YT YouTube2Text

YouTube2Text → Transcripts → Как построить бизнес с Chat GPT? Все промты в видео! Артем Круглов

Как построить бизнес с Chat GPT? Все промты в видео! Артем Круглов — Transcript

by Николай Хлебинский | AI Action Plan · 16,756 words · 2,683 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Good afternoon. My name is Nikolai
0:01Khlebinsky. You are on the Action Plan
0:03channel. Today we are discussing a
0:04million-dollar idea. What is going to
0:05happen today? Today's episode is
0:07different; it's not like our usual ones
0:09. Usually, we sit and talk, extracting
0:12the entrepreneur's experience from
0:14their past decisions, projects,
0:16failures, successes, and achievements.
0:19Today, we decided to do things
0:21completely differently. And today, we
0:23want to conduct an experiment. We want
0:32to take all the achievements we can
0:34reach from every corner of the world,
0:36from any point in our country, and use
0:38these advancements in artificial
0:39intelligence to create a business from
0:41scratch right here and now, taking
0:43those first steps that will shorten our
0:45action plan to a million rubles from
0:47the several years it used to take to
0:49start such a business. Down to, well,
0:56however long we sit here—an hour, an
0:58hour and a half, or two—chatting,
1:00drinking coffee, and making something
1:02happen. Let's see what comes of this.
1:04This message is for those who have just
1:06started watching this video. We have
1:08already been recording for about an
1:10hour and a half, and we've done things
1:12here that, well, please watch it from
1:14beginning to end. If your goal is to
1:18launch your own IT or internet-related
1:20business, we've created so much here
1:23that I am, honestly, in shock. My guest
1:26today is Artyom Kruglov, founder of the
1:29company Anquy. Artyom, welcome.
1:31Yes, hello.
1:32I realized that the world has changed.
1:34It will never be the same again.
1:36With artificial intelligence, you don't
1:38just work better; you can do the
1:40impossible. We are just a few years
1:42away from the moment when a company
1:44worth over a billion dollars—a
1:46unicorn—consisting of only one person
1:48, will appear in the world.
1:49They come and ask: "Artyom, how do I
1:51get into artificial intelligence?"
1:53The phrase: "Artificial intelligence is
1:55marketing. No such
1:56intelligence exists." This task used to
1:59take months, and now it takes seconds.
2:03If you find something trendy and
2:05high-margin before anyone else, you win
2:07. What do you do? Give us the numbers
2:15that characterize your scale, and
2:16anything else you think the audience
2:18should know about you.
2:19I am 38 years old. I live in Moscow,
2:21I’m married, and I have a wonderful
2:24son. And we, in fact, help tens of
2:27millions of people make successful
2:30purchases online. Every time you search
2:34for something in Gold Apple, Samokat,
2:3712Storeez, or TSUM, the product
2:39selections in the search and the
2:40product recommendations are generated
2:43by us; we provide independent retailers
2:45with great algorithms for search and
2:47personalization. So, guys, for those of
2:52you who aren't familiar with marketing
2:54automation tools, here's the deal.
2:59Artyom's company, Iquii, which, by the
3:01way, was just sold to Tinkoff, is one
3:04of the rare success stories of building
3:06a tech startup and achieving a
3:08successful exit in Russia. They focus
3:17on providing retailers with AI-driven
3:19technologies to help boost conversion
3:21rates, increase average order value,
3:23grow sales, improve customer lifetime
3:26value, and so on. Did I get that right?
3:31Yeah? Just one correction: we have a
3:34strategic partnership with the Tinkoff
3:36team, they hold control, but we
3:38continue to manage the company and
3:40remain active founders in the business.
3:44I think it's important to share the
3:46thought that the path we took was
3:47incredibly difficult. I mean, when we
3:50started, there weren't even developers
3:52who knew the specific technologies we
3:55needed. Whereas now, you can build a
3:58prototype in just a day or two and
4:00already start trying to sign your first
4:02contracts. Right now, I see that
4:05there's another boom in California of
4:07solo entrepreneurs—people who have
4:09built products on their own and are
4:11already running them with the help of
4:13agents. And yes, at this point, one
4:16might say, "Well, of course, if you
4:18graduated from the MSU Physics
4:20Department, maybe you can figure this
4:22stuff out." But I think the most
4:26advantageous, the coolest mindset you
4:28can have right now is that with AI, you
4:31don't just work better—you can do the
4:33impossible. The things you were afraid
4:38to do, the things people always told
4:40you were impossible—now they’ve
4:42become possible. That’s the message I
4:46want to convey to the podcast viewers:
4:48take a look at this and find the
4:50courage to do something impossible.
4:53That sounds really cool. Honestly,
4:55I’ve seen this happen a couple of
4:57times—or maybe more than a couple—
4:59and I’ve encountered AI use cases
5:01myself where I thought, "Wow, that's
5:03incredible." We are currently at the
5:06beginning of this recording. I hope
5:08we’ll have that same "wow" effect by
5:10the end. So, make sure to watch until
5:12the very end. We’re about to show you
5:16what AI is capable of today. I’ll
5:21jump in here again to dispel all the
5:22myths right away.
5:25Uh-huh. "Artificial Intelligence" is
5:28just marketing. There is no
5:30such thing as intelligence. These are
5:32technologies from the end of the last
5:34century. Everything that hasn't really
5:37changed, but what has changed in the
5:38world since the end of the last century
5:40, has become much cheaper. Computing
5:44power. What used to not even fit in
5:48this room to calculate a flight to the
5:50Moon, today fits in our pocket, in a
5:52phone, on a much larger scale. And all
5:56these things that are usually
5:58classified as artificial intelligence
6:00technologies, they are very
6:01resource-intensive. They require very
6:04large datasets. They require processing
6:08those datasets with a large number of
6:10machines. A very large number of
6:13machines. There used to be another
6:15buzzword called "big data." What is big
6:17data? Well, in my understanding, it’s
6:19a volume of data that doesn't fit into
6:21the RAM of a single computer. We have a
6:23certain number of RAM sticks in our
6:24laptop. Here we have, say, 32, 64, or
6:28128 GB of RAM. You can feed it 128 GB
6:31of data, but it won't take any more.
6:35Therefore, in the late 2000s, very
6:37accessible distributed computing
6:39systems appeared that could break down
6:41this data and combine computers into a
6:43cluster that could exchange information
6:46and calculate it together. And now it's
6:50just available right in the browser.
6:52That is, 10-15 years ago, when Artyom
6:54and I were starting, we needed very
6:56expensive engineers, just a few who
6:58actually understood how to connect all
7:01those 300 servers. These were the
7:03cutting edge of science, if you will,
7:05in the field of Data Science,
7:07right? These were the very best,
7:09right? So now any student can sit down
7:12and engage in what is already a huge
7:14trend called "vibe-coding,"
7:17right? When you sit down in a cafe, get
7:22a coffee—I got one downstairs where
7:24we're sitting—and over coffee and an
7:26interesting conversation, we whip
7:28something up, and that something will
7:30be at the level of several years of
7:32work by a highly experienced team. Just
7:36like that.
7:37It feels like now is the most unique
7:39opportunity to get into this. I mean,
7:41you and I know very well what trends
7:43and changes are happening. There was a
7:46great trend for the internet. Then
7:49there was indeed a trend for MarTech
7:51and ML, which we caught. Uh-huh.
7:53And now you can see this wave, which is
7:55causing a FOMO effect among
7:57entrepreneurs. I see a ton of
7:59entrepreneurs who are doing other
8:00things. They come and say, "Artyom, how
8:02do I get into artificial intelligence?
8:04I want to get into this, tell me how."
8:06And so on. Yeah. And so, I think we
8:08need to talk about this, explain how it
8:10can be used, and provide the main tools
8:12. We won't cover all the tools today,
8:13but we'll—
8:14we'll make another episode if it goes
8:16well. I'll gather my thoughts again.
8:17Actually, I think you could make a
8:18separate podcast just on this topic.
8:20Where shall we begin? We will start by
8:22finding a niche. We need to understand
8:23what to do. We have prepared three
8:26approaches to finding a niche using
8:27artificial intelligence. You will now
8:31watch how this is done on my screen. I
8:34have ChatGPT open. There. So, I have
8:38prepared one prompt. Artem has prepared
8:42one prompt. And I will also show one
8:44more approach that often works for me,
8:46which you can do too. The prompt Artem
8:49brought today is in English. So my
8:52first request—a prompt, by the way,
8:54is a request given to artificial
8:56intelligence, which is then processed
8:58and helps us get a result.
9:01Nevertheless, I think it is good
9:03practice to submit requests in English.
9:06The quality improves a bit, so as not
9:08to lose it. And you can always use
9:11translators, so you can write in
9:13Russian.
9:13Let's translate it now, read it in
9:15Russian, and then use the original
9:17English, because these are life hacks
9:19for you—that using it in English is a
9:21bit more effective. Translate the
9:24request into Russian, so we all
9:29understand its meaning, right?
9:30Yes, there are two types of models.
9:32Some that generate the next words, and
9:34others that know how to reason. And at
9:37the moment of reasoning, it generates
9:39about 10 or maybe even 100,000 thought
9:41threads and then chooses the best one.
9:43And this is the version called O3,
9:45right? Version O3.
9:46I see that it is currently rewriting,
9:48right?
9:49Meaning it shows its reasoning. Yes. It
9:51thought for 25 seconds. "You are a
9:53counterintuitive digital product
9:55strategist, combining market insight
9:57with bold imagination." Inject as much
10:00context as possible for ChatGPT and
10:02other neural networks so they
10:03understand what we want from them. Goal
10:06: create exactly 100 ideas solely for
10:09software products that could
10:10potentially generate $ 100,000 in
10:12monthly revenue. Each idea must be
10:15oriented toward a target audience. This
10:17is what we need to add and write in now
10:18. Originality is critical. No idea
10:21should overlap in its value proposition
10:23. Quickly write down 100 short ones. So
10:26, in short, that's the prompt. Alright,
10:28let's take it again. Open a new chat.
10:32What do we need to enter here? We need
10:34to enter something here in square
10:36brackets. It was here
10:38a bit higher up. Go ahead.
10:41Here is the target audience. Let's say
10:44for marketplace sellers.
10:46So, also let's specifically say
10:52Wildberries and Ozon sellers. Like that
11:00. I'd rather
11:05in Russia. So, place your bets. Here is
11:08the most banal idea. Most of the
11:10startups around working with sellers
11:12that I see. These are analytical tools
11:16for marketplace sellers, as well as
11:18repricers and bidders—essentially
11:20tools that manage pricing and
11:22advertising bids on marketplaces.
11:25Let’s give it a try. Yes, that’s
11:27the obvious path, and as an
11:29entrepreneur, it’s not something
11:30I’d want to do because each of these
11:33niches already has established leaders.
11:35In analytics, it’s MPStats; in ad
11:38management, it’s Xway. There you go.
11:41Second, there are dozens, maybe even
11:43hundreds of small teams already working
11:45in that space. I wouldn’t want to
11:47jump into a "Red Ocean" right from the
11:49start. But let’s take a look anyway.
11:51Since I’m often around sellers, I
11:53have a pretty good idea of their pain
11:55points and even know what kind of apps
11:57could be built. Now let’s see what
11:59niches it suggests for us. Alright,
12:01let’s go. Starting it up. We’ll
12:03make the second prompt easier,
12:05yes,
12:06significantly easier, which I’ve
12:07already prepared. I’ve got the prompt
12:09ready in Russian. So, you are a product
12:12strategist, visionary, and startup
12:14architect. Your task is to come up with
12:16a service. A service for sellers.
12:24Let’s come up with 10 product ideas
12:26in this niche. A service for
12:31marketplace sellers on Wildberries and
12:37Ozon, that people would actually enjoy
12:42using and be willing to pay for. One
12:44that has a real product-market fit.
12:50I’ll tell you about myself, and you
12:52propose 10 product ideas. Let’s make
12:57it 10, with names. Explain why it would
12:59work. Preferably with monetization
13:02examples. I need to put my info here.
13:06What should we write about ourselves?
13:09That I’m a serial entrepreneur. I
13:12study a lot, and I’m into chess. I
13:20don’t know what else to write.
13:21My name is Nikolai, most importantly,
13:23and
13:24chess. My name is Nikolai. Right, ask
13:30me a clarifying question. Don't suggest
13:32something generic. And one more thing I
13:42want to add, to make it more relevant,
13:44is that the product you propose must
13:46have the potential to reach 1 billion
13:48rubles in annual revenue. Let’s do
13:57that; it’s all important.
13:58There’s a "write more" button, so it
14:00generates all of this for us at once.
14:02What do you mean, generate?
14:03To have it generate all of this for us?
14:05Yes, yes, generate it for now. Go ahead
14:07.
14:08My profile, list it when you're done.
14:10Alright, let it work. And third, what I
14:13want to do. The third prompt. This
14:17trick has worked for me many times when
14:19the results from other prompts didn't
14:21satisfy me. If I’m not happy with the
14:26results, I assume the issue might be
14:28with the prompt itself. So, my prompt
14:34looks like this. Come up with a prompt
14:36for me. That is exactly what we are
14:38going to do now. We are doing this from
14:40audio. Hello. So, I am a serial
14:43entrepreneur. I'm here with another
14:45entrepreneur. My name is Nikolai. I am
14:47with another entrepreneur named Artyom.
14:50We are recording a podcast, and right
14:52here on the recording, we're coming up
14:54with our next business venture. We need
14:57an idea for an IT startup, for a
14:59technology company. We are looking for
15:03a niche where we can create a software
15:06product, a so-called SaaS cloud service
15:08, for sellers on the marketplaces
15:10Wildberries.ru and Ozon.ru, which could
15:12potentially reach a revenue of 1
15:14billion rubles. Per year, potentially.
15:20And one that can be built as quickly as
15:22possible, so that our first version for
15:25testing hypotheses appears as fast and
15:27as cheaply as possible. Please. Ask me
15:34what you need to ask me, and come up
15:35with a prompt yourself that I should
15:37give you so that the result turns out
15:39the best. That's it. That's about 30
15:44seconds of speech. Now this text will
15:47be recognized, or rather, this audio
15:49will be transcribed into text. I will
15:51give it to ChatGPT to think about. It
15:53will create the prompt and immediately
15:55provide the result itself. There was a
15:57recent OpenAI conference. That is the
15:59developer of ChatGPT. And their CEO,
16:03Sam Altman, known to everyone, stated
16:05that it is only a few years away until
16:08a company worth over a billion dollars
16:10—a unicorn—appears in the world,
16:12consisting of just one person doing
16:15what we are doing today. They might
16:20have coded something, but they found
16:22such a niche, such a gap between demand
16:24and supply, which was easily bridged by
16:26the kind of mechanisms we are playing
16:28with today, that they built a unicorn,
16:31maybe even went to IPO and cashed out
16:33properly, and this event is just a few
16:35years away. Maybe someone watching us
16:39right now will be that first person.
16:42We are already using agents internally,
16:45and I would say that we probably have
16:47four or five agents working. It’s
16:51just a program we run, but it would be
16:53quite possible to give them names,
16:55positions, and job descriptions. Right.
16:59But I heard another story, that the
17:01main competitor to OpenAI, a company
17:04called Anthropic, said that 90%of the
17:06system's code was written by the system
17:08itself. It writes 90%of its own code.
17:12Cool. Yeah. The next mega-trend in this
17:16story is, of course, the reduction of
17:18resources allocated to the payroll for
17:21engineers. We see American companies
17:24cutting thousands, tens of thousands of
17:27people. And these waves are already
17:29reaching us. Those in my circle
17:33involved in IT recruiting are also
17:34hinting that, guys, things are starting
17:37to look bad for us. They still blame
17:40the economy for now, but the battle for
17:42productivity is sort of beginning. For
17:44now, the code this thing writes cannot
17:46survive without a human, without an
17:48operator, without an expert. Because
17:52there's total nonsense sometimes, and
17:54there must be an expert to fix it, but
17:56the expert's labor costs are reduced
17:58significantly, by orders of magnitude.
18:01Do you remember that story about
18:03musketeers and archers when there was a
18:05war, I think between...
18:07No, tell me.
18:09I might get the location wrong. Maybe
18:12it was a war in Asia and the French
18:14musketeers were able to win tactically.
18:19And the main advantage was that to get
18:21an archer, you need to train him from
18:23the age of 10 for a long time. But a
18:26musketeer, you just give him a stick
18:28and say: "Shoot." He trained for a
18:30month. If a musketeer was killed, no
18:33big deal. The next one will come and be
18:35trained quickly. Oh.
18:36Like, muskets were less effective back
18:38then than bows? Is that what you mean?
18:40At the dawn of it.
18:43Maybe the archers were much more
18:45accurate and skilled, but there were
18:47far more musketeers. And there was a
18:50big essay which stated that...we won't
18:55lose developers, we'll have 10, 100,000
18:58, 100 times more developers. Meaning,
19:01everyone will become a developer.
19:03Well, yeah. I recently heard a thought
19:06that, you know, uh, most people...
19:10Uh-huh.
19:12are proxies between a computer and a
19:14user of some service or product.
19:17Meaning they are simply an interface.
19:20We go to a bank, there's a person
19:22sitting at a computer, typing in a form
19:24, filling it out, checking boxes,
19:26taking notes for us. Well, artificial
19:28intelligence will handle all of this
19:30very soon, and definitely better. There
19:33are people who take orders, people who
19:35perform other functions, in short,
19:37acting between the consumer and the
19:39computer. And they will be replaced
19:42soon, including by things like this.
19:45Listen, let me tell you a scary story.
19:51News just came out that in the States,
19:53young professionals can't find jobs.
19:58The demand for them is falling.
20:00Unemployment is rising sharply among
20:02developers and analysts. And maybe
20:07something is happening with their
20:08economy, but I think this is already
20:10the first wave of artificial
20:11intelligence. Yeah. And I think that we
20:17will face this in our economy in 2-3
20:19years. It was interesting that everyone
20:25said: "Go into IT, everything will be
20:27fine for you." Everyone somehow started
20:29running from there. It turns out they
20:31created a problem for themselves, right
20:33? It turns out they created their own
20:36replacement.
20:37Well, no, I don't think so. I don't
20:39know. When I was applying to university
20:41, I really wanted to get into
20:43everything related to computers. I was
20:46so interested in it. My parents told me
20:49: "No, you're not going to be a
20:51programmer." They've made so many of
20:53these programmers that nobody needs
20:55them anymore. That was the early 2000s.
20:5920-plus years ago. So, 20-plus years
21:02later, skilled engineers are still in
21:05huge demand. There. And, well, in
21:08general, people who work in IT are
21:11economically significantly more
21:12successful on average than most other
21:15professions, they are at the top.
21:18Our model is deep in thought, doing
21:20something,
21:21right? Writing code.
21:23Writing code for some reason. Yes. An
21:27idea about draining a competitor's
21:29stock just flashed by here. Well, let
21:34me tell you the story of one of my
21:36close friends who actually had this
21:38problem. What can a competitor do? What
21:43can they do on Wildberries if you're a
21:45seller? They can take it from your
21:47Moscow warehouse. You've shipped some
21:49great item to the Moscow warehouse that
21:51sells really well for you. They can
21:54order all that remaining stock to
21:56somewhere like Novosibirsk. It will
21:59take 2 weeks to get there. The seller
22:03whose goods are being shipped will be
22:05charged for the logistics. The goods
22:08will stay there in Novosibirsk at the
22:10pickup point, and nobody will pick them
22:12up. They will lie there for another
22:13week. Then they will return to the
22:15Novosibirsk warehouse, where there are
22:16no sales for this item because, for
22:18example, it's something for summer, but
22:20it's cold there, or it's just something
22:22unsuitable for Novosibirsk. In general,
22:23in MPSA, you can see which item sells
22:25in which region. During this time, this
22:27bestseller won't be sold to those who
22:29actually need it. And it will just lie
22:31there too, and you'll have to pay for
22:33storage. This is an attack that, as I
22:36understand it, competitors carry out on
22:38each other with regular frequency. And
22:40Wildberries itself doesn't do anything
22:42about it because they make money on it.
22:44They take a storage commission, so
22:45they're fine with it. So, our first one
22:47is still thinking.
22:48And my first one on another model has
22:50already generated 100 ideas. And
22:52actually, I think that this is
22:54something we should just take and
22:56publish: 100 business ideas for WB.
23:00Well, let the guys just implement some
23:02of them in the comments.
23:03Yeah. Okay, the first request is just
23:05your most powerful one. That's why it's
23:07taking a long time. Let it work. Uh, so
23:10. But the result of the second prompt
23:12just came in.
23:13Uh-huh. Let's take a look.
23:14Yeah, let's see what's happening here.
23:16The Pricing Gambit,
23:17right? So, I shouldn't have said I like
23:20chess. Uh, title and hook. Uh, MVP
23:26version of a Telegram bot where a
23:28seller uploads a sales file. The bot
23:30provides three strategic moves. So it's
23:33an advisor that will help earn a bit
23:35more. Blitz reviews. Instant review
23:39collection. The user scans a QR code on
23:41the packaging. Yeah, so this is a
23:44really important thing on WBIS.
23:46Ranking depends heavily on that, and
23:48conversion down the funnel, what your
23:50card's rating is, and in our part of
23:52the world
23:53You can stick a QR on
23:55You shouldn't, but everyone does it
23:56anyway. In our part of the world, it's
23:58just not customary to leave positive
24:00reviews. In our part of the world, it's
24:02customary to speak up,
24:03if something bad has happened. That's
24:07why you have to work on improving the
24:09card rating all the time, and sellers
24:11try to get reviews by any means
24:12necessary. And there are actually a
24:17couple of cool bots you can buy for
24:18money.
24:20There. But I don't really like them, so
24:22I would put this idea on the shortlist.
24:24A bot for collecting reviews. This is a
24:28product you can sell to all sellers on
24:30a subscription basis. There. Uh, here
24:32there's also a QR sticker generator. I
24:34see, it wrote. Yeah, points, perks,
24:37mini-prize, heat map. Well, pretty cool
24:41. So, product card bots,
24:43right? A/B testing for cards. What is
24:46an A/B test? An A/B test is when we
24:48take two versions of something, split
24:51our audience randomly in half, and one
24:53part interacts with the first version,
24:55the second with the second. And we see
24:57which one has better sales, which one
24:59has better customer reviews, to decide
25:02which version to use. Well, it's like
25:04Cherchesov at the World Cup. Either
25:06I'll call up Kokorin to the squad, or
25:08Dzyuba, and see which works better.
25:10Good.
25:12So, uh, a Chrome extension pulling card
25:15KPIs, a win-loss chart, Notion doc, but
25:18there's no proper API on the platforms
25:20to run A/B tests for cards. There's
25:23some raw internal functionality.
25:25There. So for now,
25:28what I wanted to tell you is that our
25:30next prompt will be ideas for
25:31challenges. There. So we can just
25:33quickly make some selections now.
25:35Micro-warehouse sounds really cool. A
25:38P2P network
25:39of empty garages, offices. Combining
25:42them into a platform. I think it's a
25:44very cool hypothesis. I don't know how
25:46many empty garages there are that could
25:48be turned into offices, but I know that
25:50storage businesses blew up a few years
25:52ago. I mean, those buildings that were
25:54renovated and converted into small
25:56storage units are all occupied now.
25:58These units.
25:58Well, I wanted to point one thing out
26:00here. The fact that
26:03even though you are, well, simply an
26:05expert, and you have insane experience
26:07with both tech projects and marketplace
26:09businesses, we're already seeing
26:11something new in these five ideas that
26:13we hadn't thought of.
26:16Yes, I hadn't thought about that at all
26:18. So, this thing here is cool. I don't
26:21know why it pulled us into chess.
26:23An influencer exchange
26:24with a rating system, perhaps. You see,
26:27absolutely all marketplace sellers,
26:29when I get into their community, say: "
26:31Guys, girls, what's up with bloggers?
26:33Whoever knows how to work with bloggers
26:35effectively, the one who knows, doesn't
26:37tell anyone." Actually, that's just the
26:38way it is in the marketplace world.
26:39Nobody tells anyone anything. Someone
26:41finds a good angle and stays silent.
26:43They sit there like a partisan because
26:44as soon as it's leaked, it will stop
26:46working and that's it. Uh, influencer
26:48aggregators that work well with
26:50bloggers—they don't exist. They exist
26:53in-house for big sellers, large trading
26:55companies, and major retailers. They
26:57keep their own databases, watching what
26:59worked for whom, but a general
27:01aggregator like that doesn't exist.
27:02Very cool. So, reasons for returns.
27:07Import directory, Club 64, a premium
27:10mastermind for sellers with chess
27:12sparring. So, by the way, such a
27:15business already exists. There really
27:18wasn't a landmark, cool mastermind
27:21among sellers. Now there is, it's
27:23called Big Russian Brands. Seryozha
27:25Filippov, we send our greetings to you.
27:27He was a guest on this podcast a few
27:29years ago. Now there really is the
27:31coolest mastermind. That's how it is.
27:34Okay. Well, I like what I'm seeing.
27:37Let's see what yours did.
27:38No, it doesn't want to. Switch to
27:40another model and let's restart. So,
27:43well, in short, this is the fast
27:45ChatGPT model we are currently running
27:48this prompt through. We decided to do
27:5020 ideas instead of the 100 we
27:52initially planned, just to generate
27:54them and save time. But you can do 20
27:57or 20,000. Basically, let me convey one
28:00thought here about how models differ.
28:03There are models that appeared first,
28:05that came out first. GPT-3.5 and, by
28:09the way, I think it's been a few years
28:13now, GPT-4o. These are models that, in
28:18their own way, only predicted the next
28:20word.
28:21Uh-huh.
28:22Based on the previous context. And you
28:25can think of such models as a very,
28:27very large archive of the internet.
28:29Uh-huh.
28:29So, we simply took all the knowledge
28:31that exists,
28:32Uh-huh.
28:33compressed it into a model, and you can
28:35extract it from there. With prompts.
28:38From one side, from the other. There.
28:40And here we saw that the model was
28:42already thinking for 54 seconds on the
28:44screen,
28:46and a reasoning layer was added to them
28:48, which significantly improved their
28:50quality, and they were already able to
28:52solve problems involving instructions,
28:54tasks, and executing certain
28:56directions. And such models are called
28:58agentic or reasoning models.
29:00Uh-huh.
29:00There. And there was a big breakthrough
29:02in them. Moreover, we are currently
29:04using the O4 Mini version. Uh, this is
29:06a lightweight version of the O4 model.
29:08We tried to do it on regular O4, but it
29:11took too long, and we just switched to
29:13the lightweight version here to save
29:15time and show examples. But you can try
29:17O4, the results there will be even
29:19cooler. Let's ask it to translate into
29:21Russian. We got 20 results. Translate
29:27the results into Russian. Okay, while
29:35it's thinking, let's see about our
29:37third model.
29:38Uh-huh.
29:38Uh, so, for the idea to really take off
29:41, we need to clarify six things. Then I
29:43will be able to provide the most
29:45accurate and viable one. Pain points of
29:47sellers. Okay, uh, good. Let's just
29:50answer here by voice. We'll go in
29:52dialogue mode as well. So, we need to
29:54find out the pain points of sellers. We
29:57aren't marketplace sellers, we don't
29:58know them. Therefore, we need to
30:00analyze it somehow, imagine that you
30:02are a seller yourself. And come up with
30:05it yourself, find out what problems
30:07sellers face. How many people are ready
30:10to work on the MVP right now? Right now
30:12, there are two of us sitting here. We
30:17will white-code the prototype, and then
30:19our plan is to probably outsource the
30:21development to one of the many software
30:23outsourcing companies. So, uh, do you
30:28have your own stores? No, we don't have
30:31stores. A community, yes, we have one.
30:33We will find sellers, go to a
30:35conference, we can talk to them, and
30:37when we have the product ready, we can
30:39get feedback from it. What budget and
30:42time frame do you have, and when should
30:44it start paying off? Well, we should
30:47reach self-sufficiency within the first
30:49year, and we should capture at least 1%
30:52of the total addressable market in the
30:54first year. That’s that answer. So,
30:59is borrowed money acceptable? Well, we
31:02assume that we are coding all this on
31:04our own, and we won't have access to
31:06debt financing. Especially since banks
31:08don't like giving loans to IT companies
31:10because they have nothing on their
31:12balance sheets and there's nothing to
31:13take from them later. But we are
31:16counting on venture capital. We are
31:18counting on the fact that once we have
31:20prototypes, market feedback, and some
31:22positive soft commitments for sales,
31:24we'll be able to attract venture
31:26funding, even in the current economic
31:28situation. I think that won't be a
31:29problem. So, do you plan to target the
31:31top 300 sellers, focus on a high
31:33average check, or go for the long tail?
31:35I don't know, we need to calculate it.
31:37Our goal is to reach a revenue of one
31:39billion per year over the next few
31:41years. So, you tell us which segment we
31:45should target. I don't know. We are
31:47focusing on the Russian market. Yes, we
31:49are focusing on the Russian market for
31:51now. We don't want to go anywhere else
31:53yet. Are you ready to invest in LLM
31:56features? Uh, do we need classic BI or
32:00ERP? That's it, those are very
32:02complicated words. We don't really want
32:04to dive into them right now. We just
32:07need everything to be clear, verified,
32:09launched, and tested. Uh, that's it,
32:12waiting for the request. And there's an
32:14example of the request right here,
32:15right? We, Nikolai and Artem, want a
32:18SaaS.
32:18Uh-huh.
32:19There. Right, well, this text is being
32:22recognized now, I'll send it.
32:23Let's go read the ideas that were
32:25translated.
32:26Yes, let's translate the ideas, more
32:28complex models. Artem's request.
32:31Price-pulse. Difficulties, manual price
32:34setting. Yes, a repricer, price
32:36adjustment, is indeed lying on the
32:38surface. That really does exist. And
32:41there are, well, a lot of repricers out
32:43there now. They have many interesting
32:45use cases. I've looked into this niche.
32:49Uh, the competition there is too big,
32:51it's too...
32:53There's also a threat, I think, from
32:55Wildberries and Ozon, because they want
32:57this direction. There is a conflict of
32:59interest there. There is, for example,
33:02a use case like raising the price
33:06during off-peak hours. That is one use
33:08case for a repricer. Very cool. It's
33:11not profitable for Wildberries. It
33:13works against Wildberries because they
33:15force sellers to rank higher and join
33:17promotions, and the promo price is
33:19taken as the median price for the last
33:2130 days. If you raise the price
33:24slightly during off-peak hours when
33:26demand is statistically low, your
33:27median price increases, and your promo
33:30price also increases. And that boosts
33:32your margins. But Wildberries will
33:33never implement such scenarios. It goes
33:35, it goes to the detriment of their
33:37earnings. Wildberries, they aren't
33:38interested in that. But there is
33:39something else there. Inventory planner
33:41. Inventory management. A huge problem,
33:45I'll be honest. None of the sellers
33:49really know how much stock to order,
33:51when to order more, and when to order
33:53less. I’ve talked a lot about this
33:58with people who are actually in the
33:59business. Most people have a very poor
34:03understanding of what a seasonality
34:05index is. Just, what is a seasonality
34:08index? Very few people have a clear
34:11idea of what return on equity is and
34:13how the speed of money turnover affects
34:15the business’s financial health.
34:19I’ve just said all these words. There
34:21are probably many sellers watching us,
34:23and they’ll be like, "What did he
34:24just say?" But, guys, this is the key
34:26figure in your business. Return on
34:28equity. Google it. There you go. And
34:31illiquid stock that piles up in
34:32warehouses is, I don’t know how to
34:35put it, a cancer. In short, in this
34:38business model, it just grows on its
34:40own. And these things treat it, which I
34:43think is a very strong hypothesis that
34:45we didn't have before. To build a smart
34:48tool that analyzes purchases, inventory
34:50, and sales, and turns on red lights,
34:53telling you, "Look, it's time to order
34:55this," because it’s all coming from
34:57China. Production is 15-25 days, 25-35
35:01days for logistics, a week for
35:03processing, then delivery to the
35:05marketplace in Moscow, and then some
35:07sales cycle for all of this. So we've
35:10created a production-logistics-sales
35:12cycle of 65-80 days.
35:14You've gotten a bit carried away,
35:15acting like you've already made a
35:16billion. Just go ahead and do it.
35:18I’m just describing the business
35:19model; it’s a cool thing, in short.
35:22Micro-trends are also absolute gold.
35:26Whoever cracks this won't be running
35:28this kind of business; they'll use it
35:30for themselves and won't tell anyone
35:31how it’s done. Because the most
35:34important question is what to sell on
35:36the marketplace. Everything else—if
35:39you find something trendy and
35:41high-margin before anyone else—
35:43you’ll be forgiven for lacking
35:44everything else I mentioned before.
35:47How long does a product actually last
35:49these days? I mean, if we find, say,
35:51months...
35:52It's literally called product arbitrage
35:54. Then other arbitrageurs see that
35:56it’s taking off, they order the same
35:58thing, and that's it. Okay,
36:00non-converting photos and tools for
36:02photo correction. Well, it's a cool
36:05thing, but it feels weak to me.
36:07Maybe you can make $ 10,000, but
36:09probably not more.
36:12I think it’s weak because the photos
36:14are still kind of mediocre. You can
36:18tell they're generated, and you need
36:20too much volume for anything to really
36:22work. Launching competitor promotions,
36:25tracking products. That’s a cool
36:28thing. Honestly, there still isn't a
36:31proper—well, there are Telegram bots,
36:33but it seems there's still no good,
36:35convenient solution that tracks
36:37competitor prices. Uh, price wars in
36:40marketplace retail are a really big
36:42deal. And having a simple alert system
36:45to tell you something has happened and
36:47you need to check your price—that’s
36:49important. I'm not even talking about
36:53events like promo launches and
36:54competitor actions; there's price, and
36:56there are other events too. A manager
37:00messes up and doesn't set the right
37:02price they were supposed to, so you
37:04need something to trigger an alert,
37:06because there are so many problems,
37:07like someone accidentally setting a
37:09price too low. That happens from time
37:12to time. Bots immediately buy up the
37:15entire stock. Wildberries sees the bots
37:18, instantly penalizes the seller's card
37:21, and says, "You're engaging in
37:22black-hat promotion here." At the same
37:25time, it keeps all the buyouts, and
37:26support just sends template replies. In
37:28short, it’s a total disaster. Okay,
37:31limited listings,
37:34SEO optimization. Uh, I think it's very
37:37promising.
37:39No one on WB—well, out of the sellers
37:42I talk to—understands how SEO should
37:44be structured. We actually do this
37:48professionally. We've just launched six
37:51pilots with large sellers. And for one
37:54seller, we took half their product
37:55listings and significantly increased
37:57their sales. Meaning, we split the
38:00stock into two parts.
38:01Review monitoring—listen, that’s a
38:03cool thing. The guys at MPSTATS are
38:05doing "Micro-FO."
38:06Everyone knows you need to read reviews
38:08, both yours and your competitors '.
38:10It's part of product research. It's
38:11part of the functionality for a product
38:13selection service. The point is that
38:15you need to analyze what people
38:17complain about, what they like, and
38:19what they dislike, and based on that,
38:21look for product hypotheses. Yeah. And
38:23this is actually an opportunity, so,
38:25well, in short, it’s a cool feature,
38:28I think.
38:28It’s like you start to hear the voice
38:30of the customer.
38:31Anyway, I think there’s actually no
38:33point in us continuing further. There
38:35are really some
38:35great ideas here.
38:37Very great ideas that were born in just
38:39a few minutes. Let’s try our third
38:41approach now. Uh, right. The audio has
38:44finished transcribing.
38:49Uh, let it think for a bit. Uh, so,
38:51what do we have? We have very strong,
38:56promising ideas that, with my 15 years
38:58of experience in digital business and
39:01startups, I really like. I feel that
39:05they are promising. Even though we just
39:08came up with what to do, I hadn't
39:10thought about it like this before. Uh,
39:13so, let me remind you what our idea was
39:15for this first part: to find a niche.
39:18Now we'll take any niche. We took three
39:22prompts. The first was from Artyom,
39:24which he found on the internet. I also
39:26found my own reel, found my own prompt,
39:29some reel popped up in my
39:30recommendations. And this is how I use
39:33it all the time, when I don't get a
39:35result I like, I describe the context
39:37in a short audio message and say: "Why
39:39don't you make me one." The request
39:43that I should give you. Like that. Well
39:47, here we got a cool result. We,
39:49Nikolai, Artyom, will no-code the
39:51prototype, then we'll outsource
39:53development for half a million. Yeah.
39:55We want a 12-month payback period. It's
39:58all set up great. So, uh, confirmed
40:02pain points. So I said, yeah, I told
40:04him: "Ask me everything you need to ask
40:07. What are the pain points?" I say: "I
40:09don't know what the pain points are. We
40:10decided here on something for
40:11marketplaces. You come up with what the
40:13pain points are." Uh, first pain point:
40:15opaque, erroneous fines and
40:17withholdings. Agreed. A huge problem. I
40:20can totally see an aggregator of
40:22lawyers coming next, who file
40:24complaints to get refunds. And these
40:28lawyers are actually starting to appear
40:29little by little. It's really cool.
40:32Commission growth in the summer of 2024
40:35. Well, it probably doesn't know yet
40:37that in the summer of 2025 everything
40:39grew once more. Yeah. So nobody really
40:41knows what to do. Next, poor cash flow
40:43visibility, payment delays. Payment
40:46delays, by the way, is also a very cool
40:48story, which none of the sellers know
40:50what to do about. And, well, equipment
40:52is one thing. I like this prompt. Let
40:55me say it now. I like this prompt. Go
40:58ahead, using this prompt, generate 20
41:01ideas for SaaS services for me, as I
41:03initially asked. So, we need to start
41:06thinking about which idea we'll take.
41:08Which one did you like the most? I
41:11think the point I wanted to get across
41:13is that it basically doesn't matter
41:15what we do, because we will assemble a
41:17prototype faster than has ever been
41:19possible, and we'll be able to verify
41:21it with real sellers. You can then
41:24throw it into some internal chats. So
41:28that's why I think you should just
41:30choose the one that resonates. What I
41:33would do if I were you, I would just
41:35make a post saying: "Hey guys, I have
41:38four or five ideas, please vote." Why
41:40do that? Let me explain. Because when
41:43you ask a B2B audience, then if they
41:45respond, if they vote for it, later
41:47when your sales reps call and say: "
41:51We're calling about the fines," they
41:52will get through that barrier. Let's
41:54just go ahead and do this now. Uh,
41:56first of all, please leave a comment if
41:59this topic resonates with you at all.
42:03And if it does, Artyom and I will sit
42:05down again and hash out how, before you
42:08move on to expenses, programming, or
42:10complex investments, you can actually
42:12check if a topic will take off or not.
42:16We'll sit down and share our ideas on
42:18how we do it, how we verify these
42:20things. Because today we're looking for
42:23ideas and showing how to quickly
42:25implement and prototype them, but how
42:26do you test if there's real money in it
42:28? So write in if you need this, and
42:32we'll get together again and record it.
42:34Anyway, here's the first one, the one I
42:36really wanted to show. Automated audit
42:39of fines. And I see the functionality
42:42being very simple, very cheap to
42:44develop. It will be a lawyer aggregator
42:48. Lawyers who are already employed
42:50somewhere, corporate types who want to
42:52freelance, that's where the
42:53self-employed angle comes in. It's a
42:56platform solution for sellers plus
42:58maybe some other products later on
43:00where you can sell legal services. Uh,
43:03there is a service in America called
43:05Priori Legal. How did I find out about
43:07it? When we entered the American market
43:09with Retail Rocket, the first thing we
43:11got after 2-3 weeks was a patent troll
43:13who sent us a trademark infringement
43:15letter, claiming illegal use of a
43:17trademark. They asked us for a few
43:19hundred thousand dollars. We were these
43:21penniless startup guys, not knowing
43:23what to do at all. We went to some
43:24experienced mentors. The mentors said:
43:26"Here is a lawyer for you." I show up
43:28on a Zoom with this lawyer. There was
43:30no Zoom back then, so on Skype with
43:32this lawyer. Sitting there is a man
43:35covered in gold, looking flashy, like
43:37Saul Goodman from Breaking Bad, and he
43:40says: "So, here's the deal, a $ 10,000
43:42retainer." He says: "$ 10,000," well,
43:46he was Russian but with an American
43:49accent. A retainer is a non-refundable
43:51deposit. Right. And only then will we
43:53talk. And with this exact intonation he
43:55says to me: "I've understood your
43:56situation up to this point. Overall,
43:57everything is clear to me." Well,
43:59something like that. Basically, he
44:00tells me: "Give me 10,000, and I won't
44:02promise you anything. Whether something
44:03comes of it or not, I don't know." For
44:05me, 10,000 dollars was a huge amount of
44:06money, and I didn't have it. So I poked
44:08around my friends and found this
44:09service called Priori Legal. What is
44:11that? It's an aggregator for lawyers
44:13that has a kind of office layer. You
44:17write up your case, there's a form
44:19where you check off your problem, and a
44:21text box where you write down what
44:22happened to you. And the office sorts
44:26through it—there were no AI tools
44:28back then. And it matches you with the
44:30best lawyers for your specific case.
44:32Uh-huh.
44:32Freelance lawyers. Those freelance
44:35lawyers each prepared a strategy to
44:37resolve my case. Well, each one did.
44:40They had hour-long calls with me and
44:41explained how they would handle it. All
44:43of that was free of charge. And then
44:45they offered, like, "I charge 150 an
44:47hour, I charge 200 an hour, I charge
44:50300.""Whichever of the three you liked
44:52best, just choose them and go work with
44:54them." And I was just blown away. I
44:57mean, by how quickly I received three
44:59legal opinions,
45:00with ready-made strategies right away.
45:02Technically, I could have gone and
45:03implemented them without them. But,
45:05yeah, it was pretty cool. So,
45:06Commitracker recognizes the tariff and
45:08margin. That's about analytics and ad
45:10autopilot. Listen, let's stop right
45:12here.
45:12It doesn't matter what the idea is,
45:14because the most important thing is
45:16that we can test 20 to 30 of them a
45:17month,
45:18right? But I'm saying that what's
45:19important right now is that we need to
45:21lock this in. We've got 5 to 10 really
45:23strong hypotheses here that, looking at
45:26them, I'm like, "Damn, this is great."
45:28FOMO, FOMO,
45:30honestly, maybe a little bit, maybe,
45:33maybe I should do it myself, maybe not
45:35release it to a third party, yeah,
45:37maybe not release this podcast episode
45:40and launch something myself, but
45:41we’ve started, and for me, this is a
45:44"wow." We’ve already come up with
45:46something really cool. That's it,
45:47we’re taking this one.
45:49This is where the magic happens. We
45:51just have an idea. And now we need to
45:53figure out the business model. Possibly
45:55, in which segment, with what
45:57monetization. There are different paths
45:58to take. We could attach ChatGPT so it
46:01generates the answers. We could create,
46:04I don't know, some kind of community.
46:07You could build a lawyer aggregator,
46:09there are many different ideas. Which
46:11one is best? And this is where I wanted
46:13to show you another part of the work,
46:15where we take an idea and flesh it out.
46:17Let’s do it.
46:18And what we need to do at this point—
46:20in real life, you have this idea, what
46:22would you do next? I would reach the
46:26target audience as quickly as possible.
46:29I’d go to some seller conference.
46:33I’d join some seller chat group that
46:35regularly meets up for a small fee at a
46:37hookah bar. Uh, what else? Well, I’d
46:43go to a seller training course, because
46:45there’s a seller community there.
46:48I’d start searching my professional
46:50network for those who sell on
46:51marketplaces. I’d get to the people
46:55who are doing this hands-on as fast as
46:56I could. And my task would be to
47:00validate the pain, to see if they
47:02actually have that problem. How would I
47:05do that? I have a cheat, I have this
47:08one question.
47:09Uh-huh.
47:09It goes like this: "Please rate from
47:11one to ten, how likely are you to pay
47:13to have this problem solved?" I’m not
47:15saying how I’ll solve it; I’m
47:16saying it will definitely be solved. It
47:18will definitely be profitable for you,
47:20it will definitely be convenient for
47:22you. I will fix it, because if your
47:23shoes are too tight or you have a
47:25toothache, you’ll say: "Yes, please,
47:27fix this."
47:28So, let's take that question about
47:30rating—if you are a seller, rate from
47:32one to ten the probability that you’d
47:34pay for this idea—send them 20 ideas
47:37and see which ideas...
47:39What 20 ideas to send? You have 20
47:41ideas, we need to understand which ones
47:43are cool. This question can be asked.
47:45We have several sets of 20 ideas, right
47:47?
47:47For any of them. I think we can try to
47:49find out which of the ideas is cool.
47:50Well, here are, in fact, the next steps
47:53that ChatGPT already suggested itself.
47:55Next steps. It suggests interviewing 25
47:57sellers at conferences and in chats to
48:00validate three pains from the list
48:02above. That's the main thing. And then
48:04take an MVP sprint, choose the idea
48:06with the highest buyability and ease of
48:09implementation. Uh, and it also
48:12suggested making a Telegram channel
48:14where you publicly describe how
48:15you’re building this thing, and
48:17collect a waitlist of people who want
48:19it. I already did that.
48:21Yes. Next, uh, anyway, there is
48:23literally, damn, it looks like we don't
48:26need to film the next episode. I just
48:28need to show you what’s on my screen
48:29right now. 20 to 50 pilot sellers with
48:32a nominal price. Next is unit economics
48:36. This might be unclear, what "less
48:39than 3pu" and "churn less than 5%" mean
48:43. But this is absolutely key. That is
48:47the whole point of what we are doing
48:48right now. Please, well, look into this
48:51or write it down. Again, we will still
48:53make a separate episode. That’s it,
48:56what metrics are needed for a seed
48:58investment round. Preparing for the
49:01round, what is written here, right
49:03before my eyes now, if someone had
49:05written this to me,
49:07and said: "Davy,
49:08when I was creating Retail Rocket, I
49:10would be significantly more successful
49:12and richer now." If someone had
49:14explained to me exactly what I needed
49:16to do with all these confusing
49:18abbreviations. This is incredibly
49:20important. Okay, well, fine. So, uh,
49:22I’m dictating a prompt here right now
49:25. Imagine you are a large focus group
49:28of marketplace sellers, representing
49:31100 people. Conduct all the interviews,
49:38describe the pain points for these 100
49:40people and survey each of them, asking
49:42the question: "On a scale of one to ten
49:44, how likely is each of them to pay for
49:46the solution to the problem you came up
49:49with?" and, uh, provide a list, uh, of
49:56these solutions, ideas for startups,
49:59sorted by the intensity of the pain
50:01discovered among the sellers. But this
50:05is a total game-changer, what’s going
50:07to happen now. I’ve done this a
50:10couple of times with hypotheses, when
50:12you assign it a role not as a single
50:14expert, you know, counterintuitive,
50:16multicultural, but when you tell it you
50:18are 100 people, you are a focus group.
50:22Okay, I’ll push this into all, uh,
50:24all channels now, into all three ideas,
50:27all three idea-hunting streams. Let’s
50:30see what we get, what’s happening
50:32with you. What’s happening is that
50:34I’m challenging this idea and filling
50:36it with different options. We’ll look
50:37at it now, because A
50:39scroll up, I’ll comment on it. What
50:40requests did you make, what did you
50:42actually do?
50:43Yes, what are we doing? We are putting
50:45it back into a role, the role of a
50:47product manager,
50:47right?
50:48Ah, and this is quite a large prompt.
50:50We will post it as well. There
50:52wow. That was a mile long for you.
50:54Of course, I try to make large, good
50:56instructions, because that’s what
50:58makes new models good: they know how to
51:00follow instructions. And at this point,
51:02I’m basically telling you that as
51:04Kolya Khlebinsky, try to figure out
51:06what different audiences exist,
51:08Uh-huh.
51:09that might find this interesting.
51:10Meaning, figure out what problems we
51:13are solving.
51:13Second, what this product approach is,
51:16when a smart product manager comes to
51:18you and says, "Okay, but what' jobs to
51:20be done 'are we solving here?" And you
51:22say, "No, I just want to make some
51:24money." So, accordingly, we ask it to
51:26play this smart product manager and say
51:28: "What roles are there, what consumers
51:30exist for this service, and what tasks
51:32do they want to solve?""A task" is
51:34actually what is usually written as the
51:36basis of an instruction, like, "I need
51:38to appeal fines quickly, within a week.
51:40"
51:41There. Next, an important question
51:43arises: how do we position this product
51:45? Is it as a debt collector, as an
51:47automation tool, or as a service that
51:49finds a good lawyer? We don't know. We
51:52want to sort that out. What will even
51:55go into our MVP? It could be some big
51:57platform, a small app, or a Telegram
51:59bot; we don't know what features should
52:01be inside, what the business model is,
52:03or what the revenue could be at all.
52:05These are all the questions that you
52:06and I understand perfectly, and we will
52:08actually pass this knowledge on to the
52:10viewers. This is vast entrepreneurial
52:12knowledge that we have been gathering
52:14for years. When I asked you what you
52:16would do, there’s another path where
52:18you call up an entrepreneur and say, "
52:20Listen, I have an idea, let's discuss
52:22it." And they sit down with you and
52:25start solving it on a napkin.
52:27I'll show you what I came up with while
52:28your processing is running. We imagined
52:33a focus group, described the pain
52:35points, and this service here, "
52:37Inventory Sinker," won.
52:41Uh-huh.
52:41So, I think this was about procurement,
52:44right? Uh, inventory synchronization.
52:47In short, this was our problem with...
52:49Let's find where it is. What service
52:53won here? Inventory Sinker, manual
52:55inventory management. Overselling, lost
52:58deals, dead stock. Centralized
53:00inventory management. Real-time
53:03cancellations and shipments. Yeah, this
53:05is about the warehouse. So, in your
53:07prompt, the warehouse won. Uh, in the
53:10prompt I suggested, there's a different
53:13model here. That was O4 Lite, this is
53:15O3.
53:16Uh-huh. Uh, the reasoning went
53:18differently here, so you need to use
53:20different requests for the same stories
53:22. Uh, here’s the category: uh, we had
53:2615 large sellers, 40 micro, 45
53:28mid-sized in the growth stage. Cashflow
53:32shield. Uh, every day without cash,
53:34sales come to a halt. Instant payouts.
53:37This is apparently, uh, some kind of
53:39factoring, something related to fintech
53:41.
53:41Loans, yes, secured by
53:43loans secured by sales already made.
53:46completed sales for which the money
53:47hasn’t arrived yet.
53:49I mean, we have reports that can be
53:51pulled via API from the Ozon or WB
53:53accounts regarding what was sold. We
53:55have redemption rates, we can forecast
53:57how much money will come and when, but
53:59the seller is actually waiting weeks
54:01for this money.
54:01And don't the zones provide this, or do
54:03they offer it at a high interest rate?
54:05To be honest, I don't know. I know that
54:07Ozon and Beris regularly change the
54:09rules of the game. Like, they said, for
54:11instance, we can do it once a week, now
54:14we do it once a month. And now we do it
54:17—and naturally, it says in the offer
54:20that they can delay it by 5 days past
54:22the deadline, and in reality, it turns
54:25into 10 or 15. And every couple of
54:30months, WB almost deliberately picks a
54:32cluster to delay payments to. They
54:35reimburse everything the following week
54:36, but I feel like they really play
54:38around with the money there. Yeah. Well
54:40, anyway, a fintech project won, which
54:42finances sellers against the guarantee
54:44of payouts from marketplaces that
54:46haven't happened yet. So, what else do
54:49we have here? Fines. Fines, billing
54:51errors, yes, just as I assumed, that
54:53also seems like a very cool story to me
54:56. Any seller, well, at least those I've
54:58spoken with, is willing to outsource
55:01their entire document flow with WB,
55:03lawsuits, pre-trial claims, or whatever
55:05, for a percentage of the fine—like
55:0810-20-30%of the recouped fine—because
55:10nobody wants to deal with it. Nobody
55:12has their own lawyer or a decent
55:14litigator. These are all very cool
55:17stories, and they are sorted and ready
55:19for implementation.
55:20Let me show you what we have now.
55:22Go ahead. Actually, I think we're done
55:24with the ideas, because these primary
55:26hypotheses were generated brilliantly.
55:28Let’s move on. What was happening
55:29here? Just tell me the structure of the
55:31content of this prompt. What’s in it?
55:34You could invite a really top-tier
55:36product manager from T-Bank or, I don't
55:38know, Avito, some really smart folks,
55:40and they would break it all down in
55:42detail and conduct research over three
55:44or four months.
55:45But
55:46in reality, all of this was done in a
55:48couple of minutes. So what did we do?
55:50We identified the main personas. Pro
55:52seller, and a mid-tier seller.
55:54Did you identify this, or did the AI do
55:56it? The AI did it. So you asked it to
55:57do that right away. I told you: "
55:59Identify the personas." Yes, that was
56:00your task. As a cool product manager,
56:02identify the personas, because there
56:03are different personas. So it
56:05identified the pro sellers,
56:06they have certain pain points, yes,
56:08there are small sellers, they have
56:09different pain points, and there are
56:11outsource agencies that manage many
56:13sellers.
56:13So, what else needs to be understood
56:15here is that both Artyom and I have a
56:17great deal of experience in product
56:19management. Yes,
56:21in product development. That's why the
56:23concept of a "persona" is built into us
56:25. When you see a checkmark, we were
56:28thinking about what our key target
56:30audience segments are, what their use
56:32cases for our product are, what their
56:34problems are, how they solve them, and
56:37what characterizes these personas.
56:39Those of you watching us now, well,
56:41maybe you haven't spent 15 years
56:43developing complex information systems.
56:46Maybe some, anyway, we'll just share
56:48the link to this dialogue as well. You
56:51can read it, if something is in English
56:53. You can tell ChatGPT to translate it
56:55if some term is unclear. You can copy
56:58it and say: "Explain.""I don't
56:59understand what this means." But
57:02overall, this is research work that
57:04precedes writing code or investing in a
57:07product so that we can guarantee a
57:08return on investment. It's worth noting
57:13here that it's a huge mistake not to do
57:15this when launching a startup. Because
57:17it happens like this: we come up with
57:19an idea, it pops into our head, it
57:21clicks, and that's it—we're wearing
57:22rose-colored glasses through which we
57:24look at the world, and no one can
57:26convince us that it's complete garbage.
57:29Except for the market, to which we
57:31bring a solution developed over several
57:33years, which turns out to be useless to
57:35everyone and for which nobody pays
57:37later. And we invested millions into it
57:40, either our own or our investors'. And
57:43since Artyom and I—I certainly have,
57:45and I suspect he has too—have built
57:48so many wonderful features that turned
57:50out to be completely useless. We burned
57:53so much money that we live by these
57:55practices to avoid wasting the most
57:57important thing. You have to remember
58:00that the scarcest resource in the IT
58:02business is developer resources,
58:04programmer hours. To avoid wasting them
58:06on nonsense and useless junk, all these
58:09studies are done. And this used to take
58:12months, but now it takes seconds.
58:15Yes. And I was talking with guys from
58:18one of the largest investment banks.
58:21They said they give test assignments
58:23for junior positions, where candidates
58:25need to research the airline market and
58:27solve a case study. They say that
58:31modern neural networks solve these
58:32cases better than university graduates.
58:36And for now, they are using them.
58:38Okay. So, good.
58:39So, what do we see here? We are
58:41starting to outline and detail our
58:43sketch
58:44Uh-huh.
58:45into a very clear system, right? So,
58:48what does our product look like? It's a
58:51one-click, one-click appeal, uh, and it
58:57has built-in tariff reconciliation, and
58:59it includes some form of pricing.
59:03Notice that the system immediately
59:05compared it with some competitors.
59:08Maybe these data points aren't accurate
59:10, and I apologize to the guys if the
59:12data is incorrect, though. But
59:14regardless, it presented competitors
59:16and analogues we could work with in
59:18this way, highlighting Excel as the
59:20primary one. Next, the next part is the
59:23product roadmap.
59:24Uh-huh.
59:25Here we see the engineer's hours. We
59:27move on to the business model options.
59:30It could be a percentage of the refund,
59:33a fixed fee, a connection fee for the
59:35system, or a hybrid model where you
59:37provide the service for 10,000 rubles.
59:40And additionally, you get a percentage
59:42of the recovered penalties.
59:44By the way, I just said that sellers
59:46are ready to pay 20%of recovered
59:48penalties without any issues. And that
59:50is the primary business model. I also
59:53see that the target customer
59:54acquisition cost is calculated here at
59:576,000 rubles. And the lifetime, that is
1:00:00the customer lifetime, is 12 months.
1:00:03And how much money do we plan to
1:00:05receive? The only thing I don't like is
1:00:08the 15-day payback. That I don't like,
1:00:13because on the right, the payback is
1:00:150.5 months. I think that's wrong.
1:00:19Although
1:00:19it suggests starting with a test. For a
1:00:21te-No, for a test, that's very cool,
1:00:23isn't it? So it says that
1:00:24for scaling, it's bad,
1:00:26right? Publish information about the
1:00:28service on the WB Insights channel. It
1:00:30suggests what the price should be there
1:00:32and how many leads you'll attract. With
1:00:33a 12%conversion rate. That seems
1:00:35plausible.
1:00:36Yes, yes, yes. It's a ready-made first
1:00:38step for marketing right here.
1:00:40Honestly, I'm launching an IT startup
1:00:42in Chartrack, a SaaS, myself right now.
1:00:45Uh, I was doing all of this too. But my
1:00:47approach isn't like Artyom's. I see
1:00:49what I'll be doing with my startup
1:00:51tonight after recording this podcast
1:00:53because this is cool,
1:00:55yes, the go-to-market strategy is
1:00:57already written out. And this is the
1:00:58thing you can take and start working
1:01:00with immediately. So, we just had an
1:01:02idea that we didn't quite know how to
1:01:04approach. That is the biggest challenge
1:01:06of all. You sit there thinking: "What
1:01:08if it doesn't work out?""And so on."
1:01:10But it could actually work, right? I
1:01:12mean, we know there's a 10-15%chance of
1:01:14success. That's perfectly fine; we can
1:01:16proceed down this path. And we both
1:01:18understand that this is a reasonable
1:01:20price and a solid model. There. And
1:01:22what we can do next. Look, here is
1:01:24what’s missing. All of this is still
1:01:26just internet archives. All of this is
1:01:28just some brother's advice. And we both
1:01:32know that, uh, there's nothing more
1:01:34daunting, and yet more useful, than
1:01:36advice from some expert.
1:01:38Oh, by the way, it’s here,
1:01:40yes,
1:01:41uh, in this roadmap for hitting revenue
1:01:44. A 6-month horizon, 6 months to
1:01:47revenue, 3-4 million rubles. And it's
1:01:49explicitly stated right here. Okay.
1:01:51Yes,
1:01:52in 6 months he plans to reach it. Right
1:01:54, okay, I interrupted you. Continue
1:01:56your thought.
1:01:57Yes, the important thing here is, and
1:02:01you know there are these experts who
1:02:03say: "Nothing will work for you, that
1:02:05doesn't work." I came to you once,
1:02:07asking: "What’s the maximum I can
1:02:09make in my service?""You gave me an
1:02:11estimate three times higher than what I
1:02:13had projected." Well, time passed, and
1:02:14we made four times more than you
1:02:16predicted, and 12 times more than I had
1:02:18predicted.
1:02:19Listen, here is how I view that.
1:02:21I had a situation once when I was
1:02:22launching Rocket, when we were at the
1:02:24very beginning. I approached a very
1:02:26prominent industry expert, someone we
1:02:28have both known for years. I explained
1:02:30the idea to him. I told him we were
1:02:32building this specific thing. He tells
1:02:33me: "Listen, nothing is going to work
1:02:35out for you." And I started arguing
1:02:36back with foam at the mouth. "No, look,
1:02:38I did it this way, and that way." Right
1:02:40. And he says: "Actually, that was a
1:02:42test.""If you had just given up
1:02:43immediately and said, 'Oh well, never
1:02:45mind.'""And walked away, then you
1:02:46definitely wouldn't have succeeded.""
1:02:47But if you're ready to defend your
1:02:49position with such energy after being
1:02:50told you’ll fail, then go ahead and
1:02:52do it, because nobody knows if you'll
1:02:54succeed or not."
1:02:54Nobody knows. Nobody knows. There. But
1:02:58pay attention to the fact that, uh,
1:03:00there are always these little hooks you
1:03:02give yourself,
1:03:04yes,
1:03:05like certain conditions under which
1:03:06this can be done,
1:03:07yeah. Like, I don't know, could Ufa win
1:03:11the Russian championship next year?
1:03:13Maybe, if they, I don't know, bolster
1:03:15the team with five strikers, if their
1:03:17training process changes, and so on.
1:03:20Some, like the movie Everything
1:03:21Everywhere All at Once, if you haven't
1:03:23seen it, watch it. It’s practically a
1:03:24documentary, everything is described
1:03:26there.
1:03:26There. And here, everything is
1:03:28described as well. Will we be able to
1:03:29consistently recover 60%of fines or not
1:03:31without manual verification? If that's
1:03:33required, then our service won't be as
1:03:35good.
1:03:35All of this is described. And now we
1:03:38move on to another tool, because what
1:03:40we just did is expert knowledge. But we
1:03:42need to get knowledge from the internet
1:03:44, preferably all of it. Because if
1:03:47sellers have an open vacancy for a
1:03:49lawyer,
1:03:51some discussions in open chats,
1:03:53yes, and forum threads, an article on
1:03:55VC, and so on—all this evidence of
1:03:57how it's structured, we need to gather
1:04:00it all. Ideally, if you and I have come
1:04:03up with something, if we got fired up
1:04:05about this idea,
1:04:07Uh-huh.
1:04:07you need to spend another week so that
1:04:09you can approach it later with a cool
1:04:11head. Not to jump into business
1:04:12hot-headed, but with a cool head, spend
1:04:14the time, yes, set it aside, spend the
1:04:16time
1:04:17to study it. To be honest, regarding
1:04:18the projects I have in development
1:04:20right now, I’m already itching to run
1:04:22off and do it, rather than wait a week.
1:04:24I don’t understand how to do that
1:04:25right now.
1:04:25So, there is a main tool, and it is
1:04:27available in the free version. It is
1:04:29the DeepSearch tool. We are now going
1:04:31to, unfortunately, just for 5 minutes,
1:04:33launch a big, big research project. Uh,
1:04:36I'll send this prompt again, and we
1:04:38will, uh, launch DeepSearch, we’ll
1:04:40launch a deep search across the entire
1:04:43internet to find all the information on
1:04:45this topic. We’ll literally dive into
1:04:47every nook and cranny. I’ll do that
1:04:50right here, right now.
1:04:51Well, are you ready to share all of
1:04:53this? Uh,
1:04:54there's nothing wrong with that. Anyone
1:04:55can do all of this.
1:04:56Friends, hello. So, how did you like
1:04:58this episode? I just rewatched it
1:05:00before publishing. We’re making final
1:05:02edits. Man, it’s just mind-blowing. I
1:05:06will post all the prompts, all the
1:05:08requests to the neural networks that we
1:05:10used, in one file on my Telegram
1:05:12channel as soon as the video reaches
1:05:14either 3,000 comments or 3,000 likes.
1:05:19So, if you need these materials, please
1:05:21hit the like button, leave a meaningful
1:05:24comment, and below this video, there is
1:05:26a link to my Telegram channel.
1:05:29Subscribe, all materials will be there.
1:05:31Alright, let's continue watching.
1:05:34It asks: "Are you sure you want to
1:05:36verify this problem?" Yes, that's it.
1:05:39It clarifies: "Which competitors should
1:05:41I include?" Decide for yourself. And
1:05:44the preferred language is, of course,
1:05:47Russian. Oh, look for more. Although we
1:05:54could look at the international market.
1:05:55And so the research begins; it will
1:05:57take some time now.
1:05:59Uh-huh. Good. I'll show you a use case
1:06:02that became a turning point for me. My
1:06:05world was turned upside down, and I
1:06:07realized that the world has changed. It
1:06:09will never be the same again. When I
1:06:12made this request, well, I'm subscribed
1:06:15to an aviation page. I am into civil
1:06:18aviation. I'm interested in everything
1:06:20related to planes, how they work, and
1:06:22airports. And there was a poll there.
1:06:25Just from this photo, write what kind
1:06:27of airport this is. To figure out what
1:06:32To figure out what airport this is from
1:06:34a photo is impossible. I searched
1:06:37Google, looked through several dozen,
1:06:39maybe 40 thousand airports in the world
1:06:41.
1:06:42All over the world.
1:06:43It is impossible to tell what airport
1:06:45this is. And basically, tell me which
1:06:47airport is in this photo. So, I am told
1:06:50that it is Gatwick Airport. I go check,
1:06:54yeah, it took 49 seconds. Uh-huh.
1:06:58I go to Google Maps and see that it
1:07:00also seems to have two runways, but
1:07:02those two runways are parallel to each
1:07:04other. Here you can see that this
1:07:08runway starts further away, and this
1:07:10one starts later. But at Gatwick, the
1:07:12two are right next to each other, their
1:07:14start and end points are practically
1:07:15identical.
1:07:16And so I tell it: "This is incorrect.
1:07:18At Gatwick Airport, both runways have
1:07:20the same start and practically the same
1:07:22end." In the photo, you can see that
1:07:23the start and end are offset relative
1:07:24to each other. Think again. It
1:07:28immediately finds the correct airport.
1:07:31How did it happen from such an angle?
1:07:33Here you and I,
1:07:36people who deal with data analysis and
1:07:39data monetization, would have had to
1:07:41input entities like runway start,
1:07:43runway end, runway parallelism, as
1:07:46factors, forests, houses. How can this
1:07:50even be analyzed? Because it was
1:07:52looking at satellite images, probably
1:07:53from above. And creating this with
1:07:57previous-generation approaches was...
1:08:00And when I saw that it happened in 2
1:08:02minutes by a machine with a very
1:08:04abstract request and very abstract
1:08:06input data, it just blew my mind. That
1:08:08was the moment I realized that
1:08:10everything, the world would never be
1:08:11the same again. A very non-obvious use
1:08:13case. It probably won't be useful for
1:08:16many people in life, for those watching
1:08:18us, but the level of complexity of the
1:08:20technologies that can do this is just,
1:08:22well, off the charts. How far have we
1:08:24come?
1:08:25You can send a photo, I don't know, of
1:08:27a person who took a picture somewhere
1:08:29for an avatar and ask where the photo
1:08:31was taken. And you can figure out where
1:08:34they live. Streamers who play these
1:08:35games, when a photo is shown to them
1:08:37for 3 seconds, they mark on the map
1:08:39where it is taking place. There. Yes.
1:08:41That's it,
1:08:42yes,
1:08:42that is already a product there. With
1:08:44what you are doing now, what will be
1:08:46the end result of it? I mean, that
1:08:48product research has been conducted,
1:08:50the work of a product owner, product
1:08:52director, product manager has been done
1:08:54—the person who is responsible for
1:08:56ensuring that the software code written
1:08:59with developers brings value to someone
1:09:01, is in demand, and is needed. I see
1:09:03that something grandiose is being done
1:09:05there.
1:09:05What are we going to do now? We will
1:09:07look into all, all, all, all the
1:09:09details, into all the most detailed
1:09:12corners of the internet. We will
1:09:14collect all information about
1:09:16Wildberries fines, we will get into
1:09:17Wildberries itself, we will get into
1:09:19negative reviews on Yandex Market, and
1:09:21somewhere else. From this, we will
1:09:23structure it more and challenge our
1:09:25idea. From it, we will assemble not
1:09:27some hallucination from a neural
1:09:29network, not some fragmentary data of
1:09:31ours, but a complete snapshot of what
1:09:33is happening in the world according to
1:09:35this data. And we will already have a
1:09:37completely ready-made guide with market
1:09:39research. I mean, what we are doing now
1:09:41, we are doing a full-fledged, most
1:09:43detailed market research, which,
1:09:45honestly, why didn't entrepreneurs do
1:09:47it before? Because it is expensive. But
1:09:49in reality, when some P&G launches a
1:09:51new product or someone launches a new
1:09:53type of chocolate, they actually take
1:09:55and survey a thousand people: "what do
1:09:57you need?""how is it structured?"—
1:09:59then they say: "In St. Petersburg, they
1:10:00like these kinds of chocolates, in
1:10:02Yekaterinburg, they like these kinds of
1:10:03chocolates, and so on." There. And we
1:10:05really should conduct this research. So
1:10:07, what you said, that you will do, you
1:10:08will go to a conference, you will be
1:10:09collecting all this information. And
1:10:11that is what we are doing now. This is
1:10:12a big, big prompt.
1:10:14We are trying to understand the market
1:10:16size, we will try to understand some
1:10:18main competitors, what they have, how
1:10:20they use it, we will pull out pieces of
1:10:22information from the internet, all the
1:10:24reviews, and so on. We will collect
1:10:26everything now. And then we will have
1:10:28full coverage of the pain points. We
1:10:29need to understand that despite there
1:10:30being competitors—and there certainly
1:10:32are—how can we differentiate from
1:10:33them? What should the product be to
1:10:35successfully break into the market?
1:10:36That is the most important thing. If a
1:10:38market exists, that’s great. See, he
1:10:40is currently studying a competitor's
1:10:41website.
1:10:42Studying the AUDB website.
1:10:43auditv.ru, which, as I understand, does
1:10:46the same thing,
1:10:47yes,
1:10:48what hypothesis have we chosen now?
1:10:50What positioning should we create to
1:10:51break into the market and stand out?
1:10:53And after this, what we are doing with
1:10:55you now, once we get all this
1:10:56information and facts, we will ask
1:10:58ChatGPT to write a complete guide for
1:11:00the developers, and we will launch it.
1:11:02So, we are literally 5-10 minutes away
1:11:04from having a product. Awesome. We just
1:11:06need to understand.
1:11:07Should we launch it as a low-code
1:11:08solution, or
1:11:09will we include it in a low-code build?
1:11:10Yes, a low-code build. Using Cursor
1:11:12will take a bit longer. It will be
1:11:14higher quality, but it will take longer
1:11:16. Well, basically, to launch, create a
1:11:18landing page, and start a prototype of
1:11:20the product with a demo account where
1:11:23people can log in,
1:11:24there won't be any problems.
1:11:27Moreover, when all of this is done, in
1:11:30fact, what will I, as an entrepreneur,
1:11:32do next? My mind is already on the next
1:11:36step. I will find one lawyer, run some
1:11:39test ad campaigns, collect leads, and
1:11:42explain to them what we want to do.
1:11:45I’ll find one lawyer who, manually
1:11:48and on the fly, will go through the
1:11:50process so that we can pass the fork we
1:11:53had—recovering 60%of fines,
1:11:55yes, and for 20%I’ll put the unit
1:11:57economics together. And after that, I
1:11:59won't be building a complex platform or
1:12:01complex code. I’ll create a landing
1:12:03page that collects leads. And it will
1:12:04all be done very quickly and cheaply.
1:12:06And this will be the first iteration,
1:12:08the first approach to the business. I
1:12:11think the problem with these fines is
1:12:13very relevant right now, because
1:12:15Wildberries and Ozon, firstly, often
1:12:17fine sellers unreasonably, and secondly
1:12:19, they are raising commissions, so the
1:12:22fight for efficiency and business
1:12:24economics is really important. There.
1:12:27But it's unclear what to do about it.
1:12:30Maybe an intellectual property
1:12:31protection product could be added here
1:12:33next. That’s also a very pressing
1:12:36issue. They copy content, they copy
1:12:39products, they copy texts, photos—
1:12:41everything, so in general, we need to
1:12:44protect it. Especially the Chinese
1:12:48sellers; on Ozon, already 20%of sellers
1:12:51in 2024 have crossed the 20th
1:12:53percentile, meaning 20%plus of sellers
1:12:56on Ozon today, are Chinese. That’s
1:13:01100,000 plus sellers. On Amazon, right
1:13:04now at the time of recording this
1:13:07podcast, 50%of the top sellers are from
1:13:09China. They are located in China. A
1:13:13good friend of mine was launching his
1:13:16project into the international market,
1:13:18and he was competing with the Chinese,
1:13:20and he said, "It's a completely
1:13:21different level of competition because
1:13:24you release a product, you get 10
1:13:25negative reviews instantly, and it's
1:13:27like, you know, playing a video game on
1:13:29Nightmare mode."
1:13:32And that's exactly what competing with
1:13:34the Chinese is like. But what I want to
1:13:36say is that it's not a competition,
1:13:37well, I mean, that's not what I'm
1:13:39talking about. I'm talking about how
1:13:40they copy all the content, the listings
1:13:42, and start selling what you're selling
1:13:44. On Ozon, I think there's even a
1:13:47feature to make a copy. So you can
1:13:49legally make a copy of the content
1:13:51yourself, entirely,
1:13:53of anything at all, and plug it into
1:13:54your own listing. They use this with
1:13:57pleasure, and blocking them takes some
1:13:59time. Moreover, for every block, you
1:14:01have to submit a ton of documents to
1:14:03tech support, which will then review
1:14:05everything for 10-20 days. So far, of
1:14:07course, this hasn't led to anything.
1:14:10Their sales are small, shipping is long
1:14:12, but it's clear that this is changing,
1:14:14that in America, it's already been
1:14:15Yeah.
1:14:16working for a long time. So, we are
1:14:18slowly heading in that direction too.
1:14:21Therefore, the fight for efficiency
1:14:24will intensify, and competition will
1:14:26increase. And you have to deal with
1:14:28these penalties. Let me tell you what
1:14:30happens next. So, next, once we've
1:14:32collected all the information,
1:14:33the product manager comes in and
1:14:36does this meticulous, hard work.
1:14:38Everyone thinks that being a product
1:14:40manager is a cool job. You'll be
1:14:41drinking smoothies and all that. In
1:14:43reality, there's a lot of
1:14:44responsibility, because you're actually
1:14:47providing the inputs that the most
1:14:48expensive resource in development then
1:14:51uses. And to do this, you need to take
1:14:53a few steps. I've been studying product
1:14:56management for 15 years, and I'd like
1:14:58to share some practices here. First,
1:15:00you need to clearly define the personas
1:15:02you are working with once again. Who
1:15:04they're also often called avatars,
1:15:06right,
1:15:07yes. Who exactly are the people who
1:15:09will buy your product and use it? A
1:15:11good example here is when you look at
1:15:13children's cartoons; they are
1:15:14completely different. Some are for
1:15:16girls, others are for dreamers, others
1:15:19for nerdy kids, and so on. This is also
1:15:22called target audience segments,
1:15:24representatives of the target audience
1:15:26who are described by certain
1:15:27characteristics. Here on the screen, I
1:15:29see you have characteristics:
1:15:31demographics, goals, motivation, pain
1:15:33points,
1:15:34yes,
1:15:36uh, technology immersion, that is,
1:15:39technological literacy, and typical
1:15:41usage scenarios for a specific user.
1:15:46Next, when we provide this product,
1:15:48people use it to do specific work,
1:15:51especially in B2B.
1:15:53Uh-huh.
1:15:53I mean, we all come to work, we have
1:15:55job descriptions and responsibilities,
1:15:57and we carry out these tasks.
1:15:58Accordingly, for our service, we need
1:16:01to clearly write down what task it
1:16:03solves, for which users, maybe reports,
1:16:05reimbursed expenses, and so on.
1:16:07Jobs to be Done is a very popular
1:16:09framework. You can definitely look up
1:16:11what that is. If you have never heard
1:16:13this phrase before, "Jobs to be Done,"
1:16:16just Google it or ask ChatGPT to
1:16:18explain what it is, dive into it. This
1:16:22will help make your product much more
1:16:24attractive to your clients.
1:16:28Then, once we’ve learned this, we
1:16:30must understand how we position our
1:16:32product among many others.
1:16:34This is, in short, an interesting point
1:16:37. Many think that marketing is about "
1:16:44buying traffic," getting some blogger,
1:16:46buying stories, or going out to buy
1:16:48outdoor advertising, or ordering
1:16:50something on TV during prime time, in
1:16:52the break between the halves of the
1:16:54World Cup final match—putting your ad
1:16:57there. That's marketing. No, marketing
1:17:04is the art of creating rankings where
1:17:06your company is in first place. And
1:17:11what makes you unique? What is your USP
1:17:14, your unique selling proposition?
1:17:16Unique value proposition. And here, it
1:17:19says "value proposition." What value do
1:17:22you bring to clients, and why should
1:17:24they choose you over anyone else? If
1:17:28the answer to this question is "I'm
1:17:30cheaper," you might as well give up,
1:17:32because whoever sells cheaper is
1:17:33essentially competing on price. Well,
1:17:36in price competition, the one with the
1:17:38higher purchasing power wins. And you
1:17:40can only earn something there on very
1:17:42large volumes.
1:17:43Yes, that's where economies of scale
1:17:45come into play.
1:17:46And
1:17:46the answer to the question "In what are
1:17:48you number one?" is also relevant for
1:17:50your clients. It is relevant for your
1:17:53employees and for candidates coming to
1:17:55work for you. Why—why are you number
1:17:58one as an employer? If you have an
1:18:01answer to this question, you will get
1:18:03cheaper staff, and you will get more
1:18:04qualified staff. For investors, for
1:18:07shareholders, yes, for everyone—even
1:18:10for the country. Why should the country
1:18:14, the government, or a tax inspector
1:18:16support you as a business, or whoever
1:18:18decides whether to give you IT company
1:18:21status and provide you with tax breaks
1:18:23or not? Why? A good answer to this
1:18:27question will help you everywhere.
1:18:29Next, once we have determined this, we
1:18:31need to understand what requirements
1:18:34our product has, what functions it
1:18:36performs, and what it can do. Well, for
1:18:38instance, authorization, the simplest
1:18:40version, role management, some key
1:18:42functionality in the dashboard, and so
1:18:44on.
1:18:44Then, basically, we move on to
1:18:45technical details for the developers
1:18:47about what exactly needs to be
1:18:48implemented. Business stakeholders
1:18:50usually don't think about this.
1:18:51Typically, a business stakeholder says:
1:18:53"Give me the moon from the sky." But
1:18:55they don't specify what size, what
1:18:57color, or where to put it. The
1:18:58developer has to come up with all of
1:19:00that themselves. Then, they present
1:19:02their vision to the business
1:19:03stakeholder. The business stakeholder
1:19:05looks at it and says: "Listen, that's
1:19:07not what I wanted at all, go redo it
1:19:09because I don't like it." Again,
1:19:12without diving into specific
1:19:13requirements, the developer fixes it
1:19:15based on their own understanding. This
1:19:18is endless, and in the meantime, they
1:19:20are given other similar tasks from
1:19:21other departments in the company.
1:19:23Without anything being defined, changes
1:19:25are requested, and it all snowballs. As
1:19:27a result, we end up with paralyzed IT,
1:19:30where anyone approaching the
1:19:31development team gets the answer: "We
1:19:34can take this on in a year at best."
1:19:37Our task queue is
1:19:39so long that we cannot take yours on.
1:19:41Nobody is getting anything done, nobody
1:19:44is getting anything, except for what
1:19:46they call HIPPO. Highest-Paid Person's
1:19:48Opinion. Meaning, the person who earns
1:19:50the most in the organization can walk
1:19:52up to the developers and say: "Drop
1:19:54everything you have in the queue and do
1:19:56this." And so the queue is not for the
1:19:58developers, but for these people. You
1:20:00have to prove why something needs to be
1:20:02done now to push the queue further. To
1:20:05prevent this, to avoid useless work, we
1:20:08describe requirements in detail. As
1:20:13Pavel Alferov once said—he is a
1:20:15professor at Skolkovo who teaches
1:20:17project management.
1:20:19He has this catchphrase, well, it's a
1:20:21catchphrase for me, I keep repeating it
1:20:23. The most terrible thing that can
1:20:25happen to a project manager is to make
1:20:27a project that nobody needs. And in the
1:20:29story I just told, the business
1:20:31stakeholder receives a product they
1:20:33don't need. This means the project
1:20:35management was done terribly, and we
1:20:37are excluding that here. Is the result
1:20:39there? Let's check. Yes, everything is
1:20:42there. And we won't even look at it
1:20:44right now.
1:20:45Yeah, there is a mile-long study here
1:20:47that you guys will look at later. We
1:20:49won't overwhelm you with it. The point
1:20:53is that some incredible magic happened
1:20:55here. Here, an analysis of dozens of
1:21:00sources on the internet has been
1:21:02performed, from which all information
1:21:04was gathered and a bunch of conclusions
1:21:06drawn; now we are asking to apply these
1:21:08conclusions to a product development
1:21:10framework. Which, by the way, can be
1:21:15done by simply copying and pasting the
1:21:17prompt you'll see.
1:21:18Look, this part is important, right?
1:21:20Let's go through what we did in order.
1:21:21We looked at the ideas, scored which
1:21:23one was interesting, then challenged it
1:21:25. We found a business model, meaning we
1:21:28took a seed of an idea and grew a
1:21:29flower to see what it might look like.
1:21:31Uh-huh.
1:21:32After that, we went and conducted
1:21:34research on whether this flower is
1:21:36valuable in the market or not. Now we
1:21:39say: "Okay, let's now, on the basis of
1:21:42this, uh, beautiful idea, make a
1:21:46product, let's plant a big garden."
1:21:48What could it look like? A garden made
1:21:50of these flowers. And I am writing the
1:21:52following task for it. I will give you,
1:21:54I will give you a template—the
1:21:56professional term is Product
1:21:58Requirements Document. These are
1:21:59product requirements, a "TZ" in Russian
1:22:01. Without a "TZ"[spec], you'll get "who
1:22:03knows what." Just like you were saying
1:22:05before.
1:22:05Without a spec, the result is "who
1:22:06knows what."
1:22:07Yes. Yes. The only thing is, when I
1:22:09have time, I'll definitely sit down and
1:22:11try to collect all the artifacts of the
1:22:13Soviet engineering school, because a
1:22:15huge number of things and studies that
1:22:17are being done were brilliantly
1:22:19described in the format of NIRs, OKRs,
1:22:21and so on. I—I studied at Phystech.
1:22:23And I...We laughed because Artyom told
1:22:27a joke at the beginning. He says: "What
1:22:29should I tell about myself?" I start
1:22:31listing, and I say: "Well, definitely
1:22:33say that you are from Phystech, because
1:22:35you know how to tell if a person is a
1:22:37vegan or if they studied at Phystech?"
1:22:39He says: "How?""Well, they'll tell you
1:22:41about it themselves for sure."
1:22:43Yes, I waited for 2 hours,
1:22:45it slipped out, I had to say it. It
1:22:47slipped out, yes. And I worked with our
1:22:50greatest engineers who created, I don't
1:22:52know, Russia's air defense protection.
1:22:55Some of them worked with Nobel
1:22:57laureates, and their culture is
1:22:59magnificent. So, the product culture we
1:23:01have now, which all IT people are proud
1:23:03of, no, it is, well, still quite
1:23:05youthful.
1:23:06It goes back even 300 years. It's a
1:23:08whole,
1:23:09yes,
1:23:10field of knowledge that, it seems to me
1:23:12, is somehow neglected, or something,
1:23:15and, uh, underestimated. Because if we
1:23:18were to train the model, say, on all
1:23:20the literature on TRIZ and apply it
1:23:22somewhere to our prompts about niche
1:23:24ideas, there would also be a lot more
1:23:27value-driven offshoots there.
1:23:29We launched de Persch. Right now, I’m
1:23:34giving him the format for market
1:23:36research requirements, as well as this
1:23:38big study we completed, which is the
1:23:40result of our challenge. Basically,
1:23:43I’m feeding him everything we did
1:23:45before and telling him: "Create a
1:23:46document for me." We already have the
1:23:48landing page ready. So, listen,
1:23:49everything is being recorded.
1:23:51What did you just do?
1:23:54I passed him the product description.
1:23:58The product description. He just came
1:24:00up with a product description and
1:24:02defined the avatars,
1:24:03right?
1:24:04He prepared the USP for them
1:24:06immediately. So, in essence, the
1:24:09marketing packaging for the business
1:24:11model we just invented has been
1:24:13completed. Well, not invented by us,
1:24:15but this is what's called vibe-coding.
1:24:17We just vibe-coded it.
1:24:18Let's try using a more powerful model
1:24:20now, because it can think through the
1:24:21details more thoroughly. And after that
1:24:23, we will take all these product
1:24:25requirements—you see them forming—
1:24:27and feed them into the vibe-coding tool
1:24:32, and it will build us a quick
1:24:34prototype.
1:24:34Yes. So, next, we’re going to use a
1:24:37tool called Lovable. It’s making a
1:24:43lot of noise in the US market right now
1:24:45, because it's one of the champions—
1:24:47this company is one of the champions in
1:24:49revenue growth rate, in terms of
1:24:51monthly recurring revenue. I mean, it
1:24:58took them something like 8 or 10 months
1:25:00to reach 10 million in monthly revenue.
1:25:03And that’s almost an absolute record.
1:25:05We’ll show you why right now. Just
1:25:07look at what he’s doing. He’s
1:25:09writing out the product in full detail,
1:25:11covering every aspect. Yes,
1:25:13the functionality is described right
1:25:14here. I see CSV file uploads,
1:25:17yes,
1:25:18from the Ozon personal account.
1:25:21And he goes on to score the
1:25:22functionality to determine which
1:25:24features should come first and which
1:25:25second.
1:25:26Yes,
1:25:27the information architecture is laid
1:25:29out here, and the detailed user flow is
1:25:31included. There.
1:25:32But the thing is, you still need to
1:25:33understand all of this, of course. We
1:25:35started by saying you won't need a
1:25:37super-expert, just an AI. But honestly,
1:25:39without 15 years of experience in
1:25:41product and software development,
1:25:43everything you’re doing now wouldn't
1:25:45be possible. One can repeat after you.
1:25:48One can repeat after me,
1:25:49but coming up with these prompts,
1:25:51knowing which ones are good, and
1:25:53stitching these tools together like
1:25:55this—it still feels impossible for
1:25:57most.
1:25:58I have a good friend of mine. Uh,
1:26:01but one can repeat after us. You can
1:26:03build a billion-dollar business. And
1:26:06a Muslim, he told me—he said, "How is
1:26:08it done in Islam? So, first you spend
1:26:1020 years growing up, listening to your
1:26:12parents and helping them, then the next
1:26:1520 years you create something in the
1:26:17world, and after that, you teach others
1:26:19. We’re turning 40 soon, and I think
1:26:21it’s time to start teaching others.
1:26:23When we were young, we couldn't even
1:26:25add two plus two. If only someone had
1:26:27explained that to us back then.
1:26:29That’s why I think that while it
1:26:31might be hard for some to figure out,
1:26:33I’m sure my nine-year-old son will be
1:26:35able to build endless games, services,
1:26:37and so on in just a year. I think we
1:26:41could compile everything from this
1:26:43podcast and all these dialogues—
1:26:45however many we have, five or six in
1:26:47total with ChatGPT—
1:26:51we could turn that into a neural
1:26:53network training course that explains
1:26:55step-by-step everything you need to do.
1:27:00All the prepared prompts are ready, so
1:27:01anyone can go through this training
1:27:03quickly. And that’s basically it.
1:27:06Alright, let’s move to
1:27:08We are moving to Lovable. I’ll share
1:27:10a lifehack.
1:27:11It’s a lifehack. And
1:27:13we’ll do it a little differently.
1:27:16We’ll write it for him—it's an
1:27:18international service, so it’s better
1:27:20to work with English prompts, yes.
1:27:25Generate an empty project. We are
1:27:28creating an empty project. Lovable is
1:27:33like ChatGPT, but instead of text, it
1:27:35generates websites, applications—
1:27:37everything you can view on your phone
1:27:39or in a browser, with a ready-made
1:27:42interface, menu items, and buttons you
1:27:44can click. I would say that, well, in
1:27:49development this is called a functional
1:27:51product prototype. Meaning it’s not a
1:27:53finished program that you can just go
1:27:55out and sell. It’s something you can
1:27:57test with users, show them, and ask:"
1:27:59Look, this is what it will look like,
1:28:01are you buying or not? "And so, Artem
1:28:03just entered the research results and
1:28:05the request here. What request did you
1:28:07enter?
1:28:08We went into Lovable, you can register
1:28:09there.
1:28:10You told it," Make an empty project. "
1:28:12Yes, I told it," Make an empty project.
1:28:14"And then, what we do is, it created
1:28:16the empty project. Okay, let it be like
1:28:19that. And then we go into the settings,
1:28:26yes,
1:28:28and here we enter a detailed
1:28:29description of the project we’re
1:28:31building into the knowledge base.
1:28:32Meaning,
1:28:33the one you got from ChatGPT,
1:28:35right?
1:28:35We just copied and pasted it.
1:28:37No matter where we go or what we do, it
1:28:38will always remember what the product
1:28:40is, what candles it has, and so on. It
1:28:42won't lose context; it will remember
1:28:43all of that. That’s its little
1:28:45instruction set.
1:28:45All right.
1:28:46And then, once we’ve done that,
1:28:47this is the result of ChatGPT’s work,
1:28:49yes. Uh-huh.
1:28:53Notice that I feel like we spent more
1:28:55time choosing the idea than actually
1:28:57implementing it. I mean, the
1:28:58implementation is very fast. If we had
1:29:00more time, we would have done all three
1:29:02ideas.
1:29:03I think, I think it was about the same,
1:29:05yeah. And now we’re typing" please "
1:29:07to it. Develop project. Build a service
1:29:15based on the documents you’ve been
1:29:17uploaded. That’s it, now we wait.
1:29:30Just for the sake of interest, I’ll
1:29:32run another service too. This is the V0
1:29:34service. I like it, I love it, I’m a
1:29:37fan of it. So, what is V0? V0 is a
1:29:43smart developer you can tell what you
1:29:45want to build, and it draws a really
1:29:48beautiful frontend product for you. It
1:29:52won’t have the backend part—well,
1:29:53you can do it, but it’s not really
1:29:55meant for that. But that most beautiful
1:29:57part,
1:29:58exactly. Basically, it’s also a
1:30:00functional prototype. What we used to
1:30:02build in Figma, InVision, or some other
1:30:04visual
1:30:05drag-and-drop editors like Canva, maybe
1:30:08, is now being done by AI
1:30:12through prompts. You just copied the
1:30:14same thing from the chat.
1:30:15Yes, I just copied it from there. So
1:30:17let it sit there, figure it out, and
1:30:18think for itself. It’s doing that
1:30:20right now.
1:30:20How much time does V0 take to work?
1:30:23About 5 to 10 minutes. The detail is
1:30:25that what you’ll see is that you have
1:30:27a page. And on that page, there’s,
1:30:29for example, a button
1:30:30to export a document. And it won't do
1:30:32that in the first version. So you have
1:30:34to ask:" Also, add this little detail,
1:30:36and that little detail, "and spend
1:30:38another 10 minutes on it. Okay,
1:30:40what’s happening in Replit?
1:30:43I think Replit has already reached a
1:30:44certain point. Right. Uh-huh. We have a
1:30:48landing page.
1:30:49Let’s open it right away. This
1:30:51let's
1:30:51in a separate one, yes, in a separate
1:30:54window. Framer.
1:30:55Oh, oh, oh. And here, see, there’s
1:30:58something here: home, login, dashboard.
1:31:01Right now, we have a dashboard. So,
1:31:03let’s go.
1:31:04Yes, let’s start from the beginning.
1:31:06Let’s. This is a finished landing
1:31:08page. In principle, it’s not bad, in
1:31:10my opinion. Return your fines in one
1:31:11click.
1:31:12Yes, what happened here? So, first, the
1:31:15positioning was developed, marketing
1:31:17headlines appeared—all created by
1:31:19artificial intelligence, which says:"
1:31:21Appeal fines in one click. "We automate
1:31:25WBY Azone fine appeals, helping you
1:31:27save more revenue without any manual
1:31:29work. And here, a landing page appeared
1:31:35that offers to start a free trial or
1:31:37watch a demo. So, Artem suggested he
1:31:41finish a few more things. Well, let's
1:31:44move on—here, yes, it's not just a
1:31:46landing page. What else is here? Here
1:31:48is,
1:31:49yes, let's look at the functionality
1:31:50itself, because of how it looks to the
1:31:52clients. The AI service came up with
1:31:54all of this right now.
1:31:55Automatic integration,
1:31:56yes, an appeals engine, one-click
1:31:59documents, real-time notifications, a
1:32:01performance dashboard, and an agency
1:32:03mode, which is also cool. Reviews on
1:32:06which is not only cool but for some
1:32:07reasons relevant. Right now, sellers on
1:32:10the market—it's very important on
1:32:12landing pages when you show how it will
1:32:14work. Meaning you connect, then go into
1:32:17the dashboard, check, download the
1:32:19documents, and send them off.
1:32:20Agreed. Pricing plans
1:32:2215,000 rubles per month, 10,000 rubles
1:32:25per month plus 10%.
1:32:27Uh-huh.
1:32:28Awesome.
1:32:29So, let's start with
1:32:30and the call to action at the bottom.
1:32:32Great. So, we are logging into the
1:32:34system.
1:32:34Ah, I hope it can be done just like
1:32:36that, if it lets us,
1:32:37yes. Well, generally, there is a login
1:32:40form, a registration form, and a
1:32:42ready-made personal account page that
1:32:44shows,
1:32:44how many fines you've received, how
1:32:46many came in during the selected period
1:32:48, and the potential refund. How many
1:32:50are currently in appeal and what is the
1:32:52success rate? Would you like to have a
1:32:55dashboard like this so your lawyer—
1:32:57well, in general, where it's recorded
1:32:59how many fines you have, how many are
1:33:01currently being appealed, how many were
1:33:04returned, the success rate, and a list
1:33:06of the latest fines, which I think is
1:33:08very cool.
1:33:09And right now, it's writing the
1:33:10functionality, and it will look like a
1:33:12fully ready service. And when you go to
1:33:14a seller, they won't realize it isn't
1:33:16real. Yes, you'll say it's a simulation
1:33:18of the data, and they'll think that it
1:33:20actually works.
1:33:22There's a publish button on the right.
1:33:23In general, you can publish it, and
1:33:25then it will appear at a public address
1:33:27.
1:33:27Let's see just for interest what Visero
1:33:29does. So, it's still developing it here
1:33:31.
1:33:31Visero is a more advanced tool, right?
1:33:33Here, as we can see, code is being
1:33:35written. Although V0 is being made, I
1:33:37think V0 is better, better. Yes. Get
1:33:39more messages with premium. Ah, well no
1:33:42, here it's all,
1:33:43Yes, on this screen, you can see a
1:33:44little circle spinning next to the word
1:33:46" page "in the middle column. This
1:33:49means it’s writing code right now as
1:33:51we speak, it's coming up with the code
1:33:53for this page,
1:33:54yes, and you can take a look. There it
1:33:56is, finished.
1:33:57And somewhere down there, it will
1:33:58clearly continue writing this story
1:34:00right now.
1:34:00Yes. When I talked to strong product
1:34:04managers and developers, they said that
1:34:06this is roughly one or two weeks of
1:34:07work for a highly professional team.
1:34:11Well, you have to understand that while
1:34:13it’s a week or two of work for a pro
1:34:15team, if you have a pro team, they
1:34:17aren't sitting idle,
1:34:18right,
1:34:18they have a project they’re managing,
1:34:20and they have a backlog of a million
1:34:22other things. And when you come with an
1:34:24idea, your idea gets in line with that
1:34:26team, the turn will come after some
1:34:28time, and then it will be those weeks
1:34:30of work on this story.
1:34:33It starts adding functionality.
1:34:35Well, in general, there’s an infinite
1:34:37space that follows. But, guys, what are
1:34:39we going to do with you?
1:34:40Look, here’s an integration module,
1:34:42please enter the token.
1:34:44Yes, it has already appeared here
1:34:46appeals.
1:34:47Here is a list of appeals. In short,
1:34:50there is a ready-made solution here
1:34:52that the AI has thought out, which
1:34:54looks very good. Anyway, write in the
1:34:57comments, guys, entrepreneurs, anyone
1:35:00who wants to take this on, Artem and I
1:35:02will invest. Some amount of money, I
1:35:04don’t know how much, a million rubles
1:35:05,
1:35:05yes, it's possible, whoever wants to
1:35:07take it on, come forward,
1:35:08yes. And we will, in short, join the
1:35:10board of directors of this business,
1:35:11take a share, and we’ll coach you. If
1:35:13you want to ramp this story up, or
1:35:15something else from the list, we’ll
1:35:16make a YouTube show out of it, about
1:35:18how this business develops. That will
1:35:19be cool. If you want to, mark yourself
1:35:21in the comments somehow so we can
1:35:23contact you about this, make yourselves
1:35:25known. Tell us about yourself, who you
1:35:27are, how we can find you. And that's it
1:35:28. We’ll get a topic like this going.
1:35:30Maybe someone is watching now who has
1:35:32long dreamed of having their own
1:35:34business. But we will be strict judges.
1:35:36It won't fly with us, no. We need you
1:35:38to be go-getters. There’s already
1:35:41functionality built here, personas,
1:35:43jobs to be done, marketing positioning,
1:35:46and a functional prototype has been
1:35:48made that you can click through.
1:35:50Everything necessary for engineers has
1:35:52basically been described. Artem didn't
1:35:54like the design at my suggestion just
1:35:56now. Well, yeah, it does look a little
1:35:58bit simple. I just want it to be a bit
1:36:01nicer. What Vizirov is still doing,
1:36:03what he is still up to. Well, let's not
1:36:05get distracted by them,
1:36:06let's just leave it at that. That's
1:36:08enough for an example, guys, because
1:36:09what just happened? We sat down here
1:36:12two hours ago to chat. While I was
1:36:15drinking my coffee and Artyom was
1:36:17drinking water, we found several
1:36:19promising niches. We picked out the
1:36:23ones that were the most appealing in
1:36:25our view, and in the AI's view, and
1:36:26conducted deep product research, which
1:36:28you can take a look at. We made a
1:36:31functional prototype. The next steps,
1:36:35in general, were described to us in the
1:36:37AI's plan. These are test advertising
1:36:41campaigns, first clients, starting
1:36:43development, and validating hypotheses.
1:36:48I would take this to some company that
1:36:50specializes in software development
1:36:52outsourcing. Yeah. Overall, it's not
1:36:55hard to find. There are industry
1:36:57ratings where you can find many
1:36:59outsourcers. And I would go to get an
1:37:01estimate of how much it costs to
1:37:03develop the first version of such a
1:37:05service and how long it would take.
1:37:07Then I would get confirmation from the
1:37:08market with real money. As an
1:37:10entrepreneur, I would want to get some
1:37:12pre-payments, like—it costs 15,000
1:37:14rubles——pay me 1,000, and I'll give
1:37:17you first access as soon as it's ready,
1:37:19or I'll return the money 2x, for
1:37:21example. Yes, I would collect 50 such
1:37:24pre-payments and then I would go to a
1:37:26venture fund for investment and say," I
1:37:28have deep research, I have a prototype,
1:37:31I have a development budget, I have
1:37:33market validation, and here is how much
1:37:35money I've collected. "At the same time
1:37:38, I only spent, well, 2 hours on
1:37:41vibe-coding. I spent about a week on
1:37:43Zoom calls with the potential target
1:37:45audience. Yeah. And I'm risking 50,000
1:37:49rubles. that I'd have to pay out if
1:37:53nothing happens. And that, by the way,
1:37:55is my negative motivation, my carrot on
1:37:57a stick from behind, that I will lose
1:37:5950,000 rubles. in payouts. Well, anyway
1:38:02, that's it.
1:38:04But I would still listen to the AI,
1:38:06take it, publish this landing page
1:38:08right now, and run an ad on a Telegram
1:38:11channel.
1:38:12Yes,
1:38:13right now. As usual, guys, please write
1:38:17in the comments, what do you think: in
1:38:19the 6 months the AI suggested, will it
1:38:21be possible to take this business model
1:38:24from zero to revenue? It said, by the
1:38:27way, not one million rubles, I think it
1:38:29was 3 to 4 million rubles a month.
1:38:32With market profitability. The future
1:38:34is already here. Yeah. But it's up to
1:38:37you to judge whether it will work or
1:38:38not. Let's wrap it up. If you have
1:38:41anything else important to say, go
1:38:43ahead, we need some kind of cherry on
1:38:45top.
1:38:47I will finish with what I started, what
1:38:49we were talking about, that everyone
1:38:51thinks artificial intelligence helps
1:38:53you work better, but I tell everyone
1:38:55that you can start doing the impossible
1:38:56. And what each of you can do is try to
1:39:00repeat what we've done and accomplish
1:39:01things you’ve never been able to do
1:39:03before. Let’s finish then with a
1:39:09thought, maybe a scene from an Isaac
1:39:12Asimov book, where the main character
1:39:14talks to a robot and says:" You’re a
1:39:16soulless machine, you’ll never be
1:39:18able to write a symphony, create a
1:39:20great masterpiece, or come up with a
1:39:22brilliant poem, because no matter what
1:39:24kind of artificial intelligence you
1:39:26have, it's impossible for you, you
1:39:28can't do it. "And do you remember what
1:39:34the robot answers him? The robot says:"
1:39:37But can you? "And on that note, well,
1:39:40that’s not my scene, guys. Robots are
1:39:42evolving, we can be friends with them,
1:39:44and we can earn money with their help.
1:39:46Please, leave a like, hit the bell,
1:39:48subscribe to our channel, and write
1:39:50your biggest takeaway from this video
1:39:52in the comments. That will help you
1:39:54promote it. And we will see you in the
1:39:56next video.
1:39:57Bye.

About this transcript

This page contains the full transcript of Как построить бизнес с Chat GPT? Все промты в видео! Артем Круглов by Николай Хлебинский | AI Action Plan, generated from the public captions YouTube serves with the video. The transcript has 16,756 words across 2,683 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.