker onlynovelty 2/4durability 3/4· Bret

“I'm not sure top-down is important, but actually, I'm a huge believer that especially at larger firms, it's often the people on the ground”

0.34

The most meaningful algorithmic breakthrough of the past decade was the Transformer architecture from a Google paper, Attention Is All You Need, which created the current wave of AI excitement.

factualestablishednovelty 0/4durability 4/4· Bret Taylor

“Probably the most meaningful breakthrough over the past decade was the Transformers model, which was a paper out of Google, called Attention Is All You Need”

0.33

A large amount of the textual content humanity has produced has already been used for training large language models, so future gains depend increasingly on multimodal data and on simulation and synthetic data generation.

factualcontestednovelty 2/4durability 2/4· Bret

“I think a lot of textual content -- a large amount of the textual content humanity has produced has already been used in training.”

0.32

The scaling laws, which hold that linear increases in compute and data produce roughly linear quality improvements, will eventually stop, but no one knows when.

forecastcontestednovelty 1/4durability 3/4· Bret Taylor

“no one really knows when -- the scaling laws ... will stop”

0.31

Sierra's approach to robustness includes having one AI model supervise another and layering multiple models on top of each other to produce statistically more robust results, along with an agent development life cycle process.

factualspeaker onlynovelty 2/4durability 2/4· Bret

“having one AI model supervise another AI model to layer different models on top of each other to produce statistically more robust results”

0.31

Smart speakers like Alexa and Siri have not transformed phone use because of platform limitations that predate modern LLMs; as countertop microphones, CarPlay, and car systems gain full conversational capability with personal agents, they could become genuinely effective computers rather than devices for weather and music.

forecastspeaker onlynovelty 2/4durability 2/4· Bret

“the surge we saw and excitement around smart speakers and all that might come back when those things are actually more effective computers”

0.31

Engineering is the act of applying science to reality, and much of its value comes from the moment a system meets reality, which requires accountable, high-frequency decision-making.

normativespeaker onlynovelty 1/4durability 4/4· Bret Taylor

“engineering is the act of taking science and applying it to reality”

0.30

An agent is a system where software and AI can reason, make decisions, and take action without human intervention; the idea is old but its practical effectiveness has improved so much over the past couple of years that we are now in an era of agents.

definitionestablishednovelty 1/4durability 2/4· Bret

“a system that can reason and take action autonomously is the way I think about it”

0.30

Intensity inside a company is built through urgency and caring about every detail, and values become real only when employees feel them, as Google's don't-be-evil and Facebook's move fast pace did, so the leader's intensity is reflected back by colleagues.

normativespeaker onlynovelty 1/4durability 3/4· Bret

“is it a poster on the cafeteria wall or do you feel it?”

0.30

Attention to detail in every interaction, from slides to phone calls to products, signals to customers that the unseen details are also right, which builds trust in a new and risky category like AI agents and makes product designers and engineers proud of their work.

normativespeaker onlynovelty 1/4durability 3/4· Bret

“If you feel that the details are right, you have trust that we're going to get all the details you don't see right as well.”

0.27

If rethinking a system requires going through a committee, it will never get done, whereas a small team with a clear objective and empowered people can make large changes quickly.

normativespeaker onlynovelty 1/4durability 3/4· Bret Taylor

“if you have to go through a committee to talk about rethinking something, you'll never get it done”

0.27

The scaling laws predicting roughly linear quality gains from compute and data will eventually stop, but because the smartest researchers are concentrating on algorithms, data, and infrastructure simultaneously, progress is remarkable for now.

forecastcontestednovelty 1/4durability 2/4· Bret

“no one really knows when -- the scaling laws ... will stop”

0.26

Bret hopes technology recedes into the background despite growing power because people spend too much time staring at screens, which he suspects harms personal well-being and interpersonal relationships.

normativespeaker onlynovelty 1/4durability 2/4· Bret

“I hope technology melts away from what we do.”

0.26

Customers use free-form agents in ways companies did not plan, such as a shoe retailer's agent being asked what sandals to wear to a Hawaii wedding, which pushed retailers toward product recommendation and consultative selling.

factualspeaker onlynovelty 2/4durability 2/4· Bret Taylor

“you might think I'm going to launch this agent to handle customer service, which is a great initial application, but you might find your customers are pulling you in a different direction”

0.26

The main rewrite goal for Google Maps was to reduce the JavaScript bundle a browser must download before showing a map, getting it to about 20,000 uncompressed, while making it work reliably across browsers.

factualspeaker onlynovelty 2/4durability 2/4· Bret Taylor

“We got it down to like 20,000 uncompressed”

0.25

Building a personal assistant agent requires integrating with an effectively infinite number of systems because people use different calendars, email, and tools, whereas a coding agent is narrow but deep and is judged by benchmarks of the software it produces and its robustness.

factualspeaker onlynovelty 1/4durability 3/4· Bret

“If you're building a software to be a personal assistant agent, the breadth of systems you have to integrate with is infinite”

0.25

Sierra is an applied company that engineers on top of foundation and frontier models to produce reliable customer agents, in contrast with fundamental research labs such as OpenAI.

definitionspeaker onlynovelty 1/4durability 3/4· Bret

“we're not that we're applied. We're really thinking about how do we engineer on top of these foundation and frontier models”

0.25

Early iPhone apps were flashlights and simple utilities, and it took several App Store iterations before apps like DoorDash, Uber, and WhatsApp emerged, so conversational AI will likely need time before native agentic platforms and tools mature.

forecastspeaker onlynovelty 1/4durability 3/4· Bret

“first iPhone apps were like flashlights”

0.24

Patrick notes Taylor has written about Coase's Theorem, the idea that transaction costs shape firm boundaries; Taylor's reply does not directly develop it, so the substantive claim is limited.

factualestablishednovelty 0/4durability 2/4· Bret Taylor

“I've seen you write about Coase's Theorem before”

0.24

Patrick raised Coase's Theorem and the age of the joint stock company as a frame for how technology changes firms, but Taylor did not develop the theorem itself.

definitionestablishednovelty 0/4durability 2/4· Patrick

“I've seen you write about Coase's Theorem before”

0.24

A key goal is to let people who are not AI experts express the behavior they want in their agent, just as one no longer needs a computer-science PhD to build a website, though Bret concedes this is not yet achieved.

normativespeaker onlynovelty 1/4durability 2/4· Bret

“You shouldn't have to be an AI expert to make an agent just like you shouldn't have to have a PhD in computer science to make a website.”

0.24

Personal agents integrated into car systems like CarPlay could let people complete tasks such as sending emails by voice during commutes, turning previously queued app-based time into hands-free work.

forecastspeaker onlynovelty 1/4durability 2/4· Bret Taylor

“rather than queuing things up on your commute, you're actually getting work done on your commute in a completely hands-free way”

0.24

The kindest thing anyone has done for Taylor was his wife saying yes when he proposed to marry her.

factualspeaker onlynovelty 0/4durability 4/4· Bret

“My wife saying, Yes, that she'd marry me.”

0.21

Salesforce under Marc Benioff was the most customer-centric company Taylor had experienced, with genuine accountability for customer success, and Taylor considers Benioff the greatest enterprise software leader of all time.

normativespeaker onlynovelty 1/4durability 2/4· Bret Taylor

“I saw in Salesforce the most customer-centric company I'd ever experienced”

0.21

Research labs such as Anthropic, which is a B Corp, illustrate a trend toward mission-driven structures that hold dual accountability to a mission and to shareholders, enabling both mission focus and the large capital needed to build AGI.

factualspeaker onlynovelty 1/4durability 2/4· Bret

“my understanding is that Anthropic is a B Corp, which is a benefit corporation, which means it has dual accountability to a mission and shareholders.”

0.18

A new customer typically gets a Sierra customer-facing agent live in one to three months, because Sierra uses a high-touch model that holds customers' hands so they need no AI expertise, analogous to IKEA rather than building a couch from Home Depot parts.

factualspeaker onlynovelty 1/4durability 1/4· Bret

“Between one and three months.”

0.13

Sierra's platform runs customer-facing agents for Sonos setup help, SiriusXM's Harmony agent, and WeightWatchers' 24/7 live coaching tab, showing agents deployed across hardware support, subscription service, and coaching.

factualspeaker onlynovelty 0/4durability 1/4· Bret Taylor

“if you're a WeightWatchers member, if you click on the 24/7 live coaching tab in their app, that's an AI agent they've built on our platform”