Alex Hormozi x Amjad Masad: Building in the Age of Agents

Replit

16.9K views • October 14, 2025

**Alex Hormozi x Amjad Masad: Building in the Age of AI Agents**
Alex Hormozi sits down with Replit CEO Amjad Masad for an unfiltered conversation about business fundamentals, mental resilience, and building software in the age of AI. From CAC/LTV economics to autonomous coding agents, this is a masterclass in first-principles thinking for founders.

**What You'll Learn:**
- Mental toughness frameworks: the difference between fortitude and resilience
- The $100M money model: how to eliminate burn rate and build capital-efficient businesses
- CAC optimization strategies that drove Hormozi's gym empire
- Why vertical SaaS is facing an existential crisis
- How AI agents are transforming software creation
- The real breakthrough in AI: from information systems to action systems
- Why Replit Agent 3 changes everything for non-technical founders

**Key Topics Covered:**

**Business Fundamentals**
- Why there are no rules for entrepreneurs
- The only CAC/LTV math that matters: get customers to pay for the next customer
- Drive CAC as low as possible, drive LTV so high you can outspend everyone
- Case study: How float tanks achieved negative customer acquisition costs
- The counterintuitive power of adding friction to increase conversion
- The "down sell" strategy that crushed churn and created the 2nd highest LTV customers

**Mental Frameworks**
- Mental toughness vs mental fortitude vs resilience: the 3-part model
- Emotional regulation without behavioral change

**AI & The Future**
- Why AI learns the same way humans do: reinforcement learning explained
- From information systems to action systems: the next paradigm
- Why reinforcement learning from the real world creates good agents

**Replit Agent Deep Dive**
- Hormozi's 30 hours building with Replit: from fragile to production-ready
- Why other AI coding tools are toys compared to complete feature sets
- Real customer stories: $5M VC CFO got to revenue, $200K in 6 months

**Industry Shifts**
- The innovators dilemma: AI companies born and disrupted rapidly
- Multi-agent future: specialized sub-agents working together

**Live Funnel Audit**
- Hormozi analyzes acquisition strategies in real-time
- Moving customers through the funnel: the watermelon analogy
- Scaling from initial revenue to $1M+ in profitable ad spend

**About the Guests:**
Alex Hormozi is a serial entrepreneur and investor known for scaling Gym Launch to $100M+ and author of "$100M Offers" and "$100M Leads."

Amjad Masad is the founder and CEO of Replit, building the platform that lets anyone create production-ready software using natural language.

Login or create an account to generate AI summaries

🔑 Login 👤 Sign Up

0:00 I'm gonna give you five different things

0:01 to do today, but the one that could

0:02 triple your business is going to take

0:03 five seconds.

0:04 I wonder if we can put Chromosi bot in a

0:07 funnel. See, that would be a very

0:09 valuable thing if you had a bot that

0:11 could crawl a funnel

0:12 and then make suggestions. It takes a

0:14 picture of each page.

0:15 Yeah, exactly.

0:16 And then just doodles. Yeah. And then

0:17 just says like, I think you have an

0:18 offer issue here. And it could crawl

0:20 your ads library page and the hooks are

0:22 off here or the hook isn't congruent

0:24 with the headline. There's all these

0:25 different check marks that you have to

0:26 go through. I wonder if we collaborate

0:28 with you to have like a funnel scan.

0:31 Yeah, that'd be cool. It' be an

0:32 incredible lead magnet.

0:33 Yeah,

0:36 there's so much content out there about

0:38 entrepreneurship and that's often good.

0:41 But often people think that they need to

0:42 fit a certain pattern and they miss the

0:46 things that are around them. Family

0:47 members that can help them. Uh

0:49 co-founders that might not fit the bill

0:51 of a

0:52 of of a co-founder you might expect.

0:54 Yeah.

0:54 Um and um

0:56 well, it's usually the weird ones that

0:58 win.

0:59 It is it is it is a it is a tails game,

1:02 right? Like it is not an average. This

1:05 is not the average game.

1:07 Um and so what I like to tell founders

1:09 and entrepreneurs is that there are no

1:11 rules in this game.

1:13 There are no rules.

1:13 You know, my uh hold on, I'll show you

1:16 something.

1:19 This is the first book.

1:23 There are no rules. Amazing. Amazing.

1:26 It's my only guiding principle for the

1:27 book. There are no rules.

1:28 By the way, I loved about this this book

1:31 when you start with here are some things

1:33 that people said about Alex. That

1:35 cracked me up.

1:35 Okay.

1:36 I I like how even when you started I I I

1:40 like I started watching your videos, you

1:42 know, when you first came out.

1:42 Oh, no way.

1:43 Um

1:44 Well, thank you.

1:44 Yeah, of course. And No, thank you. I

1:46 mean, I learned a lot and um

1:49 you know what attracted me to it is

1:52 like, yeah, I don't have anything to I'm

1:54 not trying to sell you anything. you

1:56 know, I'm just talking and it's clear

1:58 you have

1:59 this demeanor about you that is kind of

2:02 relaxed and I don't know where where

2:03 that comes from because a lot of fanders

2:05 are just sort of so highrung and

2:07 honestly I think it's just I have a

2:09 worldview that most people don't share

2:10 and I think it bothers a lot of people

2:11 and so I don't talk about it as much

2:13 because I just um but when I do talk

2:17 about it like I always like I get the

2:18 most meaningful messages from talking

2:20 about it which is just like I believe in

2:21 cosmic irrelevance

2:23 um and so

2:25 you know like the idea that like someone

2:27 is is famous or not famous or has status

2:30 or doesn't have status. I I just so hard

2:32 whole wholeheartedly believe that in in

2:34 four generations like I will be

2:36 completely forgotten and no one will

2:37 care.

2:39 And I think that just like

2:41 it just takes a lot of the pizzazz out

2:43 of out of like

2:44 special snowflake, you know, isism.

2:47 Yeah.

2:47 Um and it takes a lot of the anxiety out

2:50 of the day-to-day operations of a

2:51 business. The upside and the downside.

2:53 And I would say that

2:55 I think a lot of people, this is gonna

2:57 be a little bit of a side quest, but

2:58 I'll wrap it around. So like

3:01 I think many people desire emotional or

3:04 mental stability and I think it's

3:05 absolutely a prerequisite for being a

3:06 good entrepreneur because there's just

3:08 so many fires that come every day,

3:09 right? And so I think people have

3:11 difficulty with that because they don't

3:12 define what it really means. It seems

3:13 like a very amorphous thing like

3:14 emotional stability or mental stability

3:16 like what does that mean from a behavior

3:17 perspective? And so I see that there's

3:19 three components to it. So you have um

3:22 you have uh mental toughness, you've got

3:24 mental resilience, and then you've got

3:25 mental fortitude. And I'll define each

3:26 of them. So toughness I see as your

3:28 fuse. How many bad things can occur in a

3:31 row without you changing your behavior?

3:33 That would be toughness.

3:34 The next is um I'll go for two seconds

3:38 probably better. Uh fortitude is um once

3:41 once you surpass your fuse of bad things

3:44 that can happen before your behavior

3:45 changes, how steep is that curve of your

3:48 behavior change? Is it a little drop or

3:50 is it a you go ballistic and need to go

3:52 you know do heroin for five months you

3:55 know what I mean like I don't know just

3:56 like something completely crazy right

3:58 um so that's second element of it and

4:00 then the third element would be

4:01 resilience which is like okay how

4:03 quickly do you bounce back

4:05 um so what is the time between uh

4:07 baseline behavior change in behavior and

4:10 then recurrent to the original baseline

4:12 and so if you think about those three

4:14 vectors it's to me it's it's helpful so

4:16 that I can think okay do I need to

4:18 exhibit bit more mental toughness right

4:20 now as in like cuz the ideal scene is

4:21 that just

4:22 you have maxed mental toughness and so

4:24 that nothing changes your behavior but

4:26 we are human we are fallible and things

4:27 do happen right and so if I am going to

4:31 you know have my behavior change then I

4:32 want to think okay I want it to be as as

4:34 little of a steepness as possible um in

4:36 terms of my change as little intensity

4:39 and then once I do notice that there's a

4:40 change then I need to execute resilience

4:42 and say okay well how quickly can I

4:43 return to baseline and so what's

4:45 interesting about that is that all of

4:47 those are behaviors rather and moods or

4:49 emotions around it because you can say,

4:50 "All right, well, I can still be upset

4:52 about this thing, but if my behavior

4:55 returns to baseline, then I can resume

4:58 original function, which is going to

4:59 disrupt minimally the system that I'm

5:00 operating within or that I'm a key key

5:02 role in in making, you know, work." And

5:05 so, um, that's my little side quest.

5:08 Coming back to that, I think what allows

5:10 any founder is having some worldview. It

5:13 doesn't have to be mine by any means,

5:15 but like I mean I've seen super

5:16 religious people who like they're like,

5:18 "Well, all of this is happening because

5:20 God wants this to happen and this is

5:21 happening for me, not to me." Great.

5:23 That's just as strong maybe stronger of

5:25 a worldview. Um but I think you should

5:27 have some overarching um

5:30 to couch the experience.

5:31 Yeah. schema. Yeah. Exactly. That that

5:32 everything operates within. And I think

5:34 that makes it significantly easier

5:35 because I think if you lack that

5:37 um you become the center of your

5:39 universe. you this is almost like a

5:42 victim aspect of it.

5:43 Yeah. I can't believe this is happening

5:45 to me. Like all of that kind of stuff.

5:47 It's not fair. This other company's

5:48 doing better. They got this better

5:50 funding or whatever it is. Um and it's

5:52 just like it just doesn't matter.

5:54 And in in either of those worldviews,

5:55 you know, like if you have the cosmic

5:56 relevance worldview, it's like, well,

5:57 none of us matter. Uh and if you have

6:00 the other one, it's like well the

6:01 endgame was never the achievement, but I

6:03 am the I am the goal. And if I am the

6:05 goal, then all of these things happen

6:06 through me, not to me. And so the idea

6:08 is how can I use these things to my

6:11 advantage to get if you have a a god

6:14 worldview what the creator wants me to

6:16 learn from them. But both of these

6:18 things can either make it a null point

6:19 or make you better.

6:20 And I see that as at least I think the

6:23 worldview of more successful founders.

6:25 Fascinating.

6:26 Yeah. Um

6:28 that's that's a really interesting model

6:30 of uh how how to think about I mean just

6:34 hearing you talk I just see it in myself

6:37 like there there are breaking points I

6:39 get to where my behavior turns into like

6:43 sort of more addictive type behavior.

6:45 Maybe I'll spend like too much time on

6:47 Twitter or

6:49 or and I like catch myself like what am

6:52 I doing? Clearly I'm there's an escapism

6:54 aspect of there and and then how can you

6:56 and over time the resilience aspect

6:58 started

6:59 um started to get better but you need to

7:02 to focus on it. Uh and and this this

7:05 idea of cosmic relevance is is really

7:07 interesting as a as a philosophy because

7:09 I think people see it as a some people

7:12 see it as a sort of like a form of

7:13 depression or something like that. But

7:15 it but I think it is when you read the

7:18 classics

7:19 there's always you know there's the sort

7:22 of the religious worldview there's the

7:26 um cynical worldview

7:29 and not in a negative sense and there's

7:30 also the niic worldview and not in a

7:33 negative sense I think uh when you

7:36 our our age today like people don't read

7:38 philosophy because it just feels like

7:40 it's useless in some sense it feels like

7:44 you you know, Cisophian task of like

7:45 just like carrying the rock. You know, I

7:47 spent like 2 years in San Francisco.

7:50 There's this like really crappy coffee

7:52 shop that would go to every Saturday and

7:54 we would read a philosophy paper and we

7:56 would just like discuss it endlessly.

7:58 And it it just h just understanding how

8:00 people thought in the past have

8:03 have been hugely valuable because

8:04 because I I can also pick these mental

8:06 models or ideas from different

8:09 different you know schools of thought.

8:11 And the other thing and and maybe this

8:13 is where we get into the 100 million uh

8:15 money models. Um

8:18 it's so you make it sound so simple. Uh

8:23 in some sense it is simple. It's sort of

8:24 like like Warren Buffett talks about

8:27 investing is about buying a business and

8:30 holding it, you know, buying successful

8:32 business.

8:32 Pick right.

8:34 Pick right and pick things that are

8:36 going to be here in 100 years. You

8:37 and pick the right price and and and and

8:39 that's it. um in in your case uh acquire

8:43 customers that could pay for themselves

8:45 in 30 days, right? The idea of an

8:47 unckillable business that actually is is

8:50 no one in Silicon Valley thinks that

8:52 way.

8:52 Yeah. In Silicon Valley, there's this

8:54 aspect of like burn as much money as you

8:56 can, capture as m much market share,

8:58 then at some point ruck pull every

9:03 uh and and and and actually try to focus

9:05 on fundamentals because because the

9:07 public market, you know, we've seen this

9:08 with Uber and and other cases like that.

9:11 Um

9:11 I love the fundamentals are just making

9:13 money, but Yeah.

9:14 Yeah. Exactly. Yeah.

9:16 So So may maybe uh give me your your

9:18 thoughts, reflections on that. What what

9:20 made you go the your route the the the

9:23 sort of the more tried and true versus

9:24 the Silicon Valley cuz I'm sure you know

9:27 Alex you know 25 year old Alex could

9:30 could have easily gone to San Francisco

9:32 and like went that route.

9:34 Well, so a couple things. So one is I I

9:37 never once in my whole younger life

9:39 considered tech as a as a as a career

9:41 path.

9:42 No one I knew was in it.

9:44 No one that I went to school with did

9:45 it. Mhm.

9:46 So like I only discovered it far later

9:49 in my career that it was even like on

9:51 the menu of options, right? And so you

9:54 know for me coming out of college it was

9:55 just I I I didn't have this big drive to

9:58 like change the world or make some

9:59 product. It was just like I would like

10:00 to not be poor.

10:01 That would be great. And so that was

10:03 kind of my my judge was like get a job

10:04 that pays decently well. Um and so

10:09 I did that and then obviously you know

10:11 fast forward here I am now in more

10:12 traditional business because that's what

10:13 I that's what I kind of got into. Um,

10:16 but I do think that there's kind of two

10:17 equal opposite perspectives on taking

10:19 market share and I think Silicon Valley

10:21 focuses on one of them and I think

10:22 there's nothing wrong with that. I think

10:23 it's probably the better way, but

10:24 there's there's there's an equal

10:26 opposite way of doing it. So, one is

10:27 that you drive CAC as low as as as down

10:29 low as you possibly can.

10:30 Um, which I would imagine is more the

10:32 Silicon Valley viral product way, right?

10:34 The other way is that you drive LTV so

10:36 high that you can outspend everyone else

10:38 in the marketplace. And so, but it's

10:40 still cact like it's still the same

10:42 dynamic and you one of them you just

10:44 want to go to either infinitely low as

10:45 it approaches zero or infinitely high

10:47 that you can outspend everyone. Both of

10:48 them fundamentally accomplish the same

10:50 thing. It's just that most people you

10:52 know you can do this so much so much

10:54 more from a product I guess you can do

10:55 both from a product perspective. Um but

10:57 there's typically more marketing and

10:58 sales that goes into the side and um

11:01 I think it also depends on the scale

11:02 that you're looking for. like at a

11:03 certain point paid acquisition can get

11:05 increasingly costly and inefficient

11:08 and you get rock pulled by the Facebooks

11:09 of the world often and all that

11:11 right and they change the algorithm when

11:12 they know that you're making money and

11:13 they all of a sudden they just charge

11:14 you twice as much for you know

11:15 impressions etc etc but like um

11:18 it still worked very well because again

11:20 I do 80% you know of businesses in the

11:23 United States are or 78% are service

11:25 based businesses so it's interesting

11:28 because like the a huge percentage of

11:30 market share is Silicon Valley but from

11:32 a logo perspect perspective, it's

11:33 astonishingly few comparatively to the

11:35 number of businesses that exist on basic

11:37 fundamentals of charge more than it

11:38 costs you and do it over and over again,

11:40 right? Um, and so I've lived more in

11:42 that world of just always fundamentals.

11:43 And so to the same degree I um bootstrap

11:46 I've bootstrapped every business that I

11:47 have. Um, and I think that my my end

11:51 goal has always been freedom um and

11:53 maximum flexibility of options or

11:55 maximum optionality. Um, and I think as

11:58 a result that's shifted some of my

12:01 decisions because I think to a degree

12:02 like um, do you guys have do you guys

12:03 have VCs who back your stuff?

12:06 Yeah. Yeah, we have a lot of VCs. Yeah.

12:07 Yeah. Um, and so if you you have a bunch

12:10 of VCs like there's there's covenants,

12:11 there's controls, there's voting,

12:13 there's all this stuff. And um, not to

12:15 say that maybe someday in the future I

12:16 won't have that, but like that all felt

12:19 heavy to me. And if if I believe what I

12:22 say I believe, which is that in five

12:24 generations it won't matter, then I'd

12:25 prefer to ride this ride the way I want

12:27 to ride it. Right.

12:28 And and Okay. So, uh c can you introduce

12:32 the the 100 million money model?

12:34 Uh and um and you know it it it looks

12:39 very simple. Why aren't more people

12:41 doing it?

12:42 I think it takes skill. I mean, I think

12:44 it just takes almost as much skill as

12:46 probably less skill than building a

12:48 product that goes viral, but it takes

12:49 some skill.

12:50 Um, and so fundamentally, I call it

12:52 client finance acquisition, but

12:53 basically is that one customer comes

12:56 embedded within it enough gross profit

12:57 to pay for that customer the cost of

12:59 delivering to that customer. So, CAC

13:00 plus COGS and CAC plus COGS on the next

13:03 customer. If you can accomplish that,

13:05 then and within a 30-day cycle, then

13:07 almost all businesses have interest free

13:10 cash available to them on a 30-day

13:12 timeline, which is why I picked 30 days.

13:13 Now, if you can get interest free cash

13:15 on a 90-day timeline or a 60-day

13:17 timeline, um then you could expand that

13:19 same concept. Um

13:21 but the idea is that at that point, cash

13:23 is no longer constraint to growth.

13:24 You'll still have other constraints for

13:25 sure.

13:26 Um but at least cash won't be the

13:28 constraint to growth. So, how can you

13:29 grow a bootstrap business like it's

13:31 venture-backed with that kind of

13:33 aggressive growth?

13:36 You do client finance acquisition, which

13:37 is accomplished through a money model,

13:38 which is what I wrote the book about.

13:40 Um, and so like, you know, we had a

13:42 software company that we went from zero

13:43 to like almost $2 million a month in six

13:44 months. And that wasn't because we had a

13:46 viral product, but because we knew we

13:47 could outspend everybody, and that's

13:48 what we did, right? Um, and each of

13:50 these businesses and like my gyms, I was

13:52 able to fill up

13:53 all my gyms before I opened the gym. And

13:55 that's not something that's is common,

13:56 you know, common place to do. And I

13:58 would do those all in 30-day launches.

14:00 And so

14:02 that idea of well, if we can just get

14:06 the customers to pay for the next

14:07 customer, then I just need to make that

14:09 first sale. And then after that, I can

14:11 spin the wheel as fast as fast as my

14:13 operational capacity can can handle.

14:14 It's a flywheel.

14:15 Yeah. And then at that point, the only

14:17 thing that's going to happen is, you

14:18 know, CAC's going to potentially go up.

14:20 Um, but if you have, you know,

14:21 tremendous LTV, then you're pretty much

14:23 good to go still even in that in that

14:25 shorter timeline. And so all the stuff

14:26 that I talk in here is basically 15

14:28 mechanisms that I use to try and pull

14:30 cash forward uh with the customer so

14:32 that we can accelerate the cash

14:33 conversion cycle so that we can

14:34 ultimately remove cash as a limitation

14:36 for growth which I think from a Silicon

14:38 Valley you know uh VC backed perspective

14:41 even on the low side if you already have

14:43 VCs if you no longer if you eliminate

14:45 your burn rate then you gain tremendous

14:47 leverage of course right and then also

14:49 you can time your rounds so that like

14:50 maybe maybe this is a bit of a building

14:52 year and you don't you don't want to you

14:54 know you could grow faster but you know

14:56 it's not in the best interest of the

14:57 business right now. And so it's like

14:59 okay we're going to have a slightly

15:00 lower growth year because we want to fix

15:01 some of the fundamentals of the product.

15:03 But if you are burning cash then you're

15:04 you know you could potentially get

15:06 forced to raise a down round and if you

15:07 get raise a down round you're screwed

15:09 and all the momentum's gone and it sucks

15:10 right and so having these types of

15:13 skills or tools in your back pocket I

15:14 think can be tremendously valuable for a

15:16 founder who's obviously bootstrapped

15:18 because that's how you can grow really

15:19 aggressively but even if you are

15:20 somebody who has you know venture

15:22 backing it can get you out of sticky

15:24 situations.

15:25 Yeah. So I um I maintained full control

15:28 of the company. It's not like I don't

15:29 trust the VCs but

15:31 uh every round I was able to dictate the

15:34 terms partly because every round I

15:36 raised it with you know boatload of

15:38 money still in the bank.

15:39 Yeah.

15:39 Uh we're still like a money burning

15:41 business. Uh but we've been fairly

15:44 capital efficient. Uh, and I think I

15:46 think that's like the main that's the

15:48 main leverage like you you be again

15:50 going back to how you want to spend your

15:52 day, how you want to build your

15:54 business, that kind of freedom.

15:55 Um, it's very easy to build a jail for

15:58 yourself.

15:59 Yeah.

15:59 Uh, and astonishingly easy.

16:02 It's very easy to to be way unhappier as

16:05 an entrepreneur than you were as an

16:07 employee. Um there's a sense of freedom

16:10 as an employee where you know you clock

16:12 out at 5 and you don't have to worry

16:14 about anything but you can build a

16:16 pretty miserable jail for yourself uh as

16:18 a as an entrepreneur.

16:20 Um what kind of businesses have you seen

16:22 built with this with this model? Like

16:24 you give me a story of someone that

16:27 applied the recipe and was able to to to

16:30 to to grow a business.

16:32 Oh, I mean there's like tons I mean any

16:34 category you can think of that's like a

16:36 traditional service business can can use

16:37 this. Yeah.

16:38 Plumbers, HVAC, gyms, B2B consulting

16:43 agencies, like I mean it it works.

16:45 Yeah. Yeah. But do you have a story of

16:47 of you know someone um like maybe give a

16:51 like a personal story like how did they

16:52 get started or

16:54 two kids wanted to start an agency for

16:55 float tanks which is just like a random

16:58 niche to be in

16:59 and they read my gym launch book where I

17:01 talk about this concept not as elegantly

17:03 as I do in this book but I was like hey

17:04 we run these you know six week

17:05 challenges by doing that and they were

17:07 entering a flow tank space where people

17:09 were trying to get people on um like $49

17:12 or $99 a month memberships but the cost

17:14 of acquisition was closer to $200. plus.

17:16 And so they would have to wait

17:18 two or three months to break even with

17:19 gross profit in order to just recoup

17:21 payment, which was really difficult for

17:22 flowch owners that already had expended

17:23 a ton of cash to open these open up

17:25 these facilities.

17:26 And so they came in just with a

17:27 different offer, which was like a six

17:28 week stress release thing. And so they,

17:31 you know, bundled in um some mindset app

17:34 and some like journaling and like a once

17:36 a week touch base with somebody who just

17:38 text them. And by doing that, they were

17:39 able to charge $600 on the first

17:41 transaction.

17:42 So that's the attraction part,

17:43 right? That was an attraction offer. And

17:44 so they were able to do uh they did a

17:46 win your money back offer but not off of

17:48 like you know losing weight or

17:50 something. They just did it based on

17:51 behavior. So if you do these you know

17:53 you attend twice a week for at least

17:54 this period of time and you rate your

17:55 stress levels but between the beginning

17:57 end if your stress levels don't go down

17:58 by at least two out of 10

17:59 um by the end we would give you your

18:01 money back right and that was the offer.

18:02 A lot of people like oh that's pretty

18:03 compelling. Um and as a result they were

18:05 able to you know offset acquisition

18:06 costs and and the model fundamentally

18:08 worked. And so that's so that's like an

18:10 like there's hundreds like there's a

18:11 zillion examples of this. If you were a

18:12 dentist, it' be the same thing. You do

18:13 Invisalines um and it cost you a X to

18:16 get an Invisalign. Person comes in, you

18:17 make three times your money and then

18:18 after that you get them onto the

18:19 subscription, which for them is going to

18:21 be cleanings and whitening, you know, on

18:23 ongoing basis. And so almost every

18:25 business can structure it this way. It's

18:27 just it's interesting because most

18:28 businesses want to have a low barrier to

18:30 offer thing, which I think is really

18:31 good when you're talking about what the

18:34 very first thing that someone touches in

18:36 your business from an advertising

18:37 perspective, but not necessarily the

18:39 first meaningful transaction.

18:40 Yep. Um, and so I I come from the

18:43 perspective that someone is taking

18:45 action because they have some level of

18:46 deprivation or some level of problem

18:47 that they're trying to solve in that

18:48 moment. And so I think people have very

18:50 large amounts of motivation for very

18:51 short periods of time. And so when they

18:53 have these large, you know, moments of

18:55 motivation, we would like to capitalize

18:56 on that with the largest transaction

18:58 possible rather than the smallest

18:59 transaction possible. Um, and in so

19:01 doing, you know, larger commitment from

19:03 the customer, you know, and if you

19:05 structure some of the mechanisms that I

19:06 talk about, it's like you can actually

19:07 get them to be more likely to activate.

19:09 Um, even if you have a B2B software

19:10 that's, you know, higher touch and if

19:12 they integrate their, you know, their

19:13 their tech stack or whatever it is, um,

19:15 you can put rebates in there so that

19:16 they pay $5,000, but then they get

19:18 $1,000 back if they do step one, two,

19:20 three, and then all of a sudden, you

19:21 know, that 5x is LTV on the back end,

19:22 but you still liquidate the acquisition

19:24 cost. And so,

19:25 um, that's those are those are the

19:27 variables, right, that we like to play

19:29 with. And different businesses lend

19:30 themselves to different things in terms

19:32 of which structures like buy X, get Y

19:33 free, uh, works in some businesses and

19:35 less so in others. you know, a decoy

19:37 offer or pay less now pay more later

19:39 offer or when your money back offer or

19:41 giveaway offer. Like there's all these

19:42 different offer structures that work

19:44 phenomenally well um to pull cash

19:47 forward and also generate a lot of leads

19:49 which is kind of the best of everything.

19:51 You know, uh the the sort of internet

19:54 business mantra has always been low

19:57 friction. And yes, there's an aspect of

19:59 it that's very true. Low friction wins

20:02 in many ways, but there's an aspect of

20:04 it like we've had many experiences at

20:05 Replet where we we find that higher

20:07 friction leads to to more conversion

20:10 because you're asking more of the

20:11 customer upfront that leads to uh more

20:14 investment and maybe a song cost

20:16 feeling, but also it might set them up

20:18 for for success by putting in the work.

20:20 This is there's a great hypothetical

20:22 extreme that you can take when you're

20:23 like trying to explain this to someone,

20:24 which is like, okay, well, if less

20:26 friction were truly the answer, then we

20:28 should just have ads that just are just

20:31 like buy here.

20:32 Mhm.

20:33 And we know that that typically doesn't

20:34 work.

20:35 Yes.

20:35 Right. And so if that's if that's not

20:37 true or if it's just like call this

20:39 number right now and buy our thing for

20:40 this amount of money like that, we we

20:43 know that doesn't work because we don't

20:44 see that very happen very often unless

20:45 it's super lowc cost consumer products,

20:47 right? Um, and so then it means that

20:50 there's some level of friction that

20:51 literally makes it more efficient to

20:52 spend the money by adding friction. And

20:55 so then the question, at least for me

20:56 when I think about marketing in general,

20:57 is what is the sweet spot? And so that's

20:58 what I think all marketers are trying to

21:00 find is like you want to find just

21:01 enough friction that you can have as

21:03 many people exposed to the ideal path to

21:05 getting them to, you know, to ascend or

21:06 to activate.

21:07 Yeah, I love that. Um, uh, reductio at

21:10 absurdum, right? like like find the like

21:13 the most absurd case and that helps you

21:16 prove the point.

21:18 Um uh and and then uh the other parts of

21:22 the funnel, the upsell, the downell and

21:24 the continuity are all very important.

21:26 One thing that we haven't done very well

21:28 is the downell

21:30 and like after after reading about that

21:32 I'm I'm just gonna go back to office and

21:33 just think about that.

21:35 um like you know there's so many dark

21:39 patterns on on churn

21:41 where like you hide the you know I

21:43 remember I subscribed to the uh

21:46 economist uh newspaper uh and I had to

21:49 call a number in England to cancel my

21:53 subscription.

21:54 Yeah. During their work hours.

21:55 Yeah. During their work hours I was like

21:57 I'm not I'm never going to recommend

21:59 your magazine to anyone. Um but but I

22:02 think instead this idea of like hey well

22:05 okay you don't get the best value out of

22:07 the service. Well how about we we you

22:11 know give you um you know we we give you

22:14 a plan that perhaps suits you better.

22:17 Stick around longer. See if it works for

22:19 you.

22:20 Um

22:21 something you might find really

22:22 interesting. So we found this out when

22:23 we did a big data analysis at gym launch

22:26 which obviously B2B and higher ticket

22:27 but I still think with psychology works

22:28 the same way. Um, so we looked at all

22:32 the LTV of all different customer

22:33 segments across different service

22:34 categories that we had. And the second

22:37 highest LTV, you know, the first highest

22:39 was the most expensive thing of the best

22:41 avatar. I'm sure that was number one.

22:43 But number two was the one that was

22:44 super surprising to me. It was very

22:45 close to that one and it was a

22:47 significantly lower price point, but the

22:48 churn was almost nothing.

22:50 And what it was was customers who had

22:52 been proactively downsold.

22:54 So, not somebody who's going to cancel

22:56 and then we say at that point, hey,

22:58 let's, you know, let's downell you

23:00 because you're about to leave, but

23:01 instead someone that you saw that they

23:04 uh were not using the complete feature

23:06 set that there was a different service

23:07 category that matched their exact usage

23:10 case that cost less. And so, our team

23:13 reaching out and saying, "Hey, you're

23:14 not using this element of the service.

23:16 We have another service category or tier

23:18 that literally is 100% of what you're

23:20 currently doing. If you want, I can go

23:22 ahead and switch you to that." people

23:24 were so like, "Oh my god, this is

23:26 amazing."

23:26 Yeah. It generates such goodwill.

23:28 Yeah. And then they just stuck. They

23:30 just stuck. And I found that um I I

23:32 always kind of like remember that like

23:33 we always have this fear as business

23:35 owners, you know, to like I don't want

23:36 to just don't bother them. They're

23:38 paying us, right? But like you can

23:40 pure margin. But sometimes it's like but

23:42 you but if you can just crush churn and

23:44 and triple LTV with that move, then like

23:46 go for it. And also I think there's some

23:47 element of ethics in there that how much

23:49 word of mouth comes from that, how much

23:50 good sentiment, reviews, all that kind

23:51 of stuff. Less measurable. Um, but at

23:54 least we could see LTV on those

23:55 customers was the cohort that had the

23:57 second highest LTV and that was like and

23:59 it was at like less than half the price.

24:02 Yeah.

24:02 So to me I was like that's really

24:04 compelling and interesting.

24:05 Yeah, that's fascinating. Uh, I could

24:06 think of an example for for Replet where

24:09 you know some people build something,

24:11 they deploy it, they're getting value

24:12 out of the deployment, but they're

24:14 paying the $25 and maybe that's a month

24:16 and maybe they're not building every

24:18 month.

24:18 And so like downselling to say, hey,

24:20 keep your deployment, maintain your tool

24:23 that you built, but like you don't have

24:24 to continue paying for the credits for

24:26 for the agent or the AI that you're not

24:28 using. And I I wonder because even the

24:30 psychology because like I mean I love

24:32 the psychology behind this stuff but

24:33 like

24:34 even making the offer that they could

24:36 get some let's say it's $15 or $10 a

24:38 month whatever

24:39 and then if someone says no I'd rather

24:41 still have the optionality

24:43 it's almost like a reaffirmation that

24:45 they're not going to like I would bet

24:46 you even having to make the decision

24:48 that they don't want the downell would

24:51 make their current membership stick

24:52 longer.

24:54 Fascinating. I'm going to do that and

24:56 report back to you. See what happens.

24:58 That's very cool. Um, so let's let's get

25:00 into AI a little bit. Um,

25:04 asking you the questions. I don't know.

25:06 So, actually, um, not many people

25:09 remember, but you were one of the first

25:11 people to key in on Chad GPT as a as a

25:14 tool for business.

25:15 Yeah.

25:16 And I remember some of your early

25:17 videos.

25:17 It was like three three almost two three

25:19 years ago. Yeah. It was a lot.

25:20 Yeah. It was like a couple weeks after

25:21 Chad was launched, right? Uh, in like um

25:25 was it November 22 or something like

25:27 that? December 22.

25:30 Yeah. Um, so uh and I was very

25:34 interested because I I always watch your

25:35 world and I watch Silicon Valley world

25:37 and I see like what is hopping over and

25:39 when I saw that I was like okay this is

25:41 going to be huge. Uh

25:44 um so so what what what like what

25:47 sparked the interest for for chat? What

25:49 what what vision did it create for you?

25:51 I mean I I think that there's there's

25:53 fun questions of like what is it? So,

25:55 I'm such a such a like what do I geek

25:57 out on like what do I really enjoy a

25:59 lot? Um is studying learning and

26:01 behavior and you can hear it through all

26:03 my my definitions and my business books.

26:05 Like that's the through line that that

26:06 marries everything. Um and so then then

26:11 I mean AI obviously begs the larger

26:13 question of like what does it mean to be

26:14 human? And I think and I have this

26:16 philosophical slant to my own interest

26:17 and I think that's probably why I was

26:18 drawn to it disproportionately. I also

26:20 am a writer and so language is a natural

26:21 thing for me. And so the kind of the the

26:24 confluence of multiple these things

26:26 together u made chat GPT really

26:28 interesting. Um but just in like we we

26:32 as humans tend to be very romantic about

26:34 our ourselves and thinking that like no

26:36 one can do what we can do and like we've

26:37 been proven time and again that we're

26:39 we're not as special as we think we are.

26:41 Um

26:42 that fits in your also irrelevant.

26:44 Yeah. So totally in my worldview. Um a

26:47 lot of people very upset by that.

26:49 But with with each thing that it proves

26:51 it can do because if if

26:54 artificial intelligence learns almost

26:56 exactly the same way that humans do

26:57 through reinforcement training. And so

26:59 you have you do a thing and you get an

27:01 outcome. Yay or nay. And I believe at

27:03 the most foundational level when you ask

27:04 the question like why why are you so

27:06 driven or why are founders so driven? We

27:08 have gone through some reinforcement

27:09 training which we may have been aware or

27:11 not aware of that reinforce this set of

27:13 behaviors that when you know stacked

27:15 together so you see uh a skill as a

27:18 behavior chain of multiple adaptive

27:20 skills that are put in sequence uh to

27:22 create an outcome that's maybe you know

27:24 ideal or

27:24 that adaptive chain becomes more complex

27:27 set of

27:27 uh behaviors then becomes a skill sorry

27:29 um

27:30 and so to the same degree AI basically

27:33 learns the same way humans do

27:34 and so if it learns the same way humans

27:36 do then it will be able to do what

27:37 humans do Okay, that's the very first

27:39 principles kind of thinking especially

27:40 watching Tachik. It kind of sucked back

27:42 then but like being able to key on that.

27:45 it just keeps tweaking and so um

27:48 it it's it obviously starts with with

27:50 just you know language but then it just

27:52 continues to move move out and I'm

27:53 curious what you think about like Tesla

27:55 as because you know it trying to us

27:59 trying to clean data sets to train on

28:01 larger and larger you know L you know

28:03 big data that it's that it's working off

28:05 of um obviously that will have limits

28:07 and then creating high quality data will

28:09 become the constraint of the learning

28:11 and so then it it will have to learn

28:13 from first principles itself which is

28:14 going to the experimentation in the

28:15 world which means that it has to have

28:16 some sort of link to the real physical

28:19 world which is exactly how we learn

28:21 right and so we're able like we have a

28:22 so everyone who feels like you know AI

28:24 is not as smart as humans it's like well

28:25 because we've had a learning advantage

28:26 yes

28:27 because we take 10,000 inputs a day or

28:29 whatever the number is%

28:30 um if we take all of our senses times

28:32 you know hours and whatever we can take

28:33 in um as soon as we have robo babies and

28:36 I say that not a like not a weird way

28:37 but like um where it can experiment and

28:39 touch and taste and do all that stuff in

28:41 the world I think the reinforcement

28:42 loops will happen so quickly and the

28:43 other part of it that will happen is

28:44 that it doesn't forget. And I think

28:46 that's the that's the crazy part. So

28:48 we've you know we've obviously trained

28:49 AI for different functions within our

28:50 business um and business use cases. And

28:53 one of the things that was astonishing

28:54 to me is that we were trying to train a

28:55 um an SDR um so a sales development rep

28:59 who would you know do outreach or do

29:00 follow-up. And what was really

29:02 interesting to me was that um obviously

29:03 the hallucinations are still a problem

29:05 right now uh with it and I think will

29:07 always be a problem. I'm sure you I

29:09 don't know the French guy, but the guy

29:10 who Google listens to and he's some you

29:12 know

29:13 French special might be um but he

29:15 basically was saying that like LLMs will

29:17 never truly get into AGI because if you

29:19 have 3% error it gets expounded you know

29:22 uh at infin item

29:23 and so uh

29:24 he did moderate his view but I'll tell

29:26 you

29:26 oh yeah please yeah tell me um but one

29:28 of the things that uh I found so cool

29:30 for me though is that like you know it

29:32 would it would make a mistake and I

29:33 would you know we'd be like hey do this

29:35 next time and immediate and every time

29:37 it would do And I was like, man, I've

29:38 trained a lot of sales people like scary

29:41 in a cool way of just how quick you can

29:43 learn and it just doesn't make a mistake

29:44 after obviously there's hallucinations,

29:45 but in terms of following the directions

29:46 it would follow them, right?

29:47 And so

29:48 if if you It's funny because there's

29:51 there's a lot of like really cute isms

29:53 and in Silicon Valley like you know

29:56 being a master is something that like

29:58 some skills cannot be taught and only

29:59 learned. Like there are these very cute

30:00 like rhetorical devices but they're just

30:02 not they'd make no sense.

30:04 Like if you learned it you learned it

30:06 which means someone could teach it. it's

30:07 just not structured

30:08 and we didn't know how we taught it

30:09 because there's multiple factors but

30:10 like if we if you control the conditions

30:12 you can control the outcome at least

30:13 that's my viewpoint of the world and so

30:15 it just means that there's more complex

30:17 environments in order to learn and so um

30:18 I think I've just taken a lot of

30:20 interest in that because you know the

30:21 obvious of like chat GPT enters Optimus

30:23 or you know Grock enters Optimus in the

30:25 real world it's like

30:26 oh okay well it can interact with all

30:29 the tools that we've already made around

30:30 the world to to interact with as humans

30:32 and never make a mistake after that and

30:34 then that just gets into lots of

30:35 questions around existence and all that

30:37 kind of stuff which I

30:37 I I I think you're you're spot on

30:40 perhaps more than you know because um

30:44 even a lot of AI engineers would not

30:46 understand what you just said and even

30:47 some AI researchers. So what what

30:49 happened is LLM's

30:53 exactly what you're talking about train

30:54 on the internet. Yeah.

30:55 Right. Next word prediction just like

30:57 keep at keep quizzing AI about like you

31:00 know uh give it the title to a New York

31:03 Times article quiz it on the first word.

31:05 Yeah. uh if it's wrong you change the

31:08 weights if it is right you reinforce the

31:10 weights and so on

31:12 um and so that that was the paradigm and

31:14 that's how we got GBT2 three up to four

31:16 that was the paradigm skip you know

31:17 train on everything

31:18 but that created uh all sorts of

31:20 problems like hallucinations and then we

31:22 ran ran out of data like that's it like

31:25 a limited set of data

31:27 which is crazy to consider like what

31:29 that even means

31:29 yeah um but then the next paradigm which

31:32 is reinforcement learning uh in in in in

31:35 in the like the paradigm after that. So

31:38 reinforcement learning from human

31:40 feedback was the thing that created Chad

31:42 GBT because you know they it would get

31:44 an answer and then there's a human that

31:46 says okay that's not a good answer

31:48 generated again that's not a good answer

31:49 generator. So this is like voting

31:51 basically based

31:52 and so that that made it so that chat is

31:55 a good chatbot.

31:56 Um now then the next thing is learning

31:59 from experience.

32:00 This is not a human annotating the

32:03 feedback. This is actually experience

32:05 with the with the real world and this is

32:07 what's creating really profound good

32:09 coding agents right now

32:11 because the experience from the real

32:13 world and agent is actually still a

32:14 virtual world. It's a virtual machine

32:16 where it's like writing the code and

32:17 executing it and getting feedback from

32:19 it.

32:20 And now this is being applied to

32:21 robotics as well.

32:23 So you know baby robot is exactly the

32:26 right right way to think about it. We're

32:27 going to go from um you know this idea

32:30 of like humans teaching it to more of it

32:33 like touching and experiencing the world

32:34 and and growing from it.

32:37 And by the way you know this happened

32:38 before this happened with Alph Go. The

32:41 way they were trying to train Go is the

32:43 game is the game that no one was able to

32:44 beat initially with AI

32:46 but but uh Alph Go in 2015 was able to

32:49 beat it. And it was because they moved

32:51 away from training on human data

32:53 to having the the machine play itself

32:56 like a trillion times

32:58 um and and that it created an alien

33:00 intelligence

33:01 and started inventing moves that we'd

33:03 never seen before. Yeah. Really

33:04 interesting. like really artistic

33:06 creative moves

33:07 that that till this day they're studying

33:10 and they're referencing and it changed

33:12 the game of go because people start

33:14 thinking about it differently because we

33:15 had this like very alien

33:17 that is first principles perspective

33:18 because it has no biases

33:20 that's right so I think that's really

33:22 the next

33:23 step of AI where we're going to go from

33:27 information systems which why I would

33:29 call chat GPT perplexity all the systems

33:31 that we had until today that that have

33:33 achieved wide user base. Maybe there's a

33:35 billion AI users today.

33:37 Those are information systems. Uh we're

33:39 going to go to action systems. So

33:41 agents, uh the SDR you're talking about,

33:45 that's what you're training there. Like

33:46 an action system. Uh now Frano was this

33:50 big skeptic of LMS like he's like look

33:52 that's not how we're going to get to AGI

33:54 because LLMs have the hallucination is

33:57 what uh he invented this benchmark

34:00 called ARC AGI and this benchmark a lot

34:03 of the big model labs so his skepticism

34:07 uh created new set of innovations uh so

34:10 a lot of the model labs started going

34:11 from training just on human data to

34:13 training on uh on experience and that's

34:16 how we got the 01 03 models, the

34:19 thinking and reasoning models,

34:21 and then and then they started scoring

34:23 really high on his benchmark, the RKGI

34:25 benchmark, which nobody scored really

34:27 well on. Actually, there was a

34:28 breakthrough two days ago with Grock,

34:31 and it was like 79% or something like

34:33 that, and we were like at 2% like a year

34:35 or two ago.

34:36 Uh, and so he's like, okay, I could see

34:39 how LLM's could could actually scale all

34:42 the way to AGI. So, now we're entering

34:44 the action regime. Um, and this is what

34:47 gets me really excited because we're

34:48 building replet, but also this is where

34:50 I think it starts to impact the real

34:52 world.

34:53 Um, so, uh, but but even even in the

34:58 information systems, there's a huge

34:59 amount of value that still hasn't been

35:00 tapped. Uh, ACQ AI, acquisition AI,

35:05 you essentially the way I understand it

35:07 is you embodied your your knowledge into

35:10 a system that people can go and and and

35:12 and talk to. You're scaling yourself in

35:14 a way. talk to me about that.

35:16 So, um, so there's a few things. So, one

35:19 is when we were building it, I wanted to

35:20 be really clear to our dev team. I said,

35:22 I don't want this to look like a chaty

35:24 like I don't want this to just be like a

35:26 chat interface because that's not going

35:28 to be like I don't want to be compared

35:30 to open. It's like we're not we're not

35:32 going to win that game. That's not the

35:33 goal.

35:34 And so, there's a couple things that we

35:35 did that was unique. One is we added in

35:37 a context layer that would be

35:38 permanently in stuck. Um, I don't know

35:39 why all GPS don't do this, but like, um,

35:42 so everybody puts all their business

35:43 information in, which would be all the

35:44 questions that I would probably ask at

35:46 like at the at the top level before

35:48 answering a question. And so if you were

35:49 to ask because you know what our our a

35:53 normal user would come in and say, okay,

35:55 I'm going to talk to Alex AI for

35:56 example,

35:57 um, it's not it's ACQA

35:59 and say, what should I do with my

36:01 business? Well, if you were to ask me in

36:03 real life, I would say, I don't know

36:04 anything about your business, so I can't

36:05 answer the question. And so I wanted to

36:07 make sure that it would always ask

36:08 questions prior to answering. And so

36:09 it's a very questiondriven

36:12 um AI in terms of how we built it. We

36:14 did a ton of QA on it. Um but the first

36:16 thing we're going to get is all the

36:16 context around the business first.

36:18 After that, then we're going to get

36:19 context relative to whatever the the

36:22 original question the person's asking

36:23 is. And then in terms of where it gets

36:25 the answer set from, um one of the

36:27 things that we spent a really long time

36:28 on was me trying to be very clean with

36:31 the data that I was putting as inputs

36:33 and outputs within the system. um for

36:35 well obviously it has all my books and

36:37 all that stuff in there. Um and so when

36:39 it answers questions it literally

36:40 searches um in in sequence it goes

36:42 through all books and then it goes

36:44 through all playbooks and then it goes

36:46 through the notes. Now the notes are

36:47 proprietary um like no one has access to

36:50 those and it's basically all the

36:52 consultations that I did over the last

36:53 two years purposely to train it.

36:54 Okay. And I have, you know, taken

36:57 businesses from, I mean, commercial

36:58 refrigeration to

37:00 to a YouTube creator to a, you know,

37:03 school community owner to, uh, you know,

37:05 an AI company, right? Like everything

37:07 under the sun uh, in terms of what those

37:09 consultations were about. And patterns

37:11 do start to emerge obviously the same

37:12 way that you get good at doing anything.

37:14 Like you see it enough, oh, this is

37:15 another one of these and well, this

37:17 probably has two of these three

37:18 problems. Okay, let's fix it this way.

37:19 And so, as a result, it started getting

37:20 really, really good. And so, um, the

37:22 feedback, honestly, I was very

37:24 pleasantly surprised. Like, we put a lot

37:26 of work into it, but you never really

37:27 know.

37:28 Um, but our usage has remained the same

37:31 since launch.

37:31 Yeah. I couldn't sign up because it was

37:33 still overloaded, I think.

37:35 Yeah.

37:35 Uh, but it's it's it's it's um it has

37:38 stayed the it has stayed the same. And

37:39 so, I was, you know, if you have any

37:40 retention in in a in a first product,

37:42 you'd be like excited about it. But we

37:44 have

37:44 the same at launch every day getting

37:46 used right now. Um, and so

37:49 what's limiting you from opening it up

37:50 to more people?

37:51 It's more about because I did the launch

37:52 and I did it closed, you know, with

37:54 relation to the book. I'll probably

37:56 reskin how I, you know, structure the

37:57 offer, right?

37:58 Um, in terms of like what I allow people

38:00 to have access to, what data it is

38:02 trained on that it's going to answer

38:04 from. Um, and so I'm we're actively

38:08 working on that right now, but we are

38:09 still uh heavily investing in ACQ AI

38:11 because if we think about what we want

38:13 to do with ACQ, I see I think W you

38:17 might have seen this WCOM came out and

38:18 said like we're looking for full stack

38:19 AI companies. So rather than saying,

38:21 hey, build a tool for law firms, just be

38:23 a law firm that does, you know, that's

38:25 AI first. And so when we think about AI

38:29 first, and correct me if I'm wrong

38:30 because I'm super curious your

38:31 perspective, but I thought of this as

38:33 well, you can't be AI first, you need to

38:35 be data first.

38:37 And then once you're data first and have

38:38 a data uh capture and enrichment layer

38:41 first, then you can be AI enabled,

38:44 right? And so that was kind of the the

38:45 operating thesis that we've had for the

38:46 last two years is like we want to

38:48 capture data in a very specific way so

38:49 that we can train AI on it. Um, but then

38:52 with the ultimate goal of being able to

38:54 be kind of replace what I see as a

38:56 pretty big gap in the market and

38:57 obviously where I make a lot of my

38:58 content, which is kind of low midmarket.

39:00 So there's tons of people who help

39:01 people get their first five customers

39:02 and I do that for free just for anybody.

39:04 Um, and then there's obviously Fortune

39:06 100, you know, Mackenzie Gartner,

39:08 whatever. I was like, but there's this

39:09 gigantic gap between

39:10 call it 1 million and 100 million a year

39:13 where businesses in my opinion where the

39:16 a huge percentage of enterprise value is

39:17 created on private markets. uh you know

39:19 going from a $10 million business that

39:21 has you know $3 million of profit to a

39:23 $25 million business with $10 million of

39:25 profit is just a gigantic step up in

39:28 valuation and wealth for the founder and

39:31 that's where like I have so

39:32 there's no advice at all

39:33 right there's nothing there there's

39:35 nothing it's just like either you got

39:36 the whole Silicon Valley world which is

39:37 just billy your bust right

39:39 and then you've got kind of small

39:41 business world which is like how to run

39:42 a local shop but there's nothing for

39:44 like how do you go from 10 to 100 and

39:46 there's just not a lot and so we've done

39:48 it a lot of times And so we feel pretty

39:50 pretty confident to be able to help

39:51 there. And so that's what the advisory

39:53 practice has been based on. And we have

39:55 heavy human you know intervention doing

39:56 that right now. But the goal has been to

39:58 capture the data such that every day the

40:01 AI is um further and further integrated

40:03 into our processes. So right now it's

40:05 that ACQI is al is an internal tool

40:07 first. Yeah. So our associates use that

40:10 to come up with preliminary

40:11 recommendations and the advisers kind of

40:13 say okay we need to add this we need to

40:15 tweak this and then making sure that

40:16 cycle is as fast you know fast as we can

40:18 so that overall eventually becomes you

40:20 know in the beginning it's it's it gets

40:21 one third right and then it's half right

40:22 and it's 70% then eventually it's like

40:24 [ __ ] this is 90% right

40:26 is there now a version of this that we

40:27 can offer that's cheaper than what

40:30 someone else can do and better and

40:32 faster and so anyways that's what we're

40:33 working towards right now

40:34 so you know uh to to your sort of

40:37 question about how to think about full

40:39 stack AI businesses. the the main

40:42 question any founder who's building an

40:43 AI business need to answer how is this

40:45 different than chatbot 100%

40:47 uh or why couldn't openai do it yeah

40:50 right if it is lucrative they're going

40:52 so much like they just started a job

40:54 board like it's a it's clearly a company

40:55 that's so ambitious you have to worry

40:57 about what they're going to do what

40:58 they're not going to do

41:00 um and you know to me the question is

41:05 what kind of domain specific information

41:08 you have that is not on the open web,

41:11 right?

41:12 And that is your IP. That is your IP

41:14 that no one else has. Like if you're a

41:16 lawyer

41:17 that specializes in certain case law

41:20 that is very niche, no one else knows

41:22 about, you can put that information into

41:24 an AI and have that be the best AI about

41:28 um I don't know like

41:30 Yeah. IP defense for Instagram res.

41:33 Yeah. Yeah. Exactly. like the more niche

41:35 the better and there's given how big the

41:37 internet is and how big the world is

41:38 there's enough market for that for you

41:40 to scale it up

41:42 um and so you know if you have that and

41:44 I tell tell a lot of people a lot of

41:46 people ask me like okay what's going to

41:48 happen when AI take more jobs or

41:49 whatever it's like you're going to have

41:51 domain knowledge that's not going to be

41:52 out on the open web try put this into an

41:55 AI

41:56 uh you might be able actually to

41:58 generate so much money without working

42:01 and it's it's like the ultimate It's the

42:04 ultimate passive income. Yeah.

42:06 I you shouldn't go into it thinking that

42:08 way, but I think the you know,

42:10 so you know, my my view of where AI and

42:13 agents are headed,

42:15 we're going to get to a world where uh

42:19 AIs will be able to go contract out to

42:21 other AIs. So, I'm going to have an AI

42:24 or, you know, society of AIS that's

42:26 running my business. But anytime, let's

42:29 let's say we got sued by Instagram res.

42:33 Uh and it's like, okay, holy [ __ ] we

42:34 don't know how to do that. Well, our

42:37 legal agent can go out and find an agent

42:41 that is trained on that

42:43 uh and actually just facilitate the

42:45 transaction. And you can sort of imagine

42:47 how fast the world becomes

42:49 when when it's like AI to AI protocol.

42:52 Um, and I think uh I think you know

42:54 there's going to be a lot of businesses

42:56 where you don't have to hire a lot of

42:58 people in order to make a lot of money.

43:00 Um, and so I think it is such an

43:03 exciting time for entrepreneurs,

43:05 solarreneurs, small businesses.

43:08 Uh, those are the people that are most

43:10 worried today. But I think they're, you

43:12 know, those who understand it should be

43:14 the most excited. But it doesn't mean

43:16 SAS is dying. It means SAS becomes a

43:18 small business yeah

43:20 endeavor which is I think that's great.

43:22 Um and and so uh and so I I think that

43:27 if you're uh if you're someone who wants

43:30 to start a business um and you have an

43:34 experience working at a company the

43:37 first thing to think about is what kind

43:39 of domain knowledge you know that chat

43:42 doesn't know.

43:42 Yeah.

43:43 And that's a great formula for starting

43:44 a business.

43:45 Yeah. No, I mean

43:47 I obviously agree. You know what it's

43:50 interesting because um you know when we

43:52 come from the investor perspective

43:53 because I have you know that hat as well

43:55 at acquisition.com.

43:56 Um I kind of see like two operating

43:59 patterns for investing. So one is I want

44:02 to be one of the winners in AI.

44:05 Uh the alternative is asking the

44:07 question which is Basos's question which

44:08 I love which is what won't change?

44:10 Mhm.

44:11 And then doubling down on that. And I

44:12 think both of those are are great you

44:14 know operating perspectives. I think the

44:15 returns will be significantly higher in

44:17 the um winning the AI, but there will

44:19 also be tons of losses.

44:21 It's funny, there's a book that it's

44:22 called the Innovator's Dilemma. Have you

44:24 heard of it?

44:25 I've heard of it. Yeah.

44:25 Yeah. It's it's it's a really

44:27 interesting idea, which is um once

44:30 you're successful as a business, uh

44:33 you're creating the

44:35 uh the conditions for your own

44:37 disruption. Yeah. because you found a

44:40 certain success with certain customer

44:41 base and that customer base asking you

44:44 for certain features that opens up the

44:46 down market, the lower end of the market

44:49 for a disruptive technology that when

44:51 you first look at it, it feels like it's

44:53 not a threat. It's just a toy.

44:55 Yeah.

44:55 Uh but you know, pretty quickly

44:58 and all disruptive tech is like that.

44:59 Yeah. pretty quickly it'll you know

45:01 because it has more users it'll have

45:03 economies of scale or some other

45:05 advantage network effects whatever it'll

45:07 move up market and it'll start to

45:08 disrupt uh your business when uh Clay

45:11 Christensen uh which was a Harvard

45:13 business I love that yeah he's awesome

45:15 what is that what is it

45:16 jobs to be done

45:16 no he has that but he has the other book

45:19 he has which is like

45:20 what is the meaning of your life yeah

45:22 yeah yeah yeah um so yeah I recommend

45:25 people go look at his stuff uh the the

45:27 great lecture about like your your me

45:29 meaning and and your work. Um but when

45:31 he was writing this book it was in the

45:33 80s or 90s and it was in the hype era or

45:36 when storage uh became so cheap that you

45:40 had a lot of companies like SanDesk and

45:43 all these companies come up and sell

45:45 sell you know storage consumer storage

45:46 devices and he says every six months

45:48 we're getting disrupted and he says that

45:51 when I was when I look at you know you

45:53 look at biologists genet geneticists and

45:56 they want to study genetics and DNA they

45:59 study fireflies because fireflies uh get

46:02 born in and die in a matter of of a

46:05 couple days I think or or maybe 30 days

46:07 or whatever it is

46:08 lots of cycles so you can study and he's

46:10 like I like to study the hard disk

46:12 business because it was it was like a

46:14 fruitfly and I think a lot of AI

46:16 companies are like fruit flies and how

46:18 that are getting born and disrupted very

46:20 quickly die right yeah exactly um so

46:23 just to to get into uh replet a little

46:25 bit and and and broadly uh just software

46:29 uh you built so software businesses. We

46:31 just released agent 3.

46:33 Yeah.

46:33 And the main idea behind agent 3 is it

46:36 refactors the work. So refactor is a

46:37 word for uh when you're creating a mess.

46:39 Like imagine you're cooking. As you're

46:41 cooking, you create a lot of mess and

46:42 you need to clean it up in order to keep

46:44 cooking.

46:44 Uh and so with programming, you need to

46:47 be continuously cleaning up. Uh and

46:50 non-programmers do not know to prompt

46:52 the agent to do that. So now we have um

46:55 we introduced actually it's a multi-

46:57 aent system. Now we introduced an agent

46:59 called the architect

47:00 and the architect

47:01 matrix.

47:02 Yeah, exactly. The architect will look

47:03 at the code that the like main coding

47:05 agent built and like that doesn't look

47:08 good, that's not secure, that's not very

47:09 good, that's not stable, kicks it back

47:11 and the architect loops again. And we

47:13 also have a testing agent that opens the

47:15 browser window so that you don't

47:16 constantly test it.

47:17 You have to sit there every 5 minutes.

47:19 That was my that was why for me putting

47:21 30 hours in it was probably only like

47:23 uh 2 hours of actual prompts and then

47:26 like

47:27 you know 10 or 20 of just sitting

47:28 waiting for the thing and you know

47:29 let me show you the update actually it

47:31 would be pretty cool. So we we played

47:34 around our team before we got here. This

47:38 is only like a couple prompt. This is

47:39 actually just one shot this app but we

47:42 wanted to create a money models app.

47:43 Okay. So the prompt is um self-funded uh

47:47 funnel simulator AIdriven calculator to

47:49 model money model metrics. This app

47:52 would let entrepreneurs input their

47:53 offer price cost expected conversion

47:55 rate then simulate if their funnel meets

47:57 alozi selfunded criteria whether in

47:59 30-day gross profits per customer is

48:01 more than twice that the cost which

48:04 uh so this is what it created. So you

48:07 can, you know, create the uh CAC here,

48:10 you can create the offer name, um, and

48:13 then start to simulate it via

48:15 visualizing the flow.

48:16 So that's cool.

48:17 So by the way, it innovated a lot of

48:19 that stuff. So we that wasn't exactly in

48:21 the in the prompt.

48:22 So here's where the offer is. So you can

48:25 see the the sort of the funnel. You can

48:27 see the attraction, upsell, continuity.

48:29 I guess it doesn't have the downell, but

48:30 we can probably add it here. Um, and so

48:35 here's a metric dashboard. You see like

48:37 a big red thing. It's like, okay, this

48:38 bot selfunding.

48:40 I don't know if ACQ does this. Probably

48:42 does, right? Does this sort of thing?

48:43 Oh, we just do it. I mean, we we do do

48:46 it by hand, the old fashioned way, you

48:48 know.

48:48 Uh, here's the AI feature where it says

48:51 uh where you can just like click analyze

48:53 my funnel uh and your funnel needs 80%

48:57 to achieve. Um, so

48:59 consider raising the price by 20 30%.

49:01 Okay, that's it has a nice little

49:03 suggestion there.

49:04 So, I I think what we can do is we can

49:07 change that app or we can start a new

49:09 app uh if you'd like. So, just to show

49:11 you some of the new features, what what

49:12 should what feature should we add or

49:14 some of the fix? For example, we can

49:15 make this interaction to a chatbot. That

49:17 could be interesting or

49:20 Oh, you know, I wonder if um like,

49:25 you know, add another offer or something

49:27 like that to the flow would be

49:30 something we could do here. So, we can

49:32 we add another You want to add an offer?

49:33 Okay, so it has that.

49:35 And so, the issue is we need to have

49:37 something that's way more expensive. So,

49:39 it probably be like, how do we put a

49:41 $1,000 thing on here knowing that only

49:43 like 5% of people will buy it?

49:45 Like an upsell.

49:46 Yeah, like an upsell.

49:47 here. Um, miniourse upsell. Let's change

49:51 it to like full course upsell

49:54 price 997. Yeah, sure.

49:59 Um, and then conversion rate.

50:01 It's five. Yeah. Let's go five. See how

50:03 that goes.

50:05 Now we look.

50:07 All right.

50:08 So then we have

50:09 it's at point4 now because it was at 0 2

50:10 before.

50:13 So that added So we need another So we

50:16 we're we're so far behind.

50:18 That's interesting. Why are we so far

50:19 behind? Maybe because the initial price

50:22 is so low and conversion rate is low.

50:24 Could rate 15. That's pretty that's

50:25 pretty high. 15 on a on a front-end

50:28 offer.

50:29 Okay.

50:29 And then the upsell.

50:32 So that's the 997 upsell. Okay. And the

50:34 monthly membership conversion, right? We

50:36 replaced one of the offers, right?

50:38 Yeah, we replaced one of the offers.

50:39 Okay. So we replaced it. So, we need to

50:43 uh I mean, if we want to make we we

50:45 could always just change reality and be

50:46 like, "Hey, we're at 20% conversion on

50:48 our on our $1,000 thing." And then

50:50 Yeah, let's let's just see make sure it

50:52 it actually works.

50:54 So, we're closer. Um but what's our

50:58 Okay, so there is definitely a bug here

50:59 because if if cost of acquisition is one

51:02 No, no, there isn't a bug because 150 is

51:05 cost of acquisition. So that's what's

51:07 off is that I don't know where the the

51:08 150 comes from, but

51:10 Oh, it's right here.

51:11 Okay. Yeah, because if you're if you're,

51:12 you know, buying a $10 customer, your

51:14 cash probably

51:15 that's that's an awful business.

51:16 Yeah, that's that's a pretty bad Yeah.

51:17 So if you're if you're C, you're

51:18 probably calcing

51:21 like a true $10 thing on the front end.

51:23 Okay, cool.

51:24 There we go. Now we're

51:26 Yeah, we're crushing it. Now we're

51:27 crushing.

51:28 So what what kind of feature should we

51:29 add uh add to this?

51:32 Maybe let's make this into a chatbot.

51:34 Okay. Okay.

51:34 The analysis because it just gives you

51:36 it should give fun. I think it should

51:38 give offer recommendations.

51:39 Okay. Um make the a AI optimizations

51:46 suggestions. I'm just going to go into

51:48 plan mode here just to make sure that

51:50 they

51:50 Oh, cool. There's mode. So that's new.

51:53 Yeah, cuz you kind of want that. You're

51:54 like, "Hey, before you go break all this

51:55 stuff,

51:56 let's Yeah, let's brainstorm."

51:57 I don't know if you saw this meme. It

51:59 was hilarious. It was like when you fix

52:00 a chip on a car, like on paint. And so

52:03 it had like this tiny little chip and

52:04 you'd put like tape on it and then you'd

52:06 you know white out the little chip so

52:08 that you know the white matches the

52:09 little thing and then you pull off your

52:11 tape and then the whole all of the paint

52:13 comes off and it was like fixing a bug

52:15 with uh with AI%.

52:18 This is what we spend most of our time

52:20 working around. uh make the optimization

52:23 suggestions about the actual offers um

52:28 and provide

52:31 concrete and actionable

52:34 uh advice

52:35 and I would say using the uh attraction

52:38 and upsell mechanisms inside the money

52:41 models book

52:41 uh and attraction mechanisms inside the

52:45 money models book

52:47 and I would say come with one

52:49 alternative yeah I say like come with

52:50 one alternative per offer in the flow

52:52 that could potentially improve it.

52:54 Come up with one alternative at least at

52:57 least one alternative

53:00 to one of the offers.

53:02 Sweet.

53:05 So while that loads, um, how do you see

53:08 where Replet fits within the universe of

53:12 lovable cursor? Like do you see Replet

53:15 going after a different avatar? Like how

53:16 do you how do you see where you guys

53:18 fit? you know, Replet is is more

53:20 expensive. Uh, and the reason it is is

53:22 because we're really trying to get you

53:23 to production. I think a lot of these AI

53:26 tools are really kind of toys and Replet

53:28 was a toy for a long time,

53:30 but there's a few things about Replet

53:32 that are really interesting.

53:33 Uh, for example,

53:35 the feature sets are is a lot more

53:38 complete. For example, we have a

53:40 database. You can create an actual

53:42 database and you can do migration the

53:44 database. uh here I'll just create a

53:47 database here you have a development

53:49 database and you can add a production

53:53 database as well so once I deploy this

53:56 will be so this you know it is a lot

53:59 more uh production complete environment

54:02 we also have um uh this deployment

54:06 interface is actually a full cloud

54:08 system so if you open the advanced

54:10 setting you can pick autoscale you can

54:12 pick a virtual machine now you don't

54:14 have to worry about that until later on.

54:16 But once you have a million users,

54:17 you're going to have to worry about it a

54:18 little bit. And you can ask the agent to

54:20 help you with that. You can change the

54:21 machine configuration.

54:22 So no other system out there, none has

54:26 that kind of depth of the technology has

54:29 a development environment, but also the

54:31 deployment environment,

54:32 also the database,

54:34 storage. It's like a full like cloud

54:36 system. Like we have app storage, we

54:38 have all of that stuff. So um to mirror

54:42 back the question you asked earlier

54:43 about like an example of you know an

54:45 agency or whatever using you know money

54:47 models or client finance acquisition in

54:48 order to cash flow their growth um what

54:51 are some I mean you don't have to name

54:53 names if you don't want to but like what

54:55 are what are some of the biggest apps

54:56 that have been built on replet? What are

54:58 their revenues?

54:59 Yeah. So um

55:01 can you see it?

55:02 Yeah. So there there's there's one they

55:05 tell us.

55:06 So we're actually we're going to get to

55:07 a point where we're going to start to

55:08 see it. That's another thing that makes

55:10 replet different is that we're building

55:12 a lot of services for the customers. So

55:14 for example like you know you don't have

55:16 to go set up off yourself. We can give

55:19 you off built into replet so you can

55:21 manage your users here. So we're going

55:22 to add a way to like charge your users.

55:24 So we're going to be able to have like a

55:26 macro view on how how much people are

55:28 making.

55:29 But a lot of the businesses that are

55:31 making a lot of money tend to be exactly

55:32 what you talked about which is

55:35 domain specific knowledge that no one

55:36 else has.

55:37 doesn't have

55:38 there's a European VC firm the CFO

55:43 he was using like a hund different

55:45 software systems like Quickbooks and all

55:47 these different things

55:48 uh all these finance tools in order to

55:50 manage their portfolio it's like you

55:52 know I want to centralize I know the the

55:54 business of VC it is not PE it is not a

55:57 b it is not like an internet business

55:59 and so I'm just going to go to replet

56:01 and build um a piece of software

56:04 that really embodies my my my my full

56:07 knowledge

56:08 before he quit his job he had millions

56:10 of ars and commitments

56:12 because he went to every CFO friend that

56:14 he gets dinner with anyways and says

56:15 like would you pay for this

56:17 so by the time he quit his job few weeks

56:19 in he had signed 5 million euro AR

56:22 um and so that that's I think an extreme

56:24 example of like really fast takeoff

56:26 success

56:27 uh there is um

56:29 it's clear product market fit which is

56:30 still the fundamentals still still

56:32 matter like you can build whatever you

56:33 want if no one wants it doesn't really

56:34 matter Um there there's this uh product

56:38 uh general artificial intelligence

56:40 proficiency institute geni

56:43 uh that's also one of the fastest uh

56:46 scaling uh businesses on replet um and

56:50 this entrepreneur uh John Cheney what uh

56:54 he did there was two kind of catalysts

56:56 for him by the way he's he's a guy that

56:58 went the Silicon Valley route and didn't

56:59 like it raised tens of millions of

57:01 dollars all that just didn't like it

57:03 he he's just like He wants more control.

57:05 He wants to I get it.

57:06 Yeah. He doesn't want to like manage.

57:08 Come over to my world.

57:08 Yeah. Exactly. Um and so, uh he gave

57:12 himself a challenge to build like a

57:14 business on LinkedIn to build like a

57:16 business and get the first customer like

57:17 in a week or something like that.

57:19 Uh and he was out to dinner uh with with

57:23 with their neighbor and um he asked his

57:27 neighbor uh who's who's selling

57:29 tractors. So he's like lives in uh

57:31 farmer land and he asked him how to use

57:34 chachi and I was like what is chach tea

57:36 right

57:36 and then he showed him what chachi is

57:38 and he like the next the day that guy is

57:40 using it for everything

57:42 I was like wow there's so many people

57:44 that just needs a little bit of

57:45 handholding so he created this

57:47 uh platform to have certifications live

57:50 courses

57:52 um live tests you know there's a test

57:54 here question answer

57:56 and he got I think he got to 200,000

58:00 in uh in revenue in the first two weeks.

58:03 I think he's closing in on a million

58:05 dollars a month right now.

58:07 Oh wow.

58:07 Uh and he sells to individuals, but he

58:10 also sells to enterprise. A lot of

58:12 enterprises just don't know how to

58:13 implement AI.

58:15 Speaking of which, what's headcount for

58:17 you guys right now?

58:18 Uh by the way, I forgot to to kind of

58:21 kick it off. So here's uh

58:23 um

58:24 it's a robust plan.

58:26 This sounds good, but I need you to make

58:29 the to prompt the AI to do this. Create

58:35 a plan

58:37 to do that. Um, yeah, it's like, oh,

58:41 here's all the suggestions I have. Well,

58:42 no, I want you to prompt AI in the app

58:45 to do that. Um yeah so headcount uh for

58:48 replet um last 2023 24 were actually

58:52 very tough for us because we had built

58:54 all the so the reason our platform is so

58:57 deep in technology right like it has the

58:59 virtual machines the APIs AI all of that

59:01 is because I spent 10 years trying to

59:03 build this

59:05 uh and most of my time was not spent on

59:07 AI most of my time was spent on the

59:08 platform

59:10 I was like if you want to make it so

59:11 that anyone can make software yeah

59:13 you need to make a consumer grade cloud

59:15 system. No one will know how to use AWS

59:17 if you don't have a coding background.

59:19 So, we basically built like like

59:21 consumer grade AWS.

59:23 And then in 2024, we weren't making a

59:25 lot of money. We're burning a lot of

59:26 money. The team was really huge. We had

59:28 to lay off half the team.

59:30 Oh, wow.

59:31 And I told the team, look, we we just

59:34 have to solve coding.

59:36 At that time, there was no lovable bolt,

59:38 none of that. There was it was not a

59:40 market.

59:40 And I said, we just have to go solve

59:42 coding. have to go build an agent that

59:43 can do the coding for you. So we spent

59:46 six six months built that we spent six

59:48 months struggling with it because the

59:52 the open eyes of the world still haven't

59:54 really started optimizing for agents.

59:56 Really felt like we're going to get

59:58 there and actually what's interesting is

60:00 replic catalyzed that. So we launched in

60:02 September the system was very rough and

60:05 bare bones and all that

60:07 but it inspired a lot of people

60:09 including the AI researchers. It's like

60:10 huh this is possible. Yeah,

60:12 it's possible to have an agent not just

60:13 write the code but create the database,

60:15 run the SQL migration and do the

60:17 deployments.

60:18 Okay, let's go optimize against that.

60:20 And so it is sort of this feedback loop

60:22 where we inspired the researchers to go

60:24 optimize for our thing. And then a bunch

60:26 of uh you know competitors and copycats

60:29 came in and of course that's what

60:30 happens. Um and so everything uh

60:36 like you know we went from you know two

60:38 three million to like 10 million in by

60:40 the end of the year.

60:42 Uh and still there was disbelief inside

60:44 the company. Actually people continue to

60:45 leave uh after they you know you

60:49 mentioned when you get a down round that

60:50 kind of destroys the dream. Same thing

60:52 with a layoff. It really

60:53 Oh totally. It it really kind of

60:54 destroys that dream that people were

60:56 like living through. And there was this

60:58 belief that we're going to make it

61:00 despite like having a revolutionary

61:01 product. It's kind of really insane like

61:03 how human psychology is.

61:05 But the revenue kept going up. The

61:06 feedback we're getting is is

61:08 where did the big was there a big

61:10 inflection point at some point?

61:11 The big there two inflection points. Um

61:15 in uh in February we uh got out of beta

61:22 and the and the and the product started

61:24 being more stable and also we

61:26 reintroduced our mobile app and now a

61:28 lot of people started coding from their

61:29 phones. So that increased the the

61:31 market.

61:32 Uh so

61:33 I did that too.

61:33 That was because I could watch I could

61:35 like watch TV at night and just like

61:36 just give it the prompt and it comes

61:37 back in five minutes and then you just

61:39 give the next prompt.

61:39 We know our audience that they're busy.

61:41 They don't want to be in front of the

61:42 computer all the time. So that was one

61:44 and then April was version two. So

61:48 version two was a lot more stable. Uh it

61:51 started actually working and that and

61:53 people started spending a lot more on

61:54 replet and so that scaled our revenue.

61:57 Okay. With tokens and all that.

61:58 Yeah. Do you think um can replet build

62:00 something like a CRM?

62:02 Yeah. Yeah. A lot of people build CRM.

62:04 Like there's a private equity guy that I

62:05 saw. He wrote a story the other day

62:07 where he's like you know there's the

62:09 hotspots of the world. They're all

62:10 great, but I need like a private equity

62:13 CRM

62:14 CRM. And so there there's a lot of

62:16 and the market is so small and because

62:18 they don't do transactions, no one's

62:19 really like wanted to go after 1,000

62:21 customers that and they're not really

62:23 going to say, "Oh, like

62:25 for the economics to work with a

62:26 thousand customers, you got to be

62:27 charging like a million dollars a year."

62:28 And they're just not going to justify it

62:30 because they're like, "We're only going

62:31 to make eight eight investments or

62:32 whatever it is." But you still want it.

62:34 Yeah. So you end up like for us we had

62:36 to we basically retrofitted you know a

62:38 HubSpot or Salesforce whatever like out

62:40 off the off the gate just made it look

62:42 like all of our deal stage pipelines

62:44 just worked like enterprise sales but we

62:46 just had to have like more steps of like

62:47 diligence one diligence two team

62:49 diligence you know had to work work work

62:51 our deals through that pipeline um but

62:53 yeah no I mean it makes sense because

62:54 because then once once you're there it's

62:56 like what does customer success look

62:57 like

62:58 when you make an investment

62:59 right

63:00 and see it's a little bit of paradigm

63:01 shift but you know some of the some of

63:02 the the fundamentals still apply

63:04 Yeah, it's sort of like you go from

63:08 you go from like a world of like large

63:12 malls and kind of that that try to cater

63:15 to everyone or large food chains to

63:18 local uh food

63:20 local coffee shops. You Starbucks is

63:22 great

63:23 and it's going to be okay for most

63:25 people.

63:26 But uh so now it's booting up the app.

63:28 It built the feature. spinning out the

63:30 app in a in a in a browser window so it

63:33 can save us from the QA. That's great.

63:36 So, it's interesting what you bring up

63:37 though because there's this this uh it's

63:39 kind of like an accordion effect that's

63:40 really common in business is

63:42 things just do like this. They like they

63:43 centralize and they decentralize and

63:45 they centralize and they decentralize as

63:46 these new layers of kind of the

63:47 internet, you know, first is like the

63:49 internet came out and then there was all

63:51 these different forums and blogs and

63:52 then the issue becomes finding them. So,

63:53 then the search, you know, search

63:55 becomes this this big.

63:56 It looks like it it implemented it. All

63:58 right, let's see what we got.

63:59 You could tell

64:00 when you're high impact. They even rated

64:02 it high impact. What we got?

64:04 Or is it still testing it right?

64:05 It's still testing it, but but it's good

64:07 to to

64:08 to kind of look at it when it's testing

64:10 it. And you can read its thoughts and

64:12 you can read what it you can look at

64:13 what it's looking at.

64:14 Uh so that's its eyes, this like purple

64:17 thing. And

64:18 okay,

64:18 yeah,

64:19 I've successfully triggered AI analysis.

64:21 It looks like it's generating some

64:22 helpful suggestions. You know, it's like

64:24 I'm going to click reanalyze and see if

64:26 it gives me another thing.

64:28 Did you click them or Oh, it said that

64:30 it's like a robot, right? And this is

64:32 this is what we're talking about.

64:34 The action agents now. We're going to

64:36 the world where they can use the

64:37 computer

64:38 and we can just sit back and eat popcorn

64:40 to look at it.

64:41 Oh, good move. You know,

64:43 but and now it's restarting the server.

64:46 I think it it it passed the test. So,

64:48 you can see a check mark. And so, it's

64:51 going to pass back to the main agent and

64:53 says, "Okay, I tested it for you." Uh,

64:55 and and it passed. and it'll yield

64:57 control to back to us in a second. It's

64:59 interesting that you went to the

65:00 multi-agent approach because um you know

65:02 with the the STR thing, one of the

65:04 things that we were considering was um

65:06 instead of having an agent that was

65:08 master at this you know conversation

65:11 especially from especially from a text

65:12 perspective not voice is still has some

65:14 latency which is a pain but like just

65:15 from a texting perspective um it it

65:18 started to make more sense to have like

65:20 an AI that was only trained on first

65:21 message

65:22 and another AI that was trained only on

65:24 second message and like the more

65:25 specialized we can make them like and

65:27 all of them are optimized around a

65:28 singular goal which is how do I increase

65:29 the likelihood that someone responds as

65:31 the like you're not going to get someone

65:32 booked on the first message. So then we

65:33 need to optimize for a different goal.

65:34 We have to get them to respond.

65:35 One thing to try there is

65:38 you might actually find that Gemini is

65:40 better at the first one. Yeah.

65:41 Open is better the second one.

65:43 Interesting.

65:43 And even GT5 is better the third one. So

65:46 we use every model on the market

65:48 including the open source models

65:50 because like you know for example G G5

65:53 yeah

65:53 it's actually very kind of anal and

65:55 annoying

65:56 and and it's just like so it's the

65:58 architect

65:59 in this case it'll it'll be like it

66:02 looks like cla is like that sucks and

66:04 claude tries to negotiate with it. It's

66:06 like really funny to look at it. So you

66:08 end up like that's why sub agents and

66:11 and multi- aent system

66:14 that's another advantage you can have

66:16 over open AI open AI can only use open

66:18 AI models

66:19 app layer companies can use all of the

66:21 models available including the Chinese

66:23 models

66:23 right yeah

66:25 um so I I think the multi- aent approach

66:26 is the future

66:27 what do you think about local because I

66:29 know you know GPT came out with their

66:31 our openi came out with their like

66:32 locally hosted version like how do you

66:35 see how do you see that interaction with

66:37 normal mainstream business.

66:39 I don't see much of an interaction with

66:41 it. I think that

66:42 people might still use open source but

66:44 they will not host it themselves.

66:45 Okay.

66:46 Partly because there's so many good

66:47 inference systems. You've heard of Grock

66:49 with a Q.

66:50 So Gro with a Q is this um inference uh

66:54 company that hosts open source models

66:56 for you like Llama like things like

66:58 that. Another one is Sirius. It's about

67:00 to go public. So that there's a bunch of

67:02 these companies that they will do a much

67:04 better that job than you at hosting

67:06 these open source. So I think open

67:08 source has a place and it's mostly

67:10 because of cost, maybe because of

67:12 fine-tuning,

67:13 but the idea that I'm going to download

67:15 a model hosted on my server inside the

67:16 company, I don't think it's going to

67:17 happen.

67:18 Interesting. This is where like the the

67:20 training becomes so important. And

67:22 that's like with ACQA has so many of my

67:24 inferences and so many times where I'm

67:25 like

67:26 the the thing that has been hardest for

67:28 me to replicate and it'll just take time

67:30 and iteration is like I have a very good

67:32 idea of benchmarks across hundreds of

67:35 industries of like what would uh an

67:38 in-person close rate for a plumber be?

67:42 And is it low or high based on where

67:44 they currently are? And

67:46 based on that number, is that the

67:47 constraint of the business right now or

67:48 is it something else? Or is it just

67:50 their lead cost is too high? because

67:51 like CAC being high could be one of six

67:52 different problems. And so being able to

67:54 walk through that decision tree and then

67:55 use benchmarks as really quick litmus

67:58 test or triaging the answers is what

68:00 I've basically spent my whole career

68:02 doing.

68:02 Um and trying to transfer as many of

68:04 those things over because I've just had

68:05 so many experiences and repetitions and

68:07 like a lot of that data is not public.

68:10 Like no one know like no there's I don't

68:12 know any blogs that are like these are

68:14 the average conversion rates for

68:15 plumbers. And also that's different than

68:17 roofers which is different than

68:18 doortodoor which is different than

68:19 Google search which is different than

68:21 meta ads. And so like if I have a a pest

68:23 control company that comes in you know

68:24 for for our headquarters and they're

68:26 like hey we do door to door we have 100

68:28 sales guys who deploy them to every new

68:29 market. This is how we you know uh

68:31 canvas an area and then and and deploy

68:32 it or build our build our subscription

68:34 base. It's like okay we want to have

68:36 some sort of uh coverage after we move

68:39 away from the market to at least

68:40 maintain above churn so we can just keep

68:42 that market being you know profit or

68:43 whatever. And so it's like, okay, well,

68:44 we could we have a Google search

68:46 strategy, just AdWords, and then we have

68:48 a meta strategy. And I'll have to walk

68:50 them through the fact that like, yeah,

68:51 you might be able to convert 50% of your

68:53 Google search leads because intent's

68:54 super high. Now, your lead cost will be

68:56 10 times that of your meta leads, but

68:58 your metal leads uh will be onetenth the

69:02 price and you're going to convert 10%.

69:05 But they'll come in and say like, meta

69:07 meta doesn't work for us because we're

69:08 only converting like 5% of leads. It's

69:10 like, okay, well, slow down. you only

69:12 need to double this for this to be

69:13 effective and the ocean of traffic

69:15 that's there is so much higher because

69:17 intent there's only so many people

69:19 searching once you once you max that out

69:20 in a given market that's it there's

69:22 nothing out there's only so many people

69:23 searching but from an interruption based

69:26 you know marketing perspective it's like

69:27 you open up the entire market because

69:28 you can run Tik Tok you can run YouTube

69:30 you can run Facebook Instagram whatever

69:32 and so but that nuance of knowing I mean

69:35 this is what I mean this is what I've

69:37 spent my whole career doing

69:38 what you just said and it will learn

69:39 what you just said is is a is a prompt

69:41 essentially you need to structure it a

69:42 little different.

69:43 You know, I I'm you know, the other day

69:45 I was trying to come up with this

69:47 analogy of,

69:48 you know, how the Airbnb guys looked at

69:51 an unmonetized asset, which is a bedroom

69:54 in their

69:55 uh and unused bedroom. Yeah. Um unus

69:59 capacity. The cloud,

70:01 what basos did with AWS is also thing

70:04 same thing. You know,

70:05 Amazon needs all these servers to handle

70:07 Black Friday, but most of the year they

70:09 don't need any of them.

70:10 Yeah. Uh and so access capacity you can

70:12 monetize. I think domain knowledge is

70:15 the same way right now.

70:17 Yeah, it's a really great perspective.

70:18 Yeah. And so I I think people watching

70:21 should like really uh introspect and I

70:23 bet you have a domain knowledge that

70:25 perhaps

70:26 super valuable

70:26 super valuable and very few people have

70:29 and that is a prompt that you can embed

70:31 into an AI

70:33 and now a lot more people can

70:34 you can you could be consultant at scale

70:37 essentially. I mean and that's that's

70:38 what we're I mean that's what we're

70:39 trying to build here you know is how can

70:41 we help businesses grow and we do we

70:43 have developed kind of like our very

70:46 like it's a decision tree that has

70:48 probably like 20 paths and then within

70:50 those path there's another six more path

70:53 you know so there's it's just a massive

70:54 decision tree but fundamentally like

70:55 but it's finite

70:56 but it exactly it's not infinite like

70:57 there's only I mean because you can

71:00 always chunk it all the way back up

71:01 which is like we either need cheaper

71:02 leads we have to convert more of them

71:03 we're mispriced right or we have a cash

71:05 flow issue or we have a supply issue

71:07 like we can't deliver on whatever it is

71:08 that we sell. There's the issues, right?

71:10 Right. Then each of those has trees

71:12 underneath of them, but like that's what

71:13 it is. And so figuring out having the

71:16 larger that's why the context layer was

71:17 step one for us of like building this.

71:19 So it's like, okay, well, at least we

71:20 have these. Now part of the issue that

71:22 comes into it is a lot of people who are

71:23 smaller businesses don't have good data.

71:25 And so then that then you have to start

71:27 asking follow-up questions that are you

71:28 have to be

71:29 in exactly questions about it. Yeah.

71:32 And even sometimes it's like I don't

71:34 know what my CAC is. So we have to say

71:35 like okay well you know what was your

71:37 marketing spend last year? How many

71:38 customers did you get? Okay. Well we can

71:40 back into CAC that way. It'll be broader

71:42 and it's not going to be segmented by

71:43 channel. But

71:44 how much I have a question. Um how much

71:48 of of CAC could be um you're getting the

71:53 creatives wrong. you're like how much is

71:56 CAC is set in stone versus like you're

71:58 actually doing something wrong on

72:00 because you know my team tells me right

72:02 now is like Google ads is like on on

72:05 rails right now

72:06 like is that true like you can't be

72:08 creative about the way you present these

72:10 things

72:11 I mean you totally can be yeah I mean

72:13 yeah 100% I mean you're bidding for

72:15 keywords but you still have all the

72:16 headlines and the copy that's going to

72:17 go there and you have endless variations

72:18 of headlines copy

72:19 and how important is that to k

72:20 super yeah

72:21 super important

72:22 is the difference between $10 CAC and

72:24 $100 CAC.

72:26 Um I would that's a that would be a big

72:28 one, but you could definitely get

72:29 doubles, triples, sometimes 5xs. 10x is

72:31 a big one.

72:32 Um so for that kind of thing, the

72:33 difference would probably be closer to

72:34 offer than copy.

72:35 Mhm.

72:36 But the copy describes the offer. So

72:38 these are, you know, kind of

72:38 interrelated. Um and with keywords uh

72:42 and search, it's going to be it's text

72:43 based predominantly.

72:45 Uh whereas where the creative will

72:46 absolutely can have 10x 100x

72:48 differences, it's going to be on video

72:50 based, image based. That's where you can

72:52 have the huge alphas. And where where

72:54 this gets really interesting is when

72:55 companies are trying to scale their

72:56 advertising. So like school, we spend,

72:58 you know, several hundred,000 a day in

73:00 in advertising.

73:01 Um, and what's what's what's interesting

73:05 about this is that I was having a

73:07 conversation yesterday with um a friend.

73:09 He was like, I can't scale this offer

73:10 past 2,000 a day profitably. As soon as

73:12 I get above 2,000 a day, it stops

73:14 working. M

73:15 and so I was explaining to him I was

73:16 like well right now the way that you

73:18 have the offer and the copy structured

73:20 is targeting a very aware customer and

73:24 so you know there's stages of awareness

73:25 Eugene Schwarz old school direct

73:26 response marketer he talked about five

73:28 stages awareness you've got unaware uh

73:30 problem aware solution aware product

73:31 aware and then most aware which is just

73:32 your existing customers and so your most

73:35 aware customers and product aware

73:37 customers you can simply just make an

73:38 offer and they will buy but if you just

73:40 say here's my offer here's my offer

73:41 you're not going to be able to scale to

73:43 mom who's across Ross the world, she's

73:46 not going to she's not going to respond

73:47 to that. And so CS can get really high.

73:49 And so it's kind of having this layered

73:51 approach from an advertising perspective

73:52 where up to 2,000 a day just

73:54 demonstrates that there's at least with

73:56 your existing creative um you can only

73:58 really reach whatever 2,000 of let's say

74:01 it's $20 CPMs. It's like you can reach a

74:03 million people a day that um that will

74:07 respond to that level of advertising.

74:08 Now there'd be a separate campaign that

74:09 then hits probably a 10 times bigger

74:11 trunch when you go to just uh you know

74:13 solution aware right and it's like okay

74:16 uh what are the different you know um I

74:18 don't have you have the product in mind

74:20 but like uh if you're trying to lose

74:21 weight it's it's like five things five

74:24 curious habits that um are are are

74:27 wrecking your you know energy

74:29 that might be at the most unaware level

74:32 because I that could lead to a pill that

74:35 could lead to a workout that could lead

74:36 to a gym

74:37 because it's just pure curiosity when

74:39 you're at the top of the funnel, right?

74:41 And as

74:41 that's what you should aim for.

74:43 Well, if you can get there because as

74:44 you go up the funnel, you go to I kind

74:47 of think of like watermelon seeds where

74:49 like at the base at the base of the the

74:50 triangle, you have lots of seeds. So,

74:52 there's lots of high likelihood

74:53 conversions and the the platforms are

74:55 optimized to get you those conversions,

74:56 right?

74:57 As you go to,

74:58 you know, thicker and thicker slices,

75:00 the seeds are more disperate,

75:02 right? And so, it costs you more because

75:04 it takes more eyeballs to reach another

75:06 seed.

75:06 Yeah. But if the money model is right

75:09 and the creative is structured so that

75:11 you can attract that person and then

75:13 move them through the process through

75:15 education, indoctrination, whatever you

75:16 want to do, um that's where you get

75:18 these these funnels that can overnight

75:20 go from like zero to a few hundred,000 a

75:22 day and spend a million a day and spend

75:24 um and do it profitably. Obviously

75:26 payback period is also a factor of that

75:28 because I you know I design all my stuff

75:30 so that we can do it in 30 days but if

75:31 we can do it in six months I can spell I

75:33 can spend a hell of a lot more right as

75:35 long as you have revenue retention. And

75:36 so um but that's that's typically when

75:39 we have a CAC issue it's going to be

75:41 offer driven

75:43 you know number one like number two

75:45 would be it would be a creative issue

75:46 which can be a factor of just not making

75:48 enough creative which is really the

75:51 symptom or sorry the cause of not good

75:53 enough creative. Mhm.

75:54 You have to make enough to be good

75:55 enough because you just need more

75:56 variations, right? And so then you have

75:59 better creative overall. Um there's

76:01 obvious CRO within the funnel, like does

76:03 the page load quickly? Um is the is the

76:05 CTA above the fold? Like simple things,

76:07 but people miss them, right? I I had a

76:09 business the other day that came in that

76:10 was in the dating space.

76:12 Um and he had this whole funnel.

76:15 I was doing three million bucks a year

76:16 or something like that. And I went

76:18 through it with him and his his like his

76:22 final call to action to get them

76:23 scheduled for a call. It had a video and

76:26 the scheduleuler was at the bottom, but

76:27 they had just watched a video. They just

76:29 watched like a 30-minute video to get

76:30 that pitch to schedule. And he said,

76:32 "Watch this video before you book." And

76:34 I'm like, "They just watched a 30inut

76:36 video? Like, let them book." And it was

76:38 crazy cuz I I told him, I said, "As as

76:39 silly as what I'm going to say sounds,

76:41 I'm going to give you five different

76:42 things to do today." I was like, "But

76:43 the one that could triple your business

76:44 is going to take 5 seconds." Right? It's

76:46 just like it was just below the fold and

76:48 and I I'll bet people just didn't see

76:50 it.

76:50 And so sometimes it's tiny hinges like

76:52 that that make huge swings on the doors.

76:54 Um and so you know I I think of this as

76:57 like I have this gigantic bag of tricks,

76:59 right? And it's just this gigantic

77:00 checklist that you just learned through

77:02 doing this.

77:02 I wonder if we can we can put like

77:06 Hermosi bot in in like a

77:08 in a funnel like it could crawl a funnel

77:10 and then

77:10 Yeah. And it can look at the visual.

77:12 See, that would be a very valuable thing

77:14 if you had a bot that could crawl a

77:17 funnel. Yeah.

77:17 And then make suggestions on like kind

77:20 of like almost like like it takes a

77:22 picture of each page. Exactly.

77:23 And then just doodles. Yeah. And then

77:25 just says like

77:26 I think you have an offer issue here.

77:28 And it could crawl your ads library page

77:30 and say like oh your ads the the hooks

77:32 are off here or the hook isn't congruent

77:34 with the headline. Yeah.

77:35 And so again there's all these different

77:36 check marks that you have to go through.

77:37 Yeah. So when we publish here, uh we

77:40 actually let you uh do a security scan

77:44 um before before you publish

77:46 uh and um uh I don't know where it is

77:51 important.

77:52 Yeah. But I I I wonder if we can also

77:55 have funnel scan. So maybe this

77:58 something and we on the security scan,

77:59 we didn't build it ourselves.

78:01 Yeah.

78:01 Uh we collaborate with a company called

78:03 SDRP to to to run the security scan. So,

78:06 I wonder if we collaborate with you to

78:09 have like a funnel scan.

78:10 Yeah. No, that would be cool.

78:12 Yeah, that would be awesome.

78:13 It'd be an amazing uh an incredible lead

78:15 magnet.

78:16 Yeah.

78:17 Yeah. Yeah, because I think what so my

78:18 opinion when we're talking about the

78:19 software stuff, I see one of the big

78:20 opportunities is the amount of free

78:23 tools that people will start being able

78:24 to give away because I sure there's ways

78:26 that we could, you know, charge for

78:27 subscriptions, but like

78:29 a funnel scan for example is probably

78:32 not a tool that you would use on a on a

78:34 on a consistent basis. You probably use

78:36 it one or two times. You give it a huge

78:38 amount of data to get an inference. Mhm.

78:41 And so those types of products I don't I

78:44 don't those aren't the types of things I

78:46 would want to sell.

78:47 Those are the types of things I would I

78:48 would want to give away in exchange for

78:49 tons of data.

78:50 So that then we could, you know, sell

78:51 them other stuff that that we'd find.

78:53 Okay. Well, now that you've optimized

78:54 your funnel, what about your traffic,

78:55 you know, like and that's something that

78:57 would have a more consistent because

78:59 once you have the the funnel down, it's

79:00 kind of this asset that you build and

79:02 once it's there, it's like you don't

79:02 really you can always tweak it, but like

79:04 n you know, 8020 is already there and so

79:07 the the iterative process will be less

79:09 so whereas the iterations you're going

79:10 to make on traffic uh are going to be so

79:12 much uh more consistent and higher

79:14 volume that that's where you'd probably

79:15 put the more subscription related

79:16 product.

79:17 Right. Right.

79:17 Yeah. But I think this is just

79:18 interesting from a business perspective

79:19 of thinking like

79:20 which of these different problems can we

79:22 solve? what is the nature of the problem

79:23 that's being solved? How how consistent

79:25 is that problem need to be attacked? And

79:26 then that kind of separates your free

79:28 level from your paid levels and whatnot.

79:30 Yeah. The the thing about AI that that

79:32 is tough right now for us especially

79:35 is how expensive it can be.

79:37 Um and so like a herozi funnel scan bot

79:41 um might actually end up being like a

79:43 couple bucks,

79:44 three bucks, something like that.

79:46 Yeah.

79:46 So yeah, I just

79:48 like our AI costs probably a million

79:49 bucks a year to run just in server cost.

79:51 Yeah. So like the calculation on free is

79:54 cuz like you know software systems

79:56 Silicon Valley internet businesses for a

79:58 long time the marginal cost is zero.

80:00 Yeah. Now it's not that

80:01 it's not zero. Now it might be two three

80:03 four $5 sometimes more.

80:06 Uh so it's changing. I mean I'm hoping

80:09 that a lot of the Chinese models and

80:11 things like that that are coming on the

80:13 scene

80:13 that are like good enough to run some of

80:15 these free experiences and upsell them

80:16 on the like more advanced AI.

80:18 Yeah. But yeah, that changes how I I'm

80:21 thinking about because I I I scaled

80:23 rapidly to 40 million users, but just

80:26 like

80:27 really optimizing the heck out of the

80:29 free experience and just like making it

80:31 so easy.

80:32 Oh, it was great. I mean, the upsell was

80:33 so so easy cuz you you get your first 10

80:36 prompts in and then it's like, hey, used

80:37 up your tokens and you're like, damn,

80:39 now I'm halfway in, you know, I want to

80:40 finish this thing. No, it's great. It's

80:42 super good.

80:43 It was seamless. I paid for annual.

80:44 Oh, awesome. Well, we did your CA.

80:47 Yeah. Hopefully we'll get you back.

80:48 We'll get you back to try it again.

80:50 I have to get my my uh my fitness app to

80:52 to to work on my phone.

80:54 Agent 3 will be able to refactor it and

80:56 test it and I'll trust agent 3.

80:57 Yeah. Okay. Great. Great.

80:59 Appreciate you.

80:59 Awesome. Yeah, this is this was great.

81:01 This is really a master class. I think

81:02 I'm really excited because as Replet is

81:04 moving to support a lot more

81:06 entrepreneurs like the ones we talked

81:08 about, this this this stuff is is a gold

81:10 mine. So, really thank you.

81:11 No, thank you, man. I appreciate.

81:12 Congrats on the success.

81:13 Thank you.

81:14 Yeah. Remember us little people.

81:17 You know, goes both ways.

81:26 At Replet, we believe everyone should be

81:28 able to create software without limits

81:30 and without knowing how to code. Our

81:33 vision was never to just build a tool.

81:35 We set out to create a teammate, a

81:37 technical partner that runs longer,

81:40 thinks smarter, and works like a

81:42 builder. An autonomous agent for

81:44 everyone.

81:46 Previous agents, while great in many

81:48 respects, were far from autonomous. They

81:50 needed constant attention and

81:52 handholding.

81:54 That was yesterday.

81:56 Today, we're proud to announce a major

81:58 breakthrough. It's a moment we've been

82:00 building towards since day one,

82:02 introducing Agent 3. Agent 3 delivers a

82:06 10x increase in autonomy, redefining the

82:08 user experience from the ground up. That

82:11 means more progress, less micromanaging,

82:14 and more time for your ideas. Let's take

82:17 a closer look at what makes Asian 3 so

82:19 powerful. Asian 3 runs on its own for

82:22 hours, handling full tasks autonomously.

82:26 We went from 2 minutes to 20 minutes to

82:29 200 minutes of autonomy. You can track

82:32 your project's progress in real time

82:35 from anywhere with live monitoring right

82:37 on your phone.

82:39 Asian 3 can now test and fix its code,

82:42 constantly improving your app behind the

82:44 scenes. We invented our own validation

82:47 system to speed up quality control,

82:49 making it at least three times faster

82:51 and 10 times more cost effective. It

82:53 checks every button, every API, and

82:56 every data source, ensuring every part

82:59 of your software works. For the first

83:01 time ever, Agent 3 can build other

83:03 agents. agents that can automate complex

83:07 workflows using natural language and

83:10 connect to your data sources and

83:11 services, allowing you and your users to

83:14 interact with your own agent.

83:18 Agent 3 isn't just an upgrade. These new

83:21 capabilities along with everything else

83:23 Agent 3 brings will unlock entirely new

83:26 frontiers, redefining what's possible.

83:29 We're so excited about what Agent 3 can

83:32 do and we can't wait to see what you're

83:34 going to build with it.

83:36 [Music]

You have ... regenerations remaining for this video.

Length

Short (100 words) Medium (300 words) Long (500+ words)

Tone

Casual Professional Conversational Persuasive

Focus

Overview Key Points Actionable Takeaways

Language Style

Simple Standard Detailed

Additional Notes (Optional)

500 characters remaining

Save these preferences for future summaries

Length

Short (~300 words - quick blog post or Substack snippet) Medium (~600 words - standard blog post) Long (~1,000+ words - in-depth article or feature)

Tone

Casual (friendly, informal) Professional (formal, business-like) Conversational (chatty, engaging) Persuasive (call-to-action focused)

Keyword Emphasis and SEO Optimization

These keywords will be used to optimize the article for search engines

Auto-generate meta title (50-60 characters) Auto-generate meta description (150-160 characters)

Content Focus

Full Overview (use entire transcript) Actionable Takeaways (focus on practical advice or steps)

Additional Notes (Optional)

500 characters remaining

Save these preferences for future blog posts