He Built The Revenue Engines for Google, Facebook & Square

tldr

Gokul Rajaram — legendary operator (Google AdSense, Facebook Ads, Square, DoorDash) and prolific investor — joins Patrick O'Shaughnessy to discuss how AI is fundamentally reshaping product development, company durability, and career strategy. The core thesis: AI tools like Claude Code are collapsing the PM/designer/engineer roles into one, "judgment" is the only future-proof human skill because AI slop is the real danger, and legacy software companies (especially seat-based pricing like Zendesk) face existential risk. He shares direct frameworks for building ads businesses (three pillars), selecting north star metrics with check metrics, and hiring "builders" over managers. Packed with first-hand stories from Larry Page, Mark Zuckerberg, Jack Dorsey, and Tony Xu.

Key Takeaways

Product Development Is Fundamentally Different Now

PM/designer/engineer roles are merging. "PMs are starting to check in code with Claude Code into the actual production repository." When given the choice between an extra designer and extra engineer, teams are choosing the engineer — AI handles design within established design systems. PM-to-engineer ratio going from 1:10 to 1:20.
Product development is now bottoms up. "Product managers the only thing they do now is they articulate what the customer needs are at the highest level and then they are the guardian of the why." The actual product is built by engineers, researchers, PMs, and designers all working on code together.
Software is non-deterministic now. "Today you could do X, Y happens. But if you do a slight variation of X, something completely different happens." PMs must own evals — writing AI to evaluate the results of AI, because humans can't review at scale.
Prototyping interviews are now standard. Companies are adding explicit prototyping interviews to the loop that force PMs to be hands-on. You can't BS your way through a coding exercise.

Judgment: The Only Future-Proof Skill

"In an era when you can do everything, the question is which of these things matter and you should truly do." AI slop is the #1 worry for every product leader Gokul talks to. A thousand AI engineers writing code means mountains of output — but who decides what's valuable?
Three dimensions of judgment: Product side = what to build and evaluating output. Engineering side = reviewing AI-generated code for bugs and vulnerabilities. Design side = does this make sense within the broader design system?
The PM is an editor, not an author. Jack Dorsey called the PM role "product editor." "Any of us can look at a product and say here's 10 things you should build. The best product people edit down things." Rick Rubin said he wasn't a producer, he was a reducer.

Software Durability in the AI Age

Seat Pricing companies are most endangered. "Zendesk prices seats and each seat corresponds to a customer service agent. I can have an AI agent sit right next to Zendesk and slowly siphon off. Instead of paying for 50 seats, you pay for 20 and have 30 AI agents." You need to shift to outcome-based pricing — but going from $20/seat to $0.50/ticket is a brutal transformation. Many may need to go private to make this shift.
Data half-life determines durability. Slack's data has a short half-life — it's more vulnerable. NetSuite's ERP data runs the whole business — no one's ripping that out. "The software public markets are not distinguishing between these two types of companies."
System of record > system of action. "Last year everyone was like 'we can do workflows and live on top of the system of record.' I don't think that's an option anymore." Legacy companies are cutting off APIs (Slack blocked Glean), offering their own agents for free, or charging $2/API call. Agent companies have no choice but to build the full platform.
Five sources of durability: (1) Ownership of a scarce asset or license, (2) Control point over money or data, (3) Hardware that's hard to replace, (4) Part of an essential workflow, (5) Network effects. "The halflife of software today is so short that unless you have one of these, you're vulnerable."
Migration is the real moat. "Who's going to migrate it? Even for Square, for a small business — they had gift cards, customer data, loyalty data, payments data. We had to build scripts and that took us months or years." One of Gokul's portfolio companies literally hired engineers in Eastern Europe for 2 years just to build a Salesforce migration tool.

Four Sources of Stickiness

Network Effects. DoorDash is sticky not because of a beautiful app but because it's a network of restaurants, dashers, and consumers. "You can't vibe code your way to those."
Money flowing through you. Financial services + software = sticky. Banks, Mercury, Toast with payments — once money flows through, switching has regulatory and operational friction.
Hardware. Toast gives you hardware for free but charges if you return it. Software alone can be replicated; hardware can't.
Access to a unique asset. Sierra's unique asset is Brett Taylor — chairman of OpenAI who can call any company in any country and they'll take his call. "You can't really outsell Brett."

Building Ads Businesses — Three Pillars

Pillar 1: Own a coveted first-party surface. Google had search intent. Facebook had identity (could match users to customer data). ChatGPT has BOTH intent AND identity — "it's the dream of any advertising person." Multi-phase natural language queries, each building context. Google loses you after one click; ChatGPT keeps the conversation.
Pillar 2: Drive outcomes without owning inventory. AppLovin (now $100B+) drives one outcome: mobile app installs. Built the entire infrastructure — buy side, sell side, and middleware. Controls the auction for most mobile apps like Google once controlled the web.
Pillar 3: Be the exclusive provider for large demand sources. Trade Desk: P&G says "here's our non-Google/Facebook display budget, figure it out." Must be exclusive.
What doesn't work: Being a middleman on top of Google/Facebook. "Every time you build a new capability on top of Google, turns out Google learns what you're building — and Google has the best engineers on the planet." AEO companies optimizing placement in answer engines will not create durable businesses.
What would scare Gokul if he ran a big ad network: Consumer behavior change. People using agentic interfaces instead of opening apps. "If they never open the app, you lose opportunities to advertise and you lose the relationship with the customer over time because they start trusting the AI agent."

North Star Metrics + Check Metrics

NSM should not be revenue. It should be directly correlated with customer value and should lead business outcomes. Square: GPV (payment volume). Facebook: DAU. DoorDash: GMV.
Check metrics are the guard rails. "If you tell a team to optimize this north star metric, they will do what it takes to go up 100%. But many things you don't want to go down could go down." DoorDash could grow GMV by setting delivery fees to zero — destroying revenue. Check metric: maintain gross margin %.
Facebook had an engagement budget. Between the newsfeed team and ads team: "We wanted this much revenue but the check metric was we can't take more than X% dip in engagement overall for newsfeed."

Leadership Lessons

Larry Page — Think in scale, not revenue. AdSense was the fastest growing product in Google history. Larry was disappointed. Why? "What percentage of all ads on the internet are you?" Less than 1%. He didn't care about the billions in revenue — he cared that Google was involved in serving every single ad on the planet.
Larry's philosophy on risk: Move approval from upfront gatekeeping to real-time monitoring. "Why do you need to approve publishers? Let it load for 100 times. After 100 impressions, then start reviewing it." Same approach for click fraud: "You just wait and understand what click fraud is and then you solve it. Be reactive when it needs to be solved."
Eric Schmidt — Communication through images only. Annual strategy presentation constraint: no words allowed, only images. "People don't remember words. They remember how things made them feel." For YouTube: just the hockey stick graph of videos uploaded per second. No numbers needed — the shape told the story.
Every great founder needs an "Eric." Zuckerberg had Sheryl Sandberg. Jack Dorsey had Keith Rabois. Tony Xu at DoorDash had Christopher Payne. A complementary operator.
Mark Zuckerberg — Best in the world at consumer engagement. Would sit in a room, look at a product flow, and instantly see what wouldn't be compelling to users. "You say, my god, why didn't I see that before?"
Zuckerberg invented Custom Audiences. Mark Pincus (Zynga CEO) kept demanding better whale targeting. Zuck said "why can't they just upload their whales into our system?" This became the foundation of modern ad targeting — upload your customer list, find similar users. "He has something about making connections between disparate domains."
Jack Dorsey — Design = no manual needed. "Good design doesn't mean visually pleasing. It means a product designed so well that you don't have to give your customers a manual on how to use it." Every POS except Square requires days of barista training. Square: download from the App Store, start selling.
Square shifted risk from business level to transaction level. Banks reject 95% of small businesses. Square accepted 95% — but ran ML models on every transaction. "That kind of lazy but brilliant onboarding characterizes a lot of good thinkers."

Self-Serve Is Non-Negotiable

Larry Page killed internal-only tools. When he saw the internal customer management system built for large advertisers, he said "I don't want — our small customers don't have access to it? End it right now." Turns out small self-serve customers adopted advanced features faster than large accounts because "entrepreneurs and hustlers exploit the system in ways you never even know."
Self-serve is the real definition: "The customer can onboard AND use the product without ever talking to or engaging with a single member of the employee base." This forces you to nail onboarding — which is where most people drop off.
Figma infiltration story. Gokul invested in Figma, then joined Square. Tried to push Figma top-down to the design team — they refused, sticking with Sketch. Two years later, a mid-level design manager brought Figma in from their prior company. It kicked out Sketch. Self-serve creates insurgent distribution that sales motions can't replicate.

Hiring & Career

Hire builders, not managers. "CEOs have gotten too comfortable hiring middle management. You're going to see the rise of AI agents with humans who manage AI agents as ICs." Management span of control less than 10 should not be allowed. If you manage 15 people and meet them weekly, that's 15 hours — what are you doing the other 30?
Work projects > interviews. Square's corp dev work project: "Give me one company Square should buy. Analyze it. Tell us why." The best PM candidates rejected the premise entirely — went to Mint Plaza, talked to 10 Square merchants on the street, and came back saying "none of them want this. We should build this other thing."
Tony Xu's hiring test: Give candidates $10-$20 and tell them to acquire 1,000 DoorDash customers. Nobody came close — "but the goal was to see how many different things they were able to try in a few hours."
Stay long enough to have impact. "12-18 month job hoppers is one of the biggest red flags. I don't think you can achieve anything of value in 12-18 months. Minimum 3-4 years." Posted this on X — tons of hiring managers agreed it's an immediate red flag.
Founder authenticity test: "Tell me your founding story." Google, Facebook, DoorDash all started as toy problems in college from authentic curiosity. "Just going out and starting a company because you want to start a company with your friend is the wrong reason."
Idea maze question: "Why did you choose this solution? Why not these five other ways?" The best founders are students of history. The Collisons bought a book on payments and studied exactly why every prior company failed and succeeded.

Board Management

Treat board seats like marriages. "Never invite anyone to join your board before spending at least a year with them." Have them join an advisory board first. Meet the management team. Attend board dinners. Then maybe make one of 3-4 advisors a board member.
Management team attends full board meetings now. 15 years ago, just the CEO and co-founder met the board. Now management teams attend everything except executive session. This lets the board assess succession potential and lets management leverage the board.
Board buddies. Each board member becomes a buddy to a management team member. They meet monthly between board meetings. "The meetings between board meetings are actually just as important as the board meetings themselves."

CEO Communication Format

Weekly CEO email — three sections: (1) Top of mind — product, business, team. What's keeping you up at night. Spend 60-70% of time here. (2) Performance update — how the company is doing on key dimensions. (3) Miscellaneous — recognitions, customer quotes, announcements.
Don't be afraid of repetition. "Repeating it once, twice, thrice, four times is when it actually seeps into their bones."
Be more candid than you think. "If you have good talent, ask them what they think you should do. People will rise to the occasion."

Go-to-Market

Consumer: Scale influencers. "TikTok is the best local search engine." Long-tail influencers go viral unpredictably — you need tools to capitalize on those viral waves.
Enterprise: Outcome-based selling. Palantir model: "What's your most important business problem? Give us 6 months. If we solve it, pay us a lot. If not, fire us." You cannot lead with what your product does anymore — lead with the outcome you can deliver.
Lighthouse effect. "If you get JP Morgan, every bank will evaluate your product. But if you get P&G, JP Morgan doesn't care." Go deep in one vertical, win the lighthouse customer, then win the rest.

Timestamps

Time	Topic
0:00	Intro
0:35	The Changing Nature of Product Development
4:09	The Merger of Product and Design
4:54	Managing Non-Deterministic Software
9:06	Judgment: The Future-Proof Human Skill
10:41	Building Durable AI Applications
16:43	The Risk to Legacy Software Companies
21:20	Sources of Stickiness in the Age of AI
23:43	Leadership Lessons from Google
27:41	Learning from Mark Zuckerberg
31:16	Jack Dorsey and the Philosophy of Great Design
35:48	The Product Manager as Editor
40:44	Three Pillars of a Successful Ads Business
49:03	Selecting North Star and Check Metrics
56:04	Hiring Functional Experts for the AI Era
1:00:06	Advice for Managing a Career
1:01:33	Evaluating Founder Authenticity
1:05:20	Best Practices for Board Management
1:11:15	The Kindest Thing

Relevance to SupportWire & FeatureOS

Zendesk is literally called out as endangered. Gokul uses Zendesk as the example of a seat-based company that AI agents will siphon value from. SupportWire's outcome-based model (per-resolution) is exactly the pricing shift he says incumbents must make but can't. This is validation of the SupportWire thesis.
System of record question is existential for FeatureOS. If an AI agent can collect feedback and build a roadmap, what's our moat? Per Gokul's framework: we need to be the system of record for product feedback, not just a workflow layer. The data (customer requests, votes, context) is the moat — if we own it and the half-life is long, we're durable.
Self-serve is non-negotiable. Larry Page's "end it right now" moment on internal tools applies directly. Every feature built for enterprise customers should be available self-serve. This is how Figma won. This is how Square won.
PM as editor = CEO as editor. Gokul's framework maps directly to the CLAUDE.md principle. The CEO's job is curation and killing, not feature generation. "In an AI age, humans with amazing judgment — which is really editorial capabilities — are the ones that will thrive."
Weekly CEO email format. Top of mind + performance + misc. Three sections. Spend 70% on what's keeping you up at night. Don't be afraid of repetition. Consider implementing this at Skcript.
Hire builders, test with work projects. Tony Xu's $10 customer acquisition test and Square's "reject the premise" PM interview are directly implementable for hiring at FeatureOS/SupportWire.
Lighthouse effect for SupportWire launch. Don't try to be horizontal. Pick one vertical, win the best company in it, then the rest follow. "You get JP Morgan, every bank evaluates your product."

Source: https://www.youtube.com/watch?v=JUsb1FYOstA

Raw Transcript

Auto-captions from YouTube. Folded by default — expand if you need to grep the source or pull an exact quote.

0:00 The one thing I think that's going to be 0:01 truly future proof is judgment. Why? 0:04 Because you have the big challenge of AI 0:06 slop. Every product leader I've talked 0:07 to is extremely worried that because you 0:09 have these engines running rampant, 0:11 they're just going to produce lots of 0:12 code. In an era when you can do 0:14 everything, the question is which of 0:16 these things matter and you should truly 0:18 do. 0:32 I thought an interesting place to start 0:34 would be the changing nature of how 0:37 people are building products. The 0:40 biggest story by far in technology seems 0:42 to be cla or claude co-work as well. The 0:44 ease with which both technical and 0:47 non-technical people are able to build 0:49 something that they can imagine. It 0:51 seems to have been just a complete 0:52 explosion in their ability to do so. 0:55 You've built a million things. You've 0:56 invested in 700 companies watching 0:58 people build things. You're about as 0:59 prolific as they come as a product 1:01 person. Maybe just give us your sort of 1:04 state of the union of how the world 1:06 feels to you in terms of technologists 1:08 building products and how fast that's 1:10 changing. 1:11 >> What was interesting about product 1:12 development is uh that 10 years ago or 1:15 even 5 years ago there were very clearly 1:17 defined roles. Product managers 1:19 articulated what to build, designers 1:22 designed it and engineers built it. Over 1:24 the last few weeks, over the last few 1:26 months, I've been talking to many 1:27 companies, but over the last two months 1:29 in particular, December and January, 1:30 December 25 and Jan 26, it's become very 1:33 clear that something has fundamentally 1:35 changed. And what that thing is is the 1:38 notion of a long horizon, a longunning 1:40 agent. I've experienced it myself about 1:43 6 months ago. I tried to use clot code 1:45 in the early days to build something. I 1:47 call it a video transcription tool. So, 1:49 I tried to build it. It kept failing and 1:51 then I had to go in and try to debug it. 1:53 Ultimately, I gave up. Two weeks ago, 1:55 while watching some episode of some TV 1:57 show, in one hour I was able to 2:00 basically prompt my way to a good video 2:02 transcription tool because these agents 2:04 now are resilient to failure. You and 2:07 you don't have to be very technical to 2:08 use them. This changes the expectation 2:12 of product teams. After I did that, I 2:14 started talking to three kinds of 2:16 companies. One, portfolio companies, 2:18 portfolio CEOs of companies I've 2:20 invested in. Second, the large AI labs. 2:22 and third a bunch of AI native young 2:25 companies to see what the similarities 2:27 are between them. So there are few a few 2:29 things that emerge. First product 2:31 development as we know it is changing 2:33 because the models and the capabilities 2:36 are growing so fast that if you try to 2:39 be very u strict and stringent about 2:42 exact describing exactly what you're 2:44 going to build or prescribing what 2:45 you're going to build it is going to not 2:47 work. So almost everybody has gone to a 2:50 bottoms up approach where it's not 2:51 driven by product management anymore. 2:53 Product managers the only thing they do 2:55 now is they articulate what the customer 2:57 needs are at the highest level and then 2:58 they are the guardian of the why. But 3:01 the actual product is built bottoms up 3:03 by engineers, researchers and product 3:05 managers and designers all working 3:07 together on the code itself. So 3:09 capabilities and models are changing 3:10 very fast. If whatever you think of 6 3:13 months ago, if you continue thinking on 3:15 that dimension, you have fallen behind. 3:16 So it's very very important for the 3:19 product managers to be understanding of 3:23 what these models are capable of and to 3:25 be hands-on. So they sit with the 3:27 engineers and the researchers and write 3:30 code, do prototypes, do anything and 3:32 everything it needs in a hands-on way. 3:34 So the first thing we are seeing now 3:36 happen is that PMs are starting to check 3:38 in code with either codec cloud code or 3:40 whatever into the actual production 3:42 repository. uh right now engineers have 3:44 to review the code but you're going to 3:46 soon see that clot codeex and other 3:48 tools actually review the code itself 3:49 before engineers commit all the 3:51 companies are struggling with how to 3:53 evaluate these people earlier there 3:56 there was nothing called the prototyping 3:57 interview now there's explicit interview 4:00 in the interview loop called prototyping 4:02 literally forces product managers to be 4:04 hands-on second the product manager and 4:07 designer role are merging increasingly 4:09 so the designer role is an interesting 4:11 role in particular a lot companies are 4:13 going through headcount allocation this 4:15 year and I'm hearing from many teams 4:17 that when given the choice between an 4:18 extra designer and extra engineer 4:20 they're saying you know what the design 4:22 systems are already laid out now that we 4:23 have the design system already laid out 4:25 we can use AI to do work around these 4:27 design systems so we need maybe a small 4:28 number of designers at the company level 4:30 to manage the design systems and the 4:32 design language but AI can leverage the 4:35 design language to do designs so please 4:37 give us an extra engineer so the number 4:39 of designers and product managers number 4:41 of engineers when I is growing up in 4:43 product. It used to be 1 to 3 or 1 to 4:45 10. It's going to 1 to 20 now. And then 4:48 I think the other very very important 4:50 thing that's happened which is 4:51 fundamentally different is when I was 4:53 growing up products were deterministic. 4:55 There was a workflow you knew if X 4:58 happened and a user did X Y happened. It 5:00 was very clear when you did X Y 5:02 happened. Today you could do X Y 5:03 happens. But if you do slight variation 5:05 of X completely something completely 5:07 different happens. Non-deterministic 5:08 software. What that means is you have to 5:11 be on the other side an evaluation or 5:14 what is called evals in AI and someone 5:16 has to evaluate whether or not what the 5:18 software is producing is reasonable or 5:20 not across various use cases. Obviously 5:22 they can be human evals, AI evals etc. 5:24 But who owns the evals? It's the PMs. 5:27 It's the PMs and the researchers. So the 5:29 PM's job is to be very clear at a high 5:32 level about what the user needs are and 5:34 then have a very clear sense of whether 5:36 this product is good to ship or not by 5:38 evaluating it. So you've got to actually 5:39 to evaluate it many times you got to 5:41 write AI yourself to evaluate the 5:42 results of AI because humans can't. So 5:44 PMs are very good at coming up with 5:46 evaluation techniques. So it's the 5:48 non-determinism of software the speed of 5:50 which things are going and overall the 5:53 notion that these things are just the 5:55 capability frontier is being pushed out 5:57 every two months makes it an incredibly 6:00 challenging yet incredibly exciting time 6:02 of product developer. If you think about 6:03 uh my friend Zach has this great way of 6:05 thinking about AI which is we had the 6:06 industrial revolution for goods and that 6:08 basically this kicks off an industrial 6:10 revolution for services. This is an 6:12 interesting opportunity to ask about 6:13 what your philosophy of product is. Um 6:16 you're such a product ccentric person 6:17 and builder that's that we that's what 6:19 you've done that's what you've invested 6:20 in. As we face down this like industrial 6:23 revolution for services 6:25 what what is your like broadest possible 6:27 philosophy of product as we enter this 6:29 era? 6:30 >> Very simple. A product person or product 6:33 manager if you call them their job is to 6:35 balance customer needs and business 6:37 needs. The product manager there has to 6:40 be somebody at the company who's the 6:42 keeper of the why. Why are we building 6:44 it? What customer need are we solving? 6:46 Why is this a pain point? How intense is 6:48 it? How deep it is? And second, how does 6:51 this add value to the company? If you 6:52 build this thing, solving this customer 6:54 need, how does value add to the company? 6:56 And I think balancing those two is a 6:59 very delicate act. You can build 7:00 something amazing that adds a tremendous 7:02 amount of value to the customer but 7:03 doesn't build any value to the business. 7:05 And you can do something that is awesome 7:07 for the business by raising prices but 7:09 is value detracting for the customer. So 7:11 balancing customer needs and business 7:13 needs at the highest level is what I 7:14 think of the product. And what it comes 7:16 down to in my opinion over the last 10 7:19 or 15 years I've really gone down to 7:20 this notion of outcomes. Outcomes I 7:23 think are what define the best product 7:26 people and outcomes have to be defined 7:29 in the form of customer behavior. I 7:32 strongly believe that the be because 7:35 customer behaviors are leading 7:37 indicators for every business outcomes. 7:39 If you think about it, the simplest 7:41 thing that a product does is to make 7:43 somebody go from not a customer state to 7:46 becoming a customer state and from 7:48 becoming a customer state to becoming a 7:49 loyal customer and then maybe to 7:51 becoming a loyal customer to become a 7:52 paying customer. So there are all these 7:54 different product states or if you do a 7:56 poor job they can go from becoming a 7:58 loyal customer to becoming a churned 7:59 customer. So these are all behaviors. 8:02 Everything you do or build should be 8:05 attuned to the goal of what customer 8:08 state change does it lead to? What 8:10 customer behavior change does it lead 8:11 to? So I tell every CEO I meet that is 8:14 trying to hire their first PM or doing 8:15 their first product review, you need to 8:18 ask why. The only question you need to 8:20 ask is why. Why are you launching this 8:22 feature? And you should not let any 8:24 feature go out if there's not a clear 8:27 hypothesis behind this feature. And the 8:28 hypothesis has to be articulated in the 8:31 form of a customer behavior change. We 8:34 we believe that by launching this thing 8:37 the customers will go from doing X to 8:39 doing Y or from spending X minutes a 8:42 month doing this to Y minutes a month 8:44 doing this. You have to have a 8:45 hypothesis which is grounded in some 8:48 data or some something you know about 8:49 the customer, some secret about the 8:51 customer. You mentioned at the start the 8:52 difference between the video 8:54 transcription tool six months ago versus 8:56 you know more recently and how quickly 8:57 that changed. It's just such a hard 8:59 future to uh reason about given the pace 9:02 of change. So how do you reason about 9:03 it? Like is there anything that can be 9:05 truly futurep proof? 9:06 >> Yes. The one thing I think that's going 9:08 to be truly future proof is judgment. 9:10 Why? Because what is the biggest 9:12 challenge you have when you have 9:13 thousand AI engineers writing code? You 9:15 have the big challenge of AI slop. Every 9:17 product leader I've talked to is 9:18 extremely worried that because you have 9:20 these engines running rampant, they're 9:22 just going to produce lots of code. 9:24 Which of this code is even valuable? 9:26 Which of these are even valuable? When 9:27 in an era when you can do everything, 9:29 the question is which of these things 9:31 matter and you should truly do on the 9:32 product side is judgment around what 9:34 needs to be built and evaluating the 9:36 output. on the engineer side is 9:38 evaluating the code because if you don't 9:41 understand what the code says I think 9:43 you can have engineers writing AI 9:45 engineers writing beautiful code that 9:47 could be wrong that could have bugs in 9:49 it that could be vulnerable someone 9:51 needs to review it and make sure you 9:53 have to have human review at some point 9:55 especially a critical code that is in 9:57 the core of your system and similarly in 9:59 design you have to have judgment around 10:01 does this make sense does it make sense 10:03 in the broader design system so I think 10:05 judgment is the number one thing that 10:07 humans are going to bring in an era of 10:10 infinite productivity. The question is 10:12 what are the things to be productive on 10:14 and are we building the right things? 10:15 >> As you evaluate companies today, build 10:17 things yourself and just think about 10:19 this problem and the trajectory of these 10:20 tools. Maybe walk through how someone 10:22 should think about building an AI 10:24 application like if if there's so many 10:26 people excited about it feels like a 10:28 gold rush with this new technology. so 10:30 many things that we can do that we 10:31 couldn't do before or things that people 10:33 specific people couldn't do because they 10:34 weren't technical that they can now do 10:36 how should people think about attacking 10:39 building something new an application 10:41 using AI starting today 10:43 >> first and foremost it has to be a deep 10:45 and compelling problem the good news is 10:47 there's a tremendous number of deep and 10:48 compelling problems today in every 10:50 vertical in every industry why because 10:52 till today till recently software was 10:55 used more as a tool by people by humans 10:58 we finally have software that is agentic 11:00 in nature which means it can do the job 11:02 of people. So the the question you have 11:05 to ask is where are what industry are 11:08 there roles of people that are highly 11:10 paid that are doing somewhat of a 11:12 repetitive job and that can be done by 11:14 software. Every 3 months the answer gets 11:17 deeper and deeper. You couldn't have 11:18 told me that a designer's job could be 11:21 automated by AI like 6 months or 9 11:24 months ago. You couldn't have told me 11:25 that an architect's job could be 11:27 automated by AI. a lawyer's job could 11:29 you auto? It turns out increasingly in 11:31 every vertical these capabilities are 11:33 getting better and better. So you want 11:34 to start with first and foremost what 11:36 industry do you want to be in and what 11:38 kind of job do you want to do. Second 11:41 you want to target a high value 11:43 workflow. You want to target a workflow 11:46 uh a way of working that is deep that is 11:51 complex and that is u that is basically 11:54 uh that that requires custom data. I met 11:57 with the CIO of a fortune 500 company a 12:00 few weeks ago. I think one of the 12:01 challenges with this with this whole 12:03 space is that the models are becoming so 12:06 good that if you try to build a company 12:08 that is light that is not a hard problem 12:11 the foundation model companies are going 12:13 to eat you. So this CIO that I met at 12:15 this company said I I was asking him 12:18 over a few startups I had invested in 12:19 and worked with. He said look I don't 12:22 know why I would use any of these 12:23 startups. Gemini has an agent builder 12:24 product and I also use Chad GDP 12:26 enterprise and they also have an agent 12:28 builder product and I have a thousand IT 12:30 engineers who work for me. 12:31 >> They all want to be retrained as AI 12:33 engineers. 12:34 >> So I'm just going to put them using 12:36 these horizontal tools to build my AI 12:38 agents. Why do you need any startups? 12:39 And so that's the kind of thing you're 12:41 going to face that if the CIO of a 12:43 company of your target customer can 12:46 build what you're building these agent 12:48 building tools, you're not going to be 12:49 successful. So you've got to really go 12:52 one step ahead of what can be built a 12:54 multiple steps ahead and you got to 12:56 extrapolate to where can the 12:57 capabilities of these agent building 12:59 products go and you got to do something 13:00 very very different. So what that means 13:02 is you've got to have an you got to have 13:04 durability because ultimately as venture 13:06 capitalists are or even as an 13:08 entrepreneur your time horizon can't be 13:10 building something that lasts for one 13:11 year and that's the biggest challenge. 13:13 It's not building an application. It's 13:15 building an application that's durable 13:17 that basically will last a test of time. 13:19 And I think there are a few things 13:20 around durability. One, you need to have 13:23 ownership of a scarce asset. Uh a scarce 13:26 asset could be it could be a license of 13:29 some kind. It could be a a regulation of 13:32 some kind where you have unique insight 13:34 into it. Second, you might need to you 13:37 might basically own a control point. A 13:39 control point is a thing that controls 13:41 how people interact with money or with 13:44 data. So if you you want to own that. 13:47 Third, you want to maybe have hardware 13:49 which is hard to replace. Fourth, maybe 13:51 you want to be part of an essential 13:53 workflow. Fifth, you want to have 13:54 network effects. You want to think about 13:56 those things and figure out how after 13:58 you take on that workflow, you can make 14:00 it more durable. And finally, I think 14:02 your ambition has to be to replace the 14:06 entire system. In other words, 14:09 increasingly what is going to happen and 14:10 I'm seeing this more and more is every 14:12 vertical has either a legacy or somewhat 14:15 new what is called a system of record 14:17 which is a system where most of the data 14:19 is stored for that system. For example, 14:21 in legal there's a company called 14:22 Filevine or another company called Cleo. 14:24 There's a few of these companies in 14:26 sales at Salesforce. In in healthcare 14:28 it's Epic. Now for many years these 14:31 companies all had APIs that if you enter 14:34 that industry you could build an agent 14:35 company on top of these APIs. 14:38 In 2025 things changed. These companies 14:40 started seeing that these agent 14:42 companies, AI companies that are being 14:44 built, they are starting to take on the 14:46 functionality out of these companies and 14:48 are treating them like a dumb database. 14:50 So you started seeing last year that 14:52 these companies are cutting off access 14:54 to APIs. Slack has done it most 14:56 publicly. Slack is owned by Salesforce. 14:58 They cut off access to Glean where Glean 15:01 can no longer access Slack data. And the 15:04 reason is they don't want Glean to build 15:05 on top of them and then slowly suck out 15:07 the value that Slack has. And I'm 15:10 hearing from other verticals that 15:12 they're doing one of three things. 15:13 They're blocking access to APIs. They're 15:15 offering their own agents for free 15:17 bundled or they're charging these AI 15:20 agent companies to access the data. Let 15:22 just to access data. The API was free. 15:24 They're saying now it's like $2 an API 15:26 call or something like that. So they're 15:28 basically making they're trying to make 15:30 the model of these agent companies 15:32 unviable. I think it's going to be very 15:34 hard for a end customer to use multiple 15:37 companies where you have a system of 15:39 record and then you have this agent that 15:40 sometimes doesn't work with it properly. 15:42 So the agent companies have no option 15:45 but to also start building and offering 15:47 a system of record. So every company I 15:49 know is now trying to figure out how do 15:51 I build the entire platform and not just 15:54 a system that does some workflows. I 15:56 think last year everyone was like, "Oh, 15:57 we can do workflows. We can build what 15:59 is called the system of action uh and 16:01 live on top of the system of record." I 16:03 don't think that's an option anymore. 16:04 >> The Slack example is a good one of uh a 16:06 sort of last generation software company 16:08 which was very big and very successful. 16:10 One of the most interesting investor 16:11 questions and I'm curious for your 16:13 answer from the perspective of a builder 16:14 and a technologist is that uh the degree 16:17 to which these horizontal model 16:19 companies are going to destroy or be 16:21 very bad for old software companies 16:23 because over time it will be trivial to 16:26 spin up your own Slack that has features 16:29 that you want for your company and it's 16:31 very reliable in all the same ways that 16:32 Slack is and therefore Slack's in a lot 16:35 of trouble. How do you think about that 16:37 question of like obviously public 16:39 markets seem to think software is in a 16:41 lot of trouble. The multiples are really 16:42 really low. How much would you be 16:43 worried if you ran like a good solid but 16:46 older software company today? 16:47 >> There are two or three kinds of software 16:49 companies. I think the the software 16:50 companies that are should be the most 16:52 worried right now is where they are 16:53 pricing the product based on utility. 16:57 Zenesk is a good example where literally 16:59 Zenesk prices seats and each seat comes 17:02 with utility. In other words, each seat 17:05 corresponds to a customer service agent 17:06 that tax certain number of customer 17:08 tickets. So that company should be 17:10 worried. Why? Because I can have an AI 17:12 agent sit right next to Zenesk and you 17:14 can slowly siphon off. You can use 17:16 instead of paying for 50 Zenesk seats, 17:18 you can pay for 20 and I can have 30 AI 17:21 agents sitting next to Zenesk and that 17:23 siphoning can hap happen over time. You 17:25 don't have to have a all-in-one 17:26 decision. It can be a two-way door 17:27 decision. Those are the most endangered 17:29 companies in my opinion. You need to 17:31 change your pricing model to be based on 17:32 outcome and you need to actually build 17:34 the product to be based on outcome. It's 17:36 easier said than done because literally 17:38 you're going from a 20 or $30 per seat 17:41 to maybe charging a buck or 50 cents or 17:43 20 cents per ticket result and you don't 17:46 know how that's going to turn out. So 17:47 you've got to change your pricing model 17:48 and I think that's a very challenging 17:50 thing. That's why I think many of them 17:52 probably need to go private because they 17:54 have to make this business model 17:55 transformation in private. I think it's 17:56 going to be hard for them to stay 17:57 public. The companies that are less 17:59 exposed are ones where the utility is 18:01 not based on seats but it's based on 18:02 data that has been collected and 18:04 captured over a period of time and the 18:06 the more uh timeless the data is the 18:09 more protected they are. Slack for 18:11 example I would say might be in a little 18:13 bit more precarious state because the 18:14 data in Slack is half time halfife is 18:17 very short that's a great way of putting 18:18 it but if you have ERP is a great 18:21 example somebody uses Netswuite as a ERP 18:23 now I don't know if how Netswuite 18:25 actually charges but it doesn't matter 18:26 however many seats you buy the reality 18:28 is it runs your whole business and there 18:30 is no compelling reason for someone to 18:32 put their career at stake by ripping out 18:34 Netswuite I know there's a lot of now 18:36 over the last year there's been a lot of 18:37 AI enabled ERP businesses but there's 18:40 There's no compelling reason to take 18:42 Netswuite and say I'm going to rip it 18:43 out because it is career limiting to 18:45 suddenly take Netswuite out when you're 18:46 a company running on Netswuite. So I 18:48 think those companies are much more 18:50 insulated and I think obviously and you 18:52 could argue that Netswuite has more time 18:54 to build AI agents on top of it because 18:56 they have the data they have data and 18:58 they can train the AI agent on top of it 19:00 and bundle it. So you could essentially 19:02 I think the software public markets are 19:04 not distinguished between these two 19:06 types of companies. Companies where the 19:07 half level data is low and where you can 19:09 actually have you can literally take 19:11 half of the value of this company and 19:13 put it onto an AI company that sits next 19:15 to it. Well something like an ERP system 19:18 or even Salesforce for sales data and 19:20 records those are real customer records. 19:22 It's going to be hard. So what are AI 19:24 native companies doing? The first thing 19:26 you've got to do if you ever have to 19:28 compete against them is you got to spend 19:29 a year or two first building a system 19:33 that literally takes migrates your 19:36 Salesforce instance to your own 19:37 company's platform. One of the one of my 19:40 companies is Na native company. They 19:42 literally hired engineers in a European 19:44 Eastern European country for 2 years to 19:46 build this migration thing transition 19:48 tool. So you have to build the migration 19:50 tool because 19:51 >> who's going to migrate it? you can just 19:52 present your spanking new system but 19:54 this data is still there even for square 19:56 for a small business I remember they had 19:59 a point of sale they wouldn't they 20:00 wouldn't move to us even though it was 20:02 cheaper because they had gift cards 20:04 customer data loyalty data payments data 20:07 all of that you know even credit cards 20:08 so we had to build scripts and and that 20:11 took us months or years to build it for 20:13 a simple POS for something like 20:15 Salesforce you can't just say well here 20:17 I am I'm a great I'm a much better CRM 20:20 because I connect there is this thesis 20:22 which I completely agree with that if 20:24 you look at CRM what does a CRM contain? 20:26 It contains your customer record. Your 20:28 customer support system contains what 20:32 your customers are complaining about and 20:34 Jira or Atlacian contains what your 20:36 product development team is building. 20:38 Now all of these things should be linked 20:40 right because there is no linkage. You 20:42 you should be building the biggest you 20:45 should be addressing the biggest 20:46 complaints of your customers which are 20:47 in Zenesk and you should those Zen 20:50 customers you should know where they 20:51 came from who bought them who sold them 20:53 what the AM is. So all these three c 20:55 three systems should be linked together 20:56 but they're all three different 20:57 companies. So they're companies that are 20:58 trying to unify these things and it's a 21:00 great value prop but guess what none of 21:02 your customers is ever going to move 21:04 unless you build a simple seamless way 21:08 to take the Salesforce data and move it 21:10 to your instance. the data from GM move 21:12 to your instance the Zender data move to 21:14 your instance. So literally it's a 21:16 two-year effort to build migration 21:18 otherwise you've got to get Accenture. 21:20 >> How do you think about um stickiness in 21:22 this era just as a general concept when 21:25 the friction for creators to build 21:27 something net new is so easy is so low 21:29 you can do whatever you want really 21:30 fast. How do you how's anyone going to 21:33 use anything for a long period of time? 21:34 >> The age of AI stickiness I think comes 21:37 from a few sources. I think you need to 21:38 have network effects. So Door Dash is 21:40 sticky not just because it has this 21:42 beautiful app, but it's because it's a 21:44 network of restaurants and dashers and 21:47 consumers. So you can't just attack one, 21:49 you've got to go, 21:50 >> you can't vibe code your way to those 21:51 two. 21:52 >> Exactly. And so network effects. Uh 21:54 second u second example of stickiness is 21:57 when you have financial or money moving 22:00 through you. I think that's another way 22:01 to be sticky. I think uh many of the 22:04 system of records I think like for 22:05 example toast have payments going 22:08 through them and I think that really is 22:10 interesting because you can't just start 22:11 building the point of sale you also have 22:12 to have money flowing through it and I 22:14 think uh if you look at the banks banks 22:17 are a good example a business bank once 22:19 you have something like mercury as a 22:22 business bank it is hard you have money 22:23 flowing through it is hard to then 22:25 switch because you have regulations 22:26 other stuff embedded so I like things 22:29 that are combination of financial 22:30 services and software because of That 22:32 the third stickiness is from hardware. 22:34 You can actually have hardware. Toast is 22:36 a good example where toast gives you 22:38 hardware for free but if you try to give 22:41 return the hardware you have to pay 22:42 them. But either case the hardware is 22:44 there and somebody can't just build 22:46 software. They also have to take 22:47 hardware and put it into the thing and 22:48 rip out the toast hardware. The fourth 22:50 one is uh access to a uh unique asset. 22:56 Uh, and I I was thinking about a good 22:58 example and I came up with the example 22:59 of Sierra, which I think the unique 23:01 asset is Brett Taylor. I mean, they have 23:04 full control of Brett, who's one of the 23:05 best salespeople, chairman of Open AI. 23:07 He can make a call to any company, any 23:09 country, and they'll take his call. You 23:12 can't really outsell Brett. And so, I 23:14 think there's alpha in that. And so, I 23:16 think there are, you need one of these 23:18 four or five things which are basically 23:20 indicators of durability. The halflife 23:22 of software today is so short that 23:24 unless you're one of these things that 23:26 make it durable. Harrison Helmer has 23:28 this thing called uh seven powers. And 23:30 so you got to have a few of those seven 23:32 powers that that basically are embedded 23:36 in the business model from day one. 23:38 >> You you've been so lucky to work for 23:40 some of the most well-known CEOs and 23:42 founders of this sort of modern era. I'd 23:45 love the chance to ask you a little bit 23:46 about each of them and what you learned 23:48 from them and then more generally just 23:49 things you've learned about what great 23:51 leaders do to run companies. Maybe going 23:53 back all the way back to Google and 23:54 starting with Larry, Larry and Sergey, 23:56 what what did you learn from watching 23:58 them operate and lead? 24:00 >> Yeah. One of the most interesting things 24:01 about all the leaders that I've worked 24:02 with which I think have built 24:04 generational companies is that they have 24:05 a superpower that is very aligned with 24:08 what the company needs to succeed. And 24:10 the company was really shaped in their 24:12 image. the company, the culture, the 24:14 early hires, the products. When I joined 24:16 Google, I joined Google in 2003 January. 24:19 The first product I got exposed to 24:21 actually which I didn't know about was a 24:23 product called Caribou. Caribou was an 24:25 internal code name for a product that 24:26 was launched on April 1st, 2003. 24:28 Publicly, it was called Gmail. 24:30 >> I I I didn't believe that this product 24:32 existed because in the internal alpha, 24:34 it said this gives you 1 GBTE of 24:36 storage. Back then, remember, Yahoo mail 24:38 was the dominant product and it gave 10 24:40 megabytes of storage. So this thing had 24:42 100x more storage and this really 24:44 epitomizes Larry and Sergey's philosophy 24:46 which was basically built the best 24:48 technology on the planet. They were 24:50 deeply technical and every product was 24:52 held to technology and scale and I'll 24:56 never forget AdSense was the fastest 24:58 growing product in Google history and we 25:01 went in to reviews and Larry would be 25:04 disappointed in us and we asked why. 25:05 It's like what percentage of all ads on 25:07 the internet are you be like like less 25:10 than 1%. His goal was not to again he 25:13 didn't care about the revenue. He cared 25:15 that Google is involved in serving every 25:18 single ad on the planet versus like 25:21 making a making a business of like 25:22 whatever billion or two billion or 10 25:24 billion. So the focus on scale and the 25:27 focus on technological superiority and 25:30 that investment Google Street View I 25:33 mean and and basically TPUs 25:36 uh Whimo all of these I think show the 25:39 10 plus years of investment to an 25:41 uncertain future but knowing that if you 25:43 invest in technology good things are 25:45 going to happen and good things happen 25:46 but it took a decade and that's 25:48 investing in technology capabilities. 25:50 Before we leave Google, um you had this 25:52 interesting idea about communication and 25:54 Eric Schmidt obviously another key 25:56 Google person. Can you talk tell the 25:57 story about him presenting the company 25:59 strategy using nothing but images and 26:02 just like a this is like an interesting 26:03 example of communication? 26:05 >> Yeah, Eric Eric was a I mean I think one 26:07 of the interesting things I've seen is 26:09 that the other interesting thing I've 26:10 seen is that almost every great founder 26:12 or founding team needs an Eric needs an 26:15 Eric figure. If you look at it, Mark 26:17 Zuckerberg had Cheryl Sandberg, Jack 26:19 Dorsey at Keith Reoa and Tony at Door 26:22 Dash, we had Christopher Payne. So, 26:24 everyone had somebody who was 26:26 complimentary to them and really helped 26:29 uh you know they they're amazing at say 26:31 technology and scale. Eric was amazing 26:34 at bringing a team together leading and 26:36 I think Larry and Sea learned a lot from 26:38 it and Larry of course became CEO after 26:39 Eric stepped down but Eric was 26:41 incredible. So uh Eric would give a 26:44 product leader. We would become 26:45 secundered to Eric for uh the weekly 26:48 strate or the annual strategy planning 26:49 session. So I did it I think in 2007 26:52 where my job was to go to Eric and say 26:54 Eric how do you want to present the 26:55 strategy of the company? He's like well 26:57 it's very simple. I want you to go and 26:59 interview each of the folks each of the 27:01 different leaders of uh the different 27:02 teams. There's only one constraint I 27:04 have. I'm like what is that? You can't 27:06 use any words to describe what they're 27:08 doing. 27:09 >> I'm like what do you mean you have to 27:10 use words? Nope. you've got to use only 27:13 images. I'm like, why is that? He's 27:15 like, people don't remember words. They 27:17 remember how things made them feel. And 27:19 you can put words in the speaker notes 27:21 I'll use, but I want you to come up with 27:22 the most compelling image that that 27:25 exists for what they're describing to 27:27 describe. And so it was a crazy thing 27:29 because I never thought of doing a 27:31 presentation that way. And uh so I went 27:34 to you know each each of the businesses 27:37 adwords, search, YouTube, AdSense and 27:40 then had to come up with a compelling 27:42 image that was easily accessible to the 27:45 whole company yet represented what they 27:47 did. 27:47 >> Do you remember like a specific image 27:49 like an I'm so interested by this 27:51 exercise. It seems like potentially 27:53 productive for anyone to try to jam what 27:54 they're trying to say into only images. 27:56 And so I'm trying to pin down like an 27:59 image and how you how you arrived at it 28:00 or 28:01 >> I think for YouTube it was a graph. It 28:02 showed the graph the number of videos 28:04 that being uploaded every second how it 28:07 had changed from the time Google brought 28:08 them to them. So it was not even a 28:10 graph. It was literally showing this 28:11 incredible hockey stick that happened 28:13 over the last 18 months and then it it 28:15 had I think we couldn't even show the 28:17 numbers. So the the thing had to be 28:18 compelling enough that it could just the 28:20 line would have to be like a U or 28:22 something like that when it went like 28:23 that because we just show it like this. 28:25 you have to say something 100x or 28:26 something where you couldn't say that. 28:27 So, so we had to show that was the one 28:29 thing we wanted to show that Google 28:30 search has gone from being used by small 28:33 and midsize companies to being used by 28:34 the largest companies in the planet. We 28:36 showed the logo of um of I think they 28:38 had a very large Fortune or Fortune50 28:42 company that they had acquired. 28:43 >> Your finance team isn't losing money on 28:44 big mistakes. It's leaking through a 28:46 thousand tiny decisions nobody's 28:48 watching. Ramp puts guard rails on 28:50 spending before it happens. Real-time 28:51 limits, automatic rules, zero 28:53 firefighting. Try it at ramp.com/invest. 28:57 Every investment firm is unique, and 28:58 generic AI doesn't understand your 29:00 process. Rogo does. It's an AI platform 29:02 built specifically for Wall Street, 29:04 connected to your data, understanding 29:05 your process, and producing real 29:06 outputs. Check them out at 29:08 rogo.ai/invest. 29:10 The best AI and software companies from 29:12 OpenAI to Cursor to Perplexity. Use Work 29:14 OS to become enterprise ready overnight, 29:16 not in months. Visit works.com to skip 29:19 the unglamorous infrastructure work and 29:20 focus on your product. 29:22 >> What did you learn from Zuck? 29:23 >> Zuck was u and is actually I think the 29:26 greatest mind on growing building growth 29:30 and engagement in consu building 29:32 consumer products broadly. I've seen him 29:35 basically sit in a room and critique 29:37 some a product team would have come in 29:39 with a very wellthoughtout 29:41 product uh consumer product flow and he 29:44 would look at the flows and he'd say 29:45 that is not going to be compelling to 29:47 users that is not something that a user 29:49 is going to engage to change it to this 29:52 and you say my god why didn't I see that 29:54 before so he's very very good at 29:56 thinking about how consumer product 29:58 should be designed to maximize 30:00 engagement and maximize just growth both 30:03 is probably the best way to put it. The 30:05 second thing he's amazing at is learning 30:07 by following. When I joined, I was uh my 30:10 task was to lead the ads product team 30:12 and Zach at that point knew a little bit 30:14 about ads uh because he had worked with 30:17 Cheryl quite closely. Cheryl had worked 30:18 on ads before. But then within I think 30:21 about a year he shadowed us. He came to 30:24 the ads team. He basically sat with us. 30:26 He came to many of our meetings and 30:28 within a year he got to the point where 30:31 he was generating ideas for the ads 30:34 team. One of the most foundational ideas 30:36 of Facebook ads came from what is called 30:39 custom audiences. Custom audiences is 30:42 the foundation of most ad systems now is 30:44 the is the idea that you as an 30:46 advertiser you want to reach people who 30:49 are similar to your customers. So if 30:51 you're a bank and you have say 100,000 30:53 customers, how can you give this set of 30:56 customers to your ad platform and say 30:58 look instead of describing these 31:00 customers right what do what did before 31:03 they would describe their customers I 31:05 think they are 25 to 34 year old women 31:07 that's too that's not good enough 31:10 instead if you can just tell us who your 31:12 customers are and we can map it to our 31:14 users we can then find people similar to 31:16 them so uploading that data into our 31:19 system securely early and doing it in a 31:22 way that doesn't compromise an EPI was 31:24 was the key thing and it all came from 31:26 Zuck. How? Because Mark Pinkers was the 31:29 CEO of Zinga. Zinga was the largest 31:30 advertiser on Facebook. Zinga basically 31:32 wanted to like most gaming companies 31:34 they were very focused on acquiring 31:36 Wales. 31:37 >> Uh because Wales for any gaming company, 31:39 casino etc. 80% of all probably all the 31:42 betting companies 80% of all revenue for 31:44 any gaming company comes from Wales. So 31:47 he was very frustrated at us. We would 31:49 do uh these quarterly reviews with Zingo 31:51 on the ad side because they were large 31:52 spenders on ads and they would 31:54 constantly be yelling at us saying we 31:56 want to get more whales. We were like 31:57 yeah you're getting users and you it's 31:59 your idea you need to figure out how to 32:01 get whales from your games. What do you 32:02 want us to do? We can help you acquire 32:04 users. So he once I think talked to Zuck 32:06 and Zuck came to us and said why can't 32:08 they just upload their whales into our 32:11 system? We know who the whales are. Why 32:13 can't we just find them people similar 32:14 to those whales? We were like that's 32:16 interesting but we actually didn't know 32:17 who the whales were. So they needed to 32:20 tag it for us who the whales were and 32:22 and basically we started doing it 32:24 similarly. We started finding users 32:26 similar to the whales that they had and 32:28 it worked so well. Then we said why 32:30 don't we take this approach and use it 32:32 for other types of customers who we 32:34 didn't have data on and it became truly 32:37 it was a transformative thing for ads 32:38 and it was all it was all Zuck's idea. 32:40 He just has something about connect 32:42 making connections between disparate 32:44 domains which is uh pretty pretty 32:46 amazing and unique. Jack is the I mean 32:49 he's I think on par with Johnny Iv and 32:50 Steve Jobs as in terms of his thinking 32:53 with design. I understood what good 32:55 design means. Good design doesn't mean 32:57 visually pleasing. It means a a product 33:00 that is designed so well that you don't 33:03 have to give your customers a manual on 33:05 how to use it. They should be able to 33:06 see the product and use it. Think about 33:08 your point of sale. Every point of sale 33:10 except Square and things that have 33:11 copied Square, you have to train a 33:13 barista still for several days after 33:15 they join on how to use the point of 33:17 sale. Square is something you can 33:18 download from the app store and start 33:21 using it as a point of sale to run your 33:23 business. A category where you have to 33:25 you have to you have to train somebody 33:27 for weeks. That's the example of a of a 33:30 good design. He brought that to every 33:32 part of the company and removing 33:34 friction from what is traditionally I 33:36 mean Square's whole premise was removing 33:38 friction from small businesses applying 33:40 for financial services and that extended 33:42 to the product that also extended to 33:44 risk. One of the most interesting things 33:46 that I didn't realize is that Square at 33:47 its core is a risk company. when you 33:50 apply to a bank for payment processing. 33:52 In fact, the company was founded because 33:54 Jack's co-founder Jim was rejected many 33:56 many times to accept AMX uh by by banks. 34:00 He was a fairly successful glass blower 34:02 in St. Louis and uh he basically was 34:04 selling two $3,000 glass sculptures to 34:07 people who would send him checks. So, a 34:10 woman called from Panama one day and 34:11 said, "I want to buy this on his 34:13 website." He had this beautiful piece of 34:14 glass. He said, "Great." They agreed on 34:15 the price and she said, "I can you take 34:18 my credit card number?" So he said, "I 34:19 don't accept credit cards." So she said, 34:21 "Sorry, I can't send you your travelers 34:23 check or check or whatever the case is." 34:24 So he lost the sale. And so he went to 34:26 his friend Jack Dorsey. They had never 34:28 built hardware. They had never done any 34:30 of that stuff, but they brainstormed and 34:32 realized that the phone, the iPhone, 34:34 which had just been released a couple of 34:36 years ago, had this thing called the 34:38 audio jack that basically could be used 34:40 to uh put a piece of hardware in and 34:43 process cards. I I can't even imagine 34:44 the leaps you have to make to get there. 34:46 But the number one thing that they 34:48 realized is people most of most small 34:50 business are denied by banks when they 34:52 apply. Square instead said we are going 34:54 to accept 95% and but what they did was 34:58 they put risk at the transaction level. 35:01 So they accepted you as a person as a 35:03 business but then once you started 35:05 processing transactions they would then 35:07 run machine learning models and every 35:08 transaction this transaction risky this 35:10 is not. shifted the level 35:11 >> shifted the level and so that kind of 35:14 lazy but brilliant onboarding is 35:16 something that characterizes a lot of 35:17 good thinkers Sergey very similar I've 35:19 come up with this conclusion when we're 35:21 going to launch AdSense in 2003 I I'll 35:24 never forget this 2003 May was when we 35:26 were doing our final launch things 35:27 Sergey was our sponsor he came and sat 35:29 in the meetings he said what are you 35:30 guys building here we're like oh you 35:32 know website publishers are going to 35:33 apply from all across the world it's a 35:35 self-s serve product we have to review 35:36 them we have to review them and say we 35:39 should approve them not approve approve 35:40 them to run AdSense. He's like, why do 35:42 you need to approve them? We were like, 35:43 what do you mean? Our ads are going to 35:45 we are going to be running ads on on 35:47 these things. Google ads or ads powered 35:50 by Google. You don't want to be on a 35:52 porn site or something else. He's like, 35:53 why not? We didn't really have good 35:55 answer to why not. I was like, well, you 35:57 know, standards or like policies. Okay, 36:00 but what if they lie? He was right. What 36:03 if they lie? Like I could We had so many 36:06 people applying with Nike.com, for 36:07 example. It's true. It was very hard to 36:09 know who owns a domain, right? I could 36:11 apply with with your domain uh and 36:13 basically, you know, get accepted. He 36:15 was right in some ways. We were just 36:17 doing it to cover our asses, turns out. 36:19 And so he said, "Okill all this." So we 36:21 had literally spent half of our 36:23 engineering team building this complex 36:25 approval system with ops and so on. Ops 36:28 are super excited. They hired a lot of 36:30 people and now you're telling us not to 36:31 do and instead do it in real time for 36:34 every page that loads because we had the 36:37 JavaScript on it. We know what URL it 36:40 is. Look at the content at that point. 36:42 >> And we were like it's too slow. We won't 36:44 be able to look at the content because 36:45 it's billions of pages. That's fine. Let 36:48 it load for 100 times and after 100 36:50 impressions if any URL hits 100 36:52 impressions then start reviewing it. not 36:55 trying to put lots of checks up front, 36:57 >> but being intentional about where and 36:59 why. Most things don't even get to the 37:01 level where you care about. So only do 37:03 stuff. The same thing happened with 37:05 click fraud. Click fraud was one of 37:06 these biggest challenges that we faced 37:08 and where people click on their own ads 37:10 and make money. How the hell do you 37:11 solve that? The reality is you don't. 37:14 You just wait and you start 37:15 understanding what click fraud is and 37:17 then you solve it. So be reactive and 37:20 solve it when it needs to be solved at 37:22 that point versus waiting. So the square 37:24 thing was exactly move risk from the 37:26 business level to a transaction level. 37:28 The same with AdSense. Move risk from 37:30 the publisher level or the you're 37:32 basically you cannot gate because 37:35 getting somebody to come to you and sign 37:37 up is one of the rarest things in 37:39 history. Someone is coming to you and 37:40 expressing an interest and you're saying 37:42 you're going to put 10 different 37:44 barriers. That's the opposite of self-s 37:46 serve. 37:47 >> So pure self-serve product would never 37:49 have any reviews of any kind. You're 37:51 going to be immediately activated. go on 37:53 and we'll do checks in real time based 37:56 on what you're doing versus banning you 37:58 or stopping you. 37:58 >> You me in both these amazing examples 38:00 and then you also said that Jack would 38:02 do this across the company, not just in 38:04 the product. How would you sum up the 38:05 process of great design that you've 38:08 observed from the people that are the 38:10 best at design? What is the what is the 38:12 thing they're the method that they're 38:13 going through over and over again as 38:15 they apply it to different parts of the 38:17 company or product? 38:18 >> The number one thing I've seen is they 38:19 try to minimize the number of steps. 38:21 Everything should be in one page and you 38:23 need to cut down things. In fact, Jack 38:25 called the product manager role product 38:27 editor. Why? Because he believed rightly 38:30 so that the role of the product manager 38:31 is not to add more features. Any of us 38:33 can look at a product and say here's 10 38:35 things you should build. The best fe the 38:37 best designers, the best product people 38:39 edit down things. Similarly, we have 100 38:41 features. What are the two things that 38:43 really matter that will drive the 38:45 customer outcome? So the best designers 38:48 really take 10 pages of design and say 38:51 cut out all the experience. So I think 38:53 it's the process of editing and this 38:55 goes to judgment. I think this is in an 38:57 AI age humans with amazing judgment 39:00 which is really editorial capabilities 39:02 are the ones that are going to do well 39:03 and thrive. 39:04 >> Apparently uh Rick Rubin would say that 39:06 he wasn't a producer he was a reducer. 39:10 >> Great example reducer. I like that. 39:12 >> I wonder how that applies also to 39:15 communication. Um maybe this is a fun 39:17 opportunity to ask you about the format 39:20 that you've lighted on that a leader can 39:22 send to his team on a weekly basis. I 39:25 think it seems like this idea of 39:27 reducing and simplifying can be applied 39:29 in so many ways by great leaders. Talk 39:31 about it in terms of communication uh 39:33 from leadership to a team. One of the 39:34 things that people especially founders 39:37 of startups don't realize is initially 39:40 most startups start with two or three 39:42 people and then they go to people who 39:44 are all sitting in a room together. 39:45 Everyone can hear what you're saying. 39:47 But as soon as a company goes into I 39:49 think I call it two rooms where they're 39:51 not in the same room together. Then you 39:53 have to communicate. You have to you 39:55 have to let people know what's going on. 39:57 You have to bring everyone together. And 39:58 there are a few artifacts that companies 40:00 need to start putting into place. One is 40:02 a notion of an all hands where I think 40:04 an all hands it seems cliched but an all 40:07 hands is actually and it seems 40:09 unnecessary but even with a 15 20 person 40:11 company just getting together once a 40:14 week um maybe on a Friday or a Monday 40:16 depending or Thursday and and basically 40:18 just sharing what people have built have 40:21 been working on in a way and then having 40:23 the leader address uh everyone or one of 40:25 the leaders address everyone is a great 40:27 way to get people together. The second 40:29 thing is a weekly CEO email. And I think 40:32 this is a very powerful way for the CEO 40:35 to get across to the to the team what is 40:38 on their mind. The best way I think is 40:40 uh that I've done myself is during the 40:42 course of the week, you start jotting 40:44 down things that you think you want to 40:46 communicate and then you'd spend Sunday 40:48 or Saturday, whatever the case may or 40:50 taking all of those things and adding it 40:52 to two or three things that matter that 40:54 you want to get across. Most businesses 40:56 I think can be communicated along three 40:58 dimensions. Progress, product, business 41:01 and team. What's happening on the 41:03 product? How is it becoming more 41:04 remarkable or serving our customers 41:05 better? What's happening on the business 41:07 side? How are we doing better as a 41:08 business? And then what's happening on 41:10 the team front? Who have we added, 41:11 subtracted? What changes have we made? 41:12 And most importantly, don't be afraid of 41:15 repetition. Don't be afraid of 41:17 repetition because repeating it once, 41:19 twice, thrice, four times is what people 41:22 that's when people actually it seeps 41:23 into their bones. What is the literal 41:25 format that you do? So you've got three 41:27 in your email. What is the structure 41:28 that you do personally? 41:29 >> So the format I've used in the past and 41:31 what I recommend and what people I've 41:33 seen now I've seen at least 15 CEOs 41:35 adopt it and to good effect is three 41:38 sections. One is called top of mind. So 41:40 this is product, business and team. Like 41:42 what is top of mind on the product side, 41:44 on the business side, on the team side. 41:45 Doesn't need to be all three. What's top 41:46 of mind for you? What's keeping you up 41:48 at night? I think this is the thing that 41:50 literally everyone is hanging on to. I 41:52 mean because I remember seeing it from 41:54 from Jack, from Mark, from Cheryl. I I 41:57 think just seeing it put in paper or put 41:59 in an email is just so powerful. That's 42:02 one. The second thing is performance 42:04 update. I think everyone wants to truly 42:06 understand how's the company doing. 42:07 How's the company doing on the 42:08 dimensions, I think. And this is where 42:10 especially being a startup, I think most 42:12 people are one dimension removed from 42:14 how the company is doing. They all want 42:15 to know that they're doing well. And I 42:17 think this is the way. And the third is 42:18 miscellaneous is things like recognizing 42:21 specific people. It's quotes from 42:23 customers. It's maybe an off-site 42:25 announcement. But the most important 42:27 section where you should spend 60 or 70% 42:29 of your time on is top of mind. 42:31 >> How transparent should one be in that? 42:33 As a leader of a business, I could tell 42:35 you what's top of mind, but a lot of it 42:37 either might be sensitive or uh I would 42:41 worry about scaring people or worrying 42:43 people about something that I'm thinking 42:44 about or worrying about. like what keeps 42:46 me up at night might create stress in 42:48 the business. So like where where where 42:50 should one draw the line in terms of how 42:52 candid they are about 42:53 >> I personally think more candid is better 42:55 than less why but if you're more candid 42:57 what you can do is you can actually get 42:59 people you can actually ask people to 43:01 suggest ideas and that's the thing I 43:03 think you by by you just if you have 43:06 good talent at the company if you 43:08 actually ask them what do you think I 43:09 should do what do you think we should do 43:10 in this situation I think people will 43:13 rise up to the occasion especially when 43:14 the company is small we want people more 43:16 input and there's a oneway road decision 43:18 that we're going to make where making it 43:20 takes us one way or the other. I think 43:21 it'd be great to get um get get feedback 43:24 from more people. 43:25 >> I want to talk about ads and um 43:27 everything you've learned about building 43:28 like an incredible ads product. You 43:30 basically have built like the core 43:32 business the important core business 43:34 engine at multiple places at at the sort 43:36 of main character company across across 43:38 your career 43:38 >> as a company. You either die or you live 43:41 long enough to become an ads company. 43:43 And so we are seeing now with OpenAI 43:45 it's happening. Now how do you build an 43:46 ads business? There are three 43:48 fundamental ways to succeed in the ads 43:50 business. Three and only three. One, you 43:54 need to own a very coveted uh group of 43:58 users and you need to have a surface on 44:00 which those users with which those users 44:02 interact. Google search is a great 44:03 example. It's a surface on which a very 44:06 coveted set of users interact with. U 44:08 obviously they express high intent. So 44:10 Google is one of the most profitable ad 44:11 businesses. Facebook very similar. It 44:13 took us a while to figure out what was 44:15 coveted of both these users. Turns out 44:16 what was committed was the identity. We 44:18 knew who these users were and we could 44:20 match them to customer and other data 44:22 and so you could precisely target these 44:24 people with messages you wanted and you 44:25 could find people similar to them. Chat 44:27 GPT their combination of intent and 44:30 identity data is unparalleled. I mean 44:31 Google had intent data but not identity. 44:33 Facebook identity but not intent. These 44:35 things been both together. I mean it is 44:37 a it's the dream of any any advertising 44:40 person. I mean shoot I mean I don't know 44:42 how many searches they see but they 44:43 going to see they're going to see more. 44:45 And these are complex complex 44:47 multi-phase searches, right? That's the 44:49 other beautiful thing. You search or you 44:51 and each of the queries is kind of like 44:53 a search and then you search again and 44:55 you're just building up searches. At 44:57 Google, you typically search and then 44:59 you lose the person because they go off 45:01 and click and you don't hear. These are 45:03 like natural language queries ripe for 45:05 amazing amazing targeting. That's one 45:07 way of making money. But you have to own 45:08 a firstparty product. You have to be the 45:11 first party. Second, you have to drive 45:13 outcomes. That's another way of making 45:15 money where you don't own any inventory 45:16 but you can drive outcomes for 45:18 advertisers. The best example of this is 45:20 a company called Apploving. Apploving is 45:22 a 100 plus billion dollar company. They 45:24 drive one outcome really well, mobile 45:27 app installs. And no one believed that 45:29 people would need that many mobile app 45:30 installs. Turns out everyone wants to 45:32 get mobile app installs. It was 45:34 initiated only restricted to gaming. But 45:37 now every mobile app where they sold one 45:38 mobile app installed. So app loving has 45:40 built a massive infrastructure. Now they 45:43 control the buy side, they control the 45:45 sell side, they even control the 45:46 middleware. You could argue that they 45:47 kind of control the auction for most 45:50 mobile apps in a way that almost Google 45:52 used to control or people say they 45:54 control for the web. But apploving has 45:56 built an amazing engine to deliver 45:58 mobile app installs at a certain cost. 46:00 >> So that's the other way. Second way to 46:02 do it. You deliver an outcome at a 46:04 certain cost. 46:06 The third way to do it is if you are the 46:09 exclusive 46:10 provider for a large advertiser or a 46:15 large source of demand where you become 46:17 a good example is a company called the 46:18 trade desk where Proctor and Gamble for 46:21 example go to the trade desk and say I 46:23 spend with Google I spend with Facebook 46:25 all my other display budget trade desk 46:27 here you go you can figure out how to 46:29 distribute it and how to run it and so 46:31 those are the three ways but you got to 46:32 be exclusive so Those are the three ways 46:35 that you can make money. 46:36 >> What business ideas don't work in in 46:39 advertising? Like what are the business 46:40 models that just are doomed to fail? 46:42 >> Trying to be a middleman on top of these 46:44 large platforms from my understanding 46:46 work trade desk I know doesn't work on 46:48 Google or Facebook at all. Doesn't work 46:49 with Google or Facebook as a first 46:51 party, but applovin I think only little 46:54 bit works on Google and Facebook. Mostly 46:55 they do their stuff on the on the 46:57 unwashed web basically outside. So, 46:59 you've got to stay out of Google and 47:01 Facebook's ecosystems because if you're 47:03 trying to build your business on top of 47:04 Google and Facebook or probably soon 47:07 OpenAI uh as an ad company, you're going 47:10 to get squeezed over and you every time 47:13 you build a new capability on top of 47:15 Google, turns out Google learns what 47:16 you're building 47:17 >> and Google has the best engineers on the 47:18 planet. So do Facebook, they will take 47:20 your capabilities, incorporate into 47:21 their platform. There's going to be 47:23 almost certainly a cottage industry of 47:24 companies that are going to come and 47:26 say, "I'm going to help you optimize ads 47:27 and chat GPD." There's already companies 47:29 that help you optimize placement in what 47:31 is called these answer engines called 47:33 AEO instead of SEO. All of those are not 47:36 going to create durable enduring 47:37 companies. 47:37 >> As your business grows, Vanta scales 47:39 with you, automating compliance and 47:40 giving you a single source of truth for 47:42 security and risk. Learn more at 47:44 vanta.com/invest. 47:46 Ridgeline is redefining asset management 47:48 technology as a true partner, not just a 47:50 software vendor. They've helped firms 5x 47:52 and scale, enabling faster growth, 47:54 smarter operations, and a competitive 47:56 edge. Visit ridgidelineapps.com 47:58 to see what they can unlock for your 47:59 firm. 48:00 >> What would you be worried about if you 48:01 were one of these fairly monopolistic 48:03 owners of a massive ad network like the 48:05 ones we've discussed? 48:07 There Uber and Amazon in the mix, Door 48:09 Dash, Facebook, Google. If you were 48:12 there running their ads businesses, what 48:14 would scare you? 48:15 >> Consumer behavior change. Consumer 48:17 behavior change where they don't open up 48:19 the apps anymore, but they use agentic 48:22 interfaces. They use AI interfaces which 48:24 are not owned by my company, this 48:26 company to to do their transactions. If 48:29 you assume that a big percentage of 48:31 things are repeat, then could you put 48:33 those repeat things on autopilot through 48:35 an agent and you never open the app and 48:37 so you lose opportunities to then 48:39 advertise and so and you lose the 48:40 relation with the customer over time 48:42 because the customers start trusting the 48:44 AI agent. You can't bury your head in 48:46 the sand. You have to go and experiment. 48:48 That's why when Chad GPD opened up their 48:50 apps platform, all of the commerce 48:53 platforms are experimenting. And the 48:55 thing I would look for very carefully is 48:57 there are going to be early adopters 48:58 using the app. I'm going to look for 49:00 obviously they're going to connect their 49:01 account, their Uber account with the 49:03 chat GBD account. I'm going to look to 49:06 see these people who are connected. 49:07 How's their behavior on my app? Are they 49:09 going to my app or not? 49:11 >> Are they opening my app or not? Are they 49:12 opening my app much less frequently? 49:14 Because if that's the case, then 49:16 obviously this experience is so 49:18 compelling that I would then have an 49:20 choice to make. How do I make this 49:23 experience maybe not as compelling as my 49:24 app experience or how do I incentivize 49:26 them here to open up my app? 49:28 >> There's a new battle happening for that 49:30 first category which is a new interface 49:32 to be owned. We know chat GBT is sort of 49:34 does I'm curious if you think being the 49:36 first mover matters to build a new ad 49:38 network because there's there's Gemini, 49:40 there's Anthropic, there's a bunch of 49:41 people that have tons of users using 49:43 this new interface. How do you think 49:46 about the landscape of the new potential 49:47 entrance to build the next dominant uh 49:51 at network? What advice would you give 49:52 these various parties? 49:54 >> The good news is the being first doesn't 49:56 matter. Why? because you control 49:58 especially if you're in category one 50:00 which we described you control your 50:01 first party inventory in fact being 50:03 second or third you can learn from the 50:05 iterations and mistakes that the first 50:06 one makes your inventory is not going 50:08 anywhere now some might have more 50:10 urgency to monetize than others but 50:12 Gemini doesn't need to monetize anytime 50:14 soon so they can just sit back they have 50:16 a lot of ads expertise and data from 50:18 Google they can sit back and wait till 50:20 they need to monetize uh in fact a good 50:23 strategic move for them might be to say 50:24 I am I am the zero ad platform like 50:27 Apple claims or Google can claim that 50:28 Gemini has no ads in it and there is a 50:30 certain set of customers or consumers 50:32 who care about that. But the biggest 50:33 thing is I think and and OpenAI has done 50:35 a good job of articulating this ads 50:37 should not ads and content ads should 50:40 not influence the content that is served 50:42 to me or the recommendations that AI 50:44 gives to me. I think they should be they 50:46 should be relevant and but they should 50:48 not be um influencing the the 50:50 recommendations. And second, you have to 50:53 keep a high bar for engagement and 50:55 usefulness. Unfortunately, however 50:57 relevant ads are, the reality is that 51:00 this wasn't proven is that once you 51:01 start showing ads in an previously 51:03 unmonetized uh no zero ads uh surface, 51:07 engage engagement of users goes down 51:09 over time. It does because some of the 51:12 engagement gets siphoned off by ads and 51:14 some of it gets siphoned off in 51:15 different ways. But this many hold out 51:16 groups across many companies have proven 51:18 this. So the question for any one of 51:21 these companies is how much engagement 51:23 are we willing to take in exchange for 51:25 monetization. And so I think first you 51:28 need to have a hold out group. I'm sure 51:29 they're having it a hold out group of 51:30 people who never ever see any ads 51:33 because that's your fresh group that 51:34 never sees ads and you need to 51:36 understand that's their behavior. And 51:37 then you need to always understand how 51:39 uh how you know people with ads are 51:41 behaving and then you need to figure out 51:44 uh what the engagement hit is from each 51:46 quantum of ads and you need to then give 51:48 your ads team a certain engagement 51:50 budget and so that's what at at Facebook 51:53 there was an engagement budget every 51:54 year that between the newsfeed team and 51:56 the ads team we had to adhere to. In 51:58 other words, uh we yes we wanted this 52:01 much revenue but it the the check metric 52:03 on the revenue was we can't take more 52:05 than x% dip in engagement overall for 52:08 newsfeed. 52:09 >> What are the attributes of a good 52:10 northstar metric? Like what advice would 52:12 you give someone that's trying to pick 52:13 the thing around which the company is 52:15 going to optimize? 52:16 >> Yeah, the northstar metric is a is a 52:18 metric that is an indicator of company 52:21 growth and customer value. So it 52:24 actually balances customer value and 52:25 business value nicely. Nostra metrics in 52:28 my opinion should not be revenue. It 52:30 should be something that is directly 52:32 correlated with customer value. So for 52:33 example, if customers are doing well, 52:35 the Nostra metric should go up and to 52:37 the right, but it should also lead in 52:39 business the business doing well. For 52:41 example, for Square, the Nostar metric 52:43 was GPV, which is volume of payments 52:46 processed. It was not correlated to re 52:48 it was somewhat correlated to revenue, 52:49 but it most importantly showed that the 52:52 number of the amount of payment 52:53 processed to the company was continuing 52:55 to grow. At Facebook, the northstar 52:57 metric was DA us. It was actually 52:59 monthly active users. Then it over time 53:01 went to daily active users because it 53:03 was a sense it was an indication of how 53:05 engaged C users were. Now, one of the 53:08 most important things about an NSM is 53:10 that it needs to be coupled with what we 53:12 call check metrics. In other words, 53:14 NSTAR metrics if they're left alone can 53:17 as you know incentives drive behavior. 53:20 So if you tell a team go and optimize 53:21 this Nstra metric they will do what it 53:24 is going to go up 100%. But then many 53:26 things that you don't want to go down 53:28 could go down. So for example in in the 53:30 Door Dash case you could say I want to 53:32 grow GMV which is the gross 53:33 merchandising value which is the 53:34 Nordstar metric. Now GMV is the total 53:36 order of total value of all the orders 53:38 that go through the marketplace. I could 53:41 make it grow up by offering by setting 53:42 delivery fee to zero by setting 53:44 everything to zero and what happens 53:46 then? The company's revenue goes to 53:47 zero. So you basically want a check 53:50 metric that is maybe a check metric 53:52 around the health of the customer and a 53:54 check metric around the health of the 53:55 company that basically hold this that 53:57 are the guard rails around this Nstra 53:59 metric. So in the case of Door Dash it 54:02 might be I want to maintain a certain 54:03 gross margin percentage or I want to 54:05 maintain a certain customer retention 54:06 percentage something like that. Again it 54:08 might be margin is typically a good one 54:10 to use because in some ways that is a 54:13 indicator of the company health. 54:14 >> There's these two ideas that we talked 54:16 about when we first met. One was the 54:17 need the need for the very best software 54:19 companies to sort of stand alone in the 54:21 sense that someone can just go use it 54:23 without talking to a human and it just 54:25 it works for their problem. So like 54:26 fully fully self-s serve. Love to hear 54:28 you talk about that. And a related idea 54:30 was that's sort of on the builder side. 54:32 On the investor side, you mentioned to 54:34 me that all the great investments that 54:36 you've had, the companies that have 54:37 really had like explosive growth have 54:39 had a high number of one of four 54:41 qualities which is I think was gross 54:43 margins, low cost to acquire the 54:45 customer. um high retention and a tight 54:48 sales cycle which maybe match maps back 54:49 onto the self-s serve thing. 54:50 >> Yeah. 54:51 >> So talk about the relationship between 54:53 those two things. 54:54 >> The self-s serve notion actually came 54:56 from Google was Google was the first 54:59 company I worked at which achieved 55:01 massive scale and what happened at 55:03 Google was within the ads team. Uh we 55:05 basically had wide number of customers 55:08 using us the millions of customers using 55:09 us. There were a lot of small businesses 55:10 but there were also large companies. 55:12 what we ended up doing to serve the 55:14 large companies. Large companies didn't 55:15 want to use the product themselves. They 55:17 had uh agencies using it for them on 55:19 their behalf and they also had internal 55:21 people at Google support and sales and 55:23 operations people using them. So on the 55:26 product side we built a lot of tools for 55:28 this internal for our internal 55:30 colleagues for our sales and operations 55:31 colleagues to manage the system for our 55:34 large customers. One day I think we were 55:36 at a Larry review and we were showing 55:37 these what we called ICS internal 55:39 customer systems to Larry. I think we we 55:41 were not meaning to show it but I think 55:43 to show him a demo we somehow got into 55:45 it. He was like what is that? We're like 55:47 well it's a system used by our internal 55:48 teams. He's like why'd you build it? We 55:50 were like well we have to help our large 55:52 customers. He said I don't want you mean 55:55 our small customers don't have access to 55:56 it. We're like no end it right now. We 55:59 were like what do you mean end it? I 56:01 want to make sure that everything you're 56:03 building for large customers is also 56:05 available to small customers. And so we 56:08 basically had to take everything we had 56:11 built over years for in this IC system 56:14 and made it make it available to 56:16 customers. And turns out an interesting 56:17 thing happened. Turns out the smaller 56:19 customers adopted it much faster because 56:22 some of these things we're building had 56:23 advanced knobs and so on that we didn't 56:25 think they would use. Turns out the 56:27 self-s served customers were the most 56:29 sophisticated users because if you do 56:32 something that's interesting, there's 56:34 all these small agencies, entrepreneurs, 56:37 hustlers, all of these folks, they if 56:40 you can help them make more money, the 56:42 it's it's a testament to human 56:44 creativity and ability, they exploit the 56:46 system in ways that you never you never 56:48 even know and you learn a lot from 56:50 working with them. So I've seen in every 56:52 case when you open up your system to 56:54 selfs serve you learn so much more about 56:56 the capability of your product than if 56:58 you basically it's your sales team doing 57:00 it on their behalf. In fact, I'll never 57:02 forget in AdSense, I think we had some 57:04 of the largest publishers in the world 57:06 sign up and start using us on a self-s 57:09 served basis and then we engage with 57:11 them after that. And I think companies 57:13 like Atlacian, Square, I think we had 57:15 Nike, I think start signed up for a uh 57:18 for a for a Square uh device and and 57:21 sell some onboarded and start using in 57:23 one of their stores. We had I think 57:24 Whole Foods. So, I think it just changes 57:27 it does two things. One, it makes your 57:29 product better. It makes your product 57:31 better because these folks they use the 57:34 product in ways that you don't expect or 57:36 anticipate and it helps you it forces 57:39 you because what is the definition of 57:41 self-s serve? The definition of self-s 57:42 serve is the customer can onboard not 57:46 just use but onboard and use the product 57:48 without ever talking to or engaging with 57:50 a single member of the employee base at 57:52 the company. So when you do that that 57:53 means you have to think about how do 57:55 they actually get set up with the 57:56 product. So it really puts a lot of 57:58 effort on onboarding because onboarding 58:00 is one of those things where most people 58:01 drop off if you don't do a good job and 58:03 then you've got to get them to a moment 58:04 of delight very quickly. All of those 58:06 things a large if you're not building a 58:08 sales product you don't even think about 58:10 and a seller product you think about it 58:11 every day. It's like a consumer product 58:12 or sells a business product. And then 58:15 second what it does for you is it opens 58:17 up the aperture to your customers 58:19 because with say a 100 salespeople yeah 58:21 you can reach maybe 10,000 customers but 58:24 with a self-so product with the right 58:25 word of mouth you can reach millions of 58:27 customers look at cursor for example it 58:30 is used in every large company I bet 58:32 only maybe 1% of companies is maybe the 58:34 top down motion 99.9% of companies some 58:37 engineer got it great example is a 58:40 company is Figma actually after I 58:42 invested in Figma I joined square one 58:43 and a half years later I tried to 58:45 basically push Figma down top down into 58:47 the design team because learning design 58:49 I said you got to use Figma designers 58:51 refused to use it they're using a tool 58:53 called sketch and they said we're not 58:54 going to use it sketch is much better 58:56 and so I felt okay it's not my place to 58:58 tell them what tools to use so I backed 58:59 off two years later a mid-level design 59:02 manager came in and they brought in 59:04 Figma from their prior company and they 59:06 basically got got it to be used across 59:08 and it kicked out sketch so I think with 59:11 self-s serve you can get into these 59:12 things where even there's an incumbent 59:14 But you can infiltrate and be an 59:15 insurgent in a unique and powerful way 59:17 which a sales direct sales motion could 59:19 never have produced then. 59:20 >> One of the other dimensions that's 59:22 changing fast is careers. I'm curious 59:24 what you think about the sorts of people 59:26 that will thrive best in this new era. 59:29 If you're a person hiring someone, what 59:31 are the sorts of things that you would 59:33 place extra emphasis on now in the sort 59:36 of AI era? 59:36 >> The number one thing I think is going to 59:38 be the focus on doing and building. I 59:40 think CEOs have gotten too comfortable 59:42 over time and I think this is changing 59:44 hiring middle management very very 59:46 quickly and hiring sea level people 59:48 instead I think you're going to see the 59:50 rise of AI agents doing a lot of work 59:52 but then humans who manage the AI agents 59:54 and our IC's so I think what the number 59:57 one skill that is going to be relevant 60:00 two years from now probably even one 60:01 year from now is to become a functional 60:04 expert that knows how to build AI agents 60:06 to do that function and orchestrate an 60:09 army of AI agents to do that function. 60:11 Well, there was a great article the 60:13 other day I read about an PM at Meta 60:16 who's non-technical but who basically 60:18 built a bunch of AI agents to do his job 60:20 as a PM so well that even his engineers 60:23 like teach me how to use a AI agents 60:25 well. And so I think that's what you 60:26 want. You want somebody who is 60:29 essentially acting as a manager but not 60:31 of humans but of AI agents. And 60:33 management has to be a full-time job. 60:35 What I mean by that is if you manage 60:37 three, five, 10 people that's not 60:38 enough. You either need to be managing 60:40 50 humans or you need to be an IC. And 60:43 so there would there's something called 60:45 span of control which means how many 60:47 people you manage in some ways. And so 60:49 span of control less than 10 should not 60:51 be allowed at any company at this point. 60:53 I think everyone should have I think a 60:55 full-time because think about it if 60:56 you're managing even 15 people maybe you 60:59 meet with them once a week that's 15 61:01 hours. What are you doing for the other 61:02 25, 30, 40 hours? 61:04 >> You should be working. And on the 61:05 company side, don't hire managers as 61:08 long as possible. Hire doers. Hire 61:11 builders. 61:12 >> How do you what is your favorite way to 61:14 assess whether or not someone is is that 61:17 in interviewing them or learning about 61:18 them? 61:19 >> Best way is to give them a work project. 61:21 Engineering does a great job. 61:22 Engineering has always done a great job. 61:23 Every company I've been at, they would 61:25 have engineering coding interviews, 61:26 programming interviews. 61:27 >> Yeah. Do stuff. Everywhere else you can 61:30 just BS your way without doing stuff. 61:31 You can just talk and talk is not you 61:34 got to actually do stuff. Produce an art 61:36 artifact. So at square we established 61:38 work projects where even for corp dev I 61:40 remember corpdeev our our work project 61:42 was tell me about give me one company 61:44 that square should buy and analyze the 61:47 company and tell us why we should buy it 61:48 and tell us what the synergy should be. 61:49 So the best candidates had to do that. 61:51 So every every function needs to have a 61:53 work project that you need to put them 61:55 in a room without AI and get them to do 61:57 the project. Get them to do the work 61:58 that is ideally very similar to the work 62:01 they're going to do. I we would almost 62:03 give them for product managers we would 62:05 take a product we were thinking about 62:06 and we would just say here's a product 62:08 we're thinking about figure it out 62:10 should we build it. The first and most 62:12 important thing you want for these kind 62:13 of thing is especially for customer 62:15 facing roles they need to take the voice 62:17 of the customer. In other words, they 62:18 need to justify the why. The best PM 62:20 candidates rejected the premise 62:22 completely 62:23 >> and they did it in a beautiful way. They 62:25 went and talked to 10 customers on the 62:26 street. It's so brilliant. They said, "I 62:28 talked to 10 customers. I they were all 62:30 square users, which is so easy. Mint 62:32 plaza, you go there and we found that 62:34 none of them want a pre want this 62:35 premium insights product. So, we don't 62:37 build it. We're going to build this 62:37 other thing." And said it was amazing. 62:40 That's what you want to see. You want 62:41 agency. You don't want people to just 62:43 say, "Give me what to do and I'll do You 62:46 want people to reject the premise or 62:47 question the premise in the first place. 62:50 Square should not buy a company. That 62:52 would be great. Why? Tell me why. And so 62:53 that's the kind of thinking you're 62:55 looking for. 62:55 >> What was Tony's thing? 62:56 >> Tony's thing was he would give people 62:58 either $10 or $20 and ask them to 63:01 acquire a,000 customers. A,000 customers 63:04 for Door Dash consumers. And literally 63:06 some people would say, "I'm not going to 63:08 take this challenge. I'm not ready for 63:10 it or something." and great that if you 63:12 literally opt out of it and then some 63:13 people would take it and the goal nobody 63:15 even came close to acquiring a thousand 63:17 or even 100 I think but the goal was to 63:19 see how many different things they were 63:21 able to try in the course of few hours 63:23 someone went to the gym printed flyers 63:25 out and gave it out people tried all 63:27 kinds of things it it was a brilliant 63:28 way to just just filter out people who 63:32 didn't want to do stuff 63:33 >> is there any other advice that you would 63:34 give the person building the career we 63:38 talked about you know evaluating and and 63:40 uh be a builder and all these sorts of 63:42 things. How should one think about 63:43 managing a career in the AI era? 63:46 >> Stay at every job long enough to have 63:49 impact. I have over the last 18 24 63:53 months I've been seeing this phenomenon 63:54 of job hoppers or job optimizers I call 63:57 them who stay at a job for 12 to 18 63:59 months and then they move to the next 64:00 job and then they say 12 to 18 months 64:02 and move to the next job. I think that 64:04 that is one of the biggest red flags as 64:06 a hiring manager that I see because I 64:09 don't think you can achieve anything of 64:11 value. You can't have any impact on a 64:13 company in 12 to 18 months. I think it 64:15 takes minimum 3 to four years to have 64:18 impact on a company. So my top advice is 64:21 stay long enough to have an impact, 64:23 build a network, have fun. Don't from 64:26 the moment you start a job, don't be 64:28 thinking about what my next job is. once 64:30 in a while maybe one job it didn't work 64:32 out uh amongst a series of jobs okay you 64:35 left it 18 months but if I'm seeing two 64:37 or three jobs back to back immediate red 64:39 flag I posted this on X and tons of 64:42 managers wrote to me saying it's an 64:44 immediate red flag so you do a massive 64:46 disservice and you won't even know the 64:48 problem is you'll get rejected you won't 64:50 know what happened it's that people want 64:51 people who stick around and build who's 64:53 going to hire you if they see that's 64:54 your behavior so I think it's a very 64:57 shortserving or it's it's a very 64:59 short-term thinking you got to build 65:01 something of value and that comes with 65:03 time. 65:03 >> So much of the theme here has been uh 65:05 identifying a superpower, having one in 65:07 the first place, evaluating one, 65:08 matching it to a problem with a leader 65:11 and so on with your investor hat on and 65:13 your new firm marathon. How do you 65:15 assess the capacity or existence of a 65:18 superpower in a person? Like what what 65:20 how have you learned to do that? Well, 65:22 >> the most important thing I look for is 65:23 founder authenticity. If you think about 65:25 it, three of the four companies I worked 65:26 with, Google, Facebook and and Door 65:29 Dash, all started in school, all started 65:32 in colleges and they all started as a 65:34 way to just a toy problem almost that 65:37 that the founders are curious about and 65:39 they started with an authentic 65:40 curiosity. Can this be built and then it 65:42 became it got built and it started and 65:44 similarly with Jack and Jim, they 65:45 started solving a real problem. So my 65:47 first question to every founder is tell 65:49 me your founding story. Why did you 65:51 decide to start this company? And so the 65:54 founding story in my opinion is what a 65:56 lot of it expresses why they chose this 66:00 problem and ideally it should touch on 66:02 what the superpower is and what 66:03 compelled them to work on this problem. 66:05 I really I I've had many people work 66:08 with me or for me who have gone out to 66:09 start companies with the only reason 66:11 being well I have my buddy and we both 66:13 want to start a company together. I 66:14 really advise them not to do that 66:16 because just going out and starting a 66:17 company because you want to start a 66:18 company with your friend is the wrong 66:20 reason. So I want to understand is there 66:22 an authentic lived experience that 66:24 they've had in their life that compels 66:26 them to work on this product. Dylan uh 66:28 from Figma, if you talk to him, he's 66:31 seeped in design. He thinks about the 66:33 design of things. He thinks about how to 66:35 make things more compelling and it was 66:37 very clear that he had a vision for what 66:40 this thing would be. And a good example 66:41 is a company called Fair. It's a B2B 66:43 marketplace. Max RHS the CEO worked for 66:46 me at Square. And Max when he left 66:48 Square he actually tried many different 66:50 ideas and turns out and none of them 66:52 were authentic to him and to fair. Turns 66:54 out the idea that worked was fair. Why? 66:57 Because when he was a undergrad student 66:59 he had an umbrella company that he 67:02 created and this umbrella company he was 67:04 trying to get distribution for it in 67:06 local retail and it was extremely hard 67:09 for a brand. How do you get local 67:11 retail? There's so many of them. How do 67:12 you go in and pitch to them? So he 67:13 realized that that problem is the one he 67:15 wanted to focus on other manufacturers 67:18 who wanted to get access to local 67:19 retail. 67:20 >> Are there any other questions that you 67:22 love to ask in a first meeting learning 67:24 about a company other than tell me your 67:25 origin story? 67:26 >> I think the other one is idea maze. Tell 67:27 me about how you navigated the idea 67:29 maze. Yes, you want to tackle this 67:30 problem because again this is a classic 67:32 product thing. You start with the 67:34 problem but then there are many 67:35 different solutions, many different ways 67:36 to solve it. Why did you choose this 67:38 solution? Why did you choose this way 67:39 versus the other way? So I will 67:41 basically throw try to throw them off 67:43 course or offkilter by asking them five 67:46 six other ways to solve the same problem 67:48 and ask understand if they are if they 67:51 are students of either history or their 67:52 industry to say why this problem why 67:55 this problem could not be better tackled 67:56 in this way. So I want to understand 67:59 that they have studied alternate 68:01 approaches historical approach to solve 68:03 this problem. I think good example is 68:04 the Collisons I think bought a book on 68:06 payments and they studied exactly why 68:08 all the payments companies did what they 68:10 did and how they failed and how they 68:11 succeeded and I think the best founders 68:13 are students of history in that industry 68:17 and they understand why all the prior 68:19 companies took the decision and ideally 68:21 they stand on the shoulders of giants 68:23 they're able to build this company the 68:24 other thing I always recommend to CEOs 68:25 is a board role is like a marriage uh 68:28 once you get into it it's very hard to 68:29 get out of so never ever ever invite 68:33 anyone to join your board before 68:34 spending at least a year with them. 68:36 >> Have them join an advisory board. Have 68:39 them meet with everybody in the on the 68:41 management team. Spend time with them. 68:43 Have them come to a few board meetings. 68:45 Have them meet with the other board 68:46 members. Come to a board dinner. But 68:47 don't and have three or four people in 68:50 your advisory board and then make one of 68:52 them a board member. If you like them, 68:53 if you feel they're adding value, if 68:55 your team feels they're adding value, 68:56 etc. The other thing I've seen with 68:58 boards over the last 15 years is the 69:01 management team getting involved. 15 69:03 years ago, it would just be the CEO, the 69:05 co-founder maybe, and the board. We'd 69:06 meet for four or five hours, discuss 69:08 topics, maybe bring in the management 69:10 team person for one slice, the CFO, and 69:12 then they would leave. Now, most 69:14 companies, they have the management team 69:16 attend the entire board meeting, entire 69:19 board meeting except for what is called 69:20 the executive session. And I think that 69:22 is awesome. Why? Because I think 69:24 management team and board get to meet 69:26 each other. As part of a board, you want 69:29 to understand who's on the management 69:30 team. Who could be successor to the CEO? 69:33 What are the capabilities of different 69:34 parts of the management team? And then 69:35 as the management team, you want the 69:37 management team to be able to leverage 69:39 the board for help. I think one of the 69:41 best practices I've seen done and I've 69:44 I've now tried to push other companies 69:45 to do it is a notion of a board buddy. 69:47 So everyone on the board should become a 69:49 buddy to manage team member and and they 69:52 would then meet with that managing team 69:53 member uh multiple times between board 69:56 meetings. So once a month or even text 69:58 with them and anything they're almost 70:00 like a sounding board anything the 70:02 management member has. So that you can 70:03 see that the different board personas I 70:05 described they map nicely. So I 70:07 generally am the management the buddy 70:09 for the head of product or the head of 70:11 engineering. Somebody else is a buddy to 70:13 the CFO someone else is the head buddy 70:14 to the CRO etc etc. So it's a I think 70:17 the meetings in between the board 70:20 meetings are actually just as important 70:21 as a board meeting themselves because 70:22 that's when you are cuz a board meeting 70:24 can there's a lot of things going on. 70:26 Yeah. And so but but those relationship 70:29 that's the other thing I realized it's 70:30 not the board meeting that truly 70:31 matters. It's all the things between the 70:33 board meetings that that are the real 70:35 real thing when things get done. I think 70:37 the only thing we haven't talked about 70:38 in this like grand art of company 70:41 building and and and product creation is 70:44 the the job of acquiring the customer, 70:47 positioning the product, marketing, the 70:49 way it sort of presents itself to the 70:51 outside world. What's the dispatch from 70:53 like the cutting edge that you're seeing 70:55 of how people do this? All these things, 70:59 position, brand, customer acquisition, 71:02 the ways they do that. What does like 71:04 new excellence look like to you across 71:06 this the many many companies that you 71:08 get to see? 71:09 >> One of the most interesting things now 71:10 it's different between enterprise 71:11 focused and consumer focus. For consumer 71:13 focused companies the big thing is how 71:15 to scale influencers. I think 71:17 influencers have become much much much 71:19 more every year they become much more 71:21 powerful in how people especially 71:23 younger people consume products and and 71:26 even choose products. Somebody said that 71:27 Tik Tok is the best local search engine 71:29 and I think that's right. My kids have 71:31 discovered crazy when you go traveling, 71:33 crazy restaurants on Tik Tok that Google 71:35 Maps would not really show or Yelp 71:36 doesn't show, etc. So, how do you reach 71:38 influencers on Tik Tok? And there's a 71:40 set of companies that's coming out 71:42 that's essentially making it easy. The 71:44 problem is influencers on Tik Tok 71:46 obviously there's head influencers, but 71:48 there's a long tail that go viral for 71:50 different reasons and you want to 71:51 capitalize on those viral waves if 71:53 possible. So there is a set of companies 71:55 that is building products to see if they 71:57 can help brands connect with these 72:00 influencers in scalable ways. On the 72:03 enterprise side, I think the most 72:05 interesting thing I'm seeing it's not 72:06 really a um acquisition channel as much 72:10 as it is a u onboarding channel. It is 72:13 basically presenting an outcome to 72:16 customer and saying let's collaborate on 72:18 outcomes. Palunteer does that very well. 72:20 Palunteer goes to customers and say 72:21 what's your most important business 72:23 problem? Oh, here it is. Okay, great. 72:25 Give us 6 months to solve it. Engage 72:27 with us. If we can't solve it, fire us. 72:28 Don't pay us anything. If we solve it, 72:30 pay us a lot of money. So, it's truly 72:32 taking ownership. And I think this goes 72:34 to outcome based pricing. How your 72:36 product is priced and your confidence in 72:38 your ability to deliver that outcome of 72:39 course. Um, so I think outcomebased 72:42 selling is I think one of the most 72:44 interesting ways of changing. And in 72:46 fact, I I've one of the top piece of 72:48 advice I have for founders reaching out 72:50 to companies is you cannot lead with 72:52 what your product does anymore. You've 72:54 got to lead with what is the outcome you 72:56 can deliver or ideally even have 72:57 delivered. I'll never forget this uh 73:00 this example and what is crazy is that 73:02 companies always look to other companies 73:04 in the vertical. This never will change. 73:06 So for example, if you get JP Morgan to 73:08 use your product, I promise you every 73:11 single bank will then evaluate your 73:12 product. But if you get Proctor and 73:14 Gamble, JP Morgan doesn't care if Prot 73:16 and Gamble use your product. So even 73:17 when you go to market, you've got to 73:19 target instead of trying to be too 73:21 horizontal unless it's bottoms up. On a 73:23 sales side, you've got to try to go 73:24 after one or two very specific verticals 73:26 because there is a very clear lighthouse 73:28 effect. You want to go after the best 73:29 one and get the best one and then you 73:32 basically win all the other ones in your 73:33 in that vertical. 73:34 >> I think you might know my traditional 73:35 closing question uh that I ask 73:37 everybody. What is the kindest thing 73:38 that anyone's ever done for you? 73:40 >> There are so many. I think uh the the 73:42 best one is a guy called Bob McDonald. I 73:44 was basically a student uh a business 73:46 school student on the east coast. I 73:48 really wanted to uh get a job. I was in 73:51 a visa. I wanted to get a job in Silicon 73:53 Valley. I was somewhat unqualified. I 73:55 was I'd never been a product manager 73:56 before. I'd been an engineer and never 73:58 worked in photonics optical networking 74:00 before. And Bob basically saw a spark in 74:03 me and said, "You know what? I'm going 74:04 to make a bet on you and I'm going to 74:06 hire you and I'm going to bring you to 74:07 Silicon Valley and you're going to be he 74:09 was a Sequoia funed company, one of the 74:11 hottest companies in the valley. He 74:12 could have had any pick of anyone but he 74:14 bet on me. So I basically have taken 74:17 this approach that I try to pay it 74:19 forward and I have no expectation when I 74:21 do something for someone. 74:22 >> What created the spark in you? 74:24 >> Like what about your life? Where did the 74:27 spark come from? Uh for me it's all 74:29 about just knowing how fortunate I am to 74:31 be healthy. Uh to have a family that 74:34 loves me and to know that in almost 74:37 every every run of the simulation I 74:39 could be in one a million different 74:42 worst circumstances that I am today. And 74:45 so just gratefulness and gratitude about 74:48 where I'm sitting. I mean we are sitting 74:49 in literally the top 1% of the 1% of the 74:52 1% situations right now and breathing. 74:56 And so literally I think I feel pain 74:59 when I see somebody suffering and I see 75:01 as they say there for the grace of God 75:03 go I in some ways and but for the grace 75:06 of God and you basically realize that 75:08 you're very lucky to be given this one 75:09 life and you have a responsibility to 75:12 the world and yourself to be grateful 75:14 and to to lead the best life you can. 75:16 >> Koko, this was incredibly fun. Thank you 75:18 so much for your time Patrick. 75:20 >> Thank you. Thank you my friend. 75:26 You know how small advantages compound 75:28 over time? That's true in investing and 75:30 just as true in how you run your 75:31 company. Your spending system is your 75:33 capital allocation strategy. Ramp makes 75:35 it smarter by default. Better data, 75:37 better decisions, better economics over 75:38 time. See how at ramp.com/invest. 75:41 As your business grows, Vant scales with 75:43 you, automating compliance and giving 75:45 you a single source of truth for 75:46 security and risk. Learn more at 75:48 vanta.com/invest. 75:50 Every investment firm is unique and 75:51 generic AI doesn't understand your 75:53 process. Rogo does. It's an AI platform 75:56 built specifically for Wall Street 75:57 connected to your data, understanding 75:59 your process and producing real outputs. 76:01 Check them out at rogo.ai/invest. 76:03 The best AI and software companies from 76:05 OpenAI to Cursor to Perplexity. Use Work 76:07 OS to become enterprise ready overnight, 76:09 not in months. Visit works.com to skip 76:12 the unglamorous infrastructure work and 76:14 focus on your product. Ridgeline is 76:16 redefining asset management technology 76:18 as a true partner, not just a software 76:20 vendor. They've helped firms 5x in 76:22 scale, enabling faster growth, smarter 76:24 operations, and a competitive edge. 76:26 Visit ridgelineapps.com to see what they 76:28 can unlock for your firm.

Key Takeaways

Product Development Is Fundamentally Different Now

Judgment: The Only Future-Proof Skill

Software Durability in the AI Age

Four Sources of Stickiness

Building Ads Businesses — Three Pillars

North Star Metrics + Check Metrics

Leadership Lessons

Self-Serve Is Non-Negotiable

Hiring & Career

Board Management

CEO Communication Format

Go-to-Market

Timestamps

Relevance to SupportWire & FeatureOS

Connected (32)