Maksym Liamin

My name is Maksym Liamin. I built an AI company to more than $1,000,000 ARR in less than 12 months with the BIGGEST automotive brands as our clients while being 21 years old. It’s my first post on X, so get to know me: I was born in Odessa Ukraine, went to a public school in Kotovski district where I always excelled at classes and ended up liking to learn foreign languages. I learnt English and German to fluent levels during my school years. At 16 I won a grant while being one of three total winners from my country and got my ticket to free studies in Germany, namely in Hannover. At 17 I left my home to study in Hannover, was living in a student residency, where later at 18 met Emilio, cofounder for all my future ventures. He was already doing business back then and taught me everything he knew. My first business was an online cake shop. I found 2 Ukrainian ladies from whom I had ordered a cake once and then offered them to work together where I was handling marketing and sales and they, the cakes. In less than a week we filled the calendar with orders for the next 2,5 months, Ukrainian Facebook groups were too powerful back then. I still remember my first cash paycheck of 60€ which I picked up delivering the cake by myself to another city. Later I was doing marketing for local nightclubs and finally united with Emilio and one more guy to do business together. We didn’t know what to do, but picked to do YouTube marketing for entrepreneurs. We never made money with it and lost our third cofounder, but met our mentor Itay, a big businessman and YC alumni who helped us a lot in our growth. Emilio’s visa ended and he flew to his home country Mexico. After that we tried tons of different things but ended up doing TikTok reposting where we got to work with one of the biggest crypto influencers. During that time we got offered a contract with a retail chain of jewelry in Mexico City to put them online. I flew all the way from Germany to Mexico to work on that thing. We had a 2 months trial to make it work and we failed miserably. During that time though, we gathered a huge network of UGC creators and tried selling those services which also didn’t end up profitable. Lastly, we did market research for one CPG company founder. After that we got completely fed up with marketing and decided to pivot to AI. None of us knew how to program. We decided to automate the market research that we were doing with AI. We built it, but it was constantly breaking and our client didn’t end up buying it, but we managed to sell some of those automated reports to other people and make some money. Afterwards, I wanted to learn to program better and took time off business to fully dedicate myself to coding. I was programming day and night, started publishing my projects open source on GitHub and found Brandon Hancock’s YouTube channel, which taught me a lot. During one call with Brandon I told him he should do a community around AI and he did it. A lot of people joined in and I was the first user and the biggest contributor to it. I met a lot of great people there, Brandon himself helped me get my first contract as a software engineer, later I also worked with him on his course launch. After getting some cash with AI, I still wasn’t convinced in my skills so I decided to get a corporate job as a software engineer to really test myself. I started sending resumes without getting a single answer, then started lying in all of them and in less than 4 weeks I ended up getting a job at one of the biggest banks in Germany for a position that didn’t even existed and was opened specially for me. So I’m working my corporate job part time, study full time and at the same time I also get an offer from one of the Brandon’s community managers to work on AI product for forensic psychiatry. I didn’t have a clue about what it is but went really deep into it and due to this project made my first 6 figures. At that time I was 20 years old making $20k per month.

It literally looks like an iPhone. And I mean that in the worst possible way

@IterIntellectus Looks like iPhone 5c

what is this GARBAGE?! 😭

@levelsio Isn’t it same what happened with AI models? DSLR is to train / fine tune your own model, iPhone camera is to pay to OpenAI/Anthropic/Google for tokens

This was directionally accurate

@BoilerPlateCPA Got it! You don’t partner with universities to get fresh graduates to go directly to you for junior roles?

@maksliamin We’ve used a combination of self posted job listings, plain old social media posts, and recruiters Can find em anywhere but not easy

@adilmania 📝📝📝 IFYKYK

steve jobs was not just a CEO. he was taste. rebellion. the artist against the beige box. “the crazy ones” was not an ad campaign. it was a recruitment spell.

@paulg the irony is, if you solve a problem for a set of users who need it so desperately that they'll use it even if no one else does, they'll eventually share it to their group of peers and you'll start getting traction

A startup idea that only works if there are already a significant number of people using it is not a valid startup idea. There has to be some subset of users who need what you're making so desperately that they'll use it even if no one else is.

@sharran idea means nothing without execution. walk the talk

Ideas are cheap and getting cheaper. I can tell within about 10 minutes of meeting someone whether they're a builder or a talker. Builders ship. Talkers have nicer decks.

Founders complain all the time that legacy workforces refuse to adopt new technology. They say 50 year old workers are too stubborn to learn AI. That is just a lie founders tell themselves to cover up bad product design. We currently have a 38.5 percent DAU/MAU ratio. And our core users are 30 to 65 year old car salespeople in Mexico. They adopt our AI at the exact same rate as the 22 year old rookies. They are not anti technology. They just hate friction. When my cofounder and I first built our product, we put it in a beautiful web app. We showed it to the dealerships. Nobody used it. That is when we realized that AI adoption is not a software engineering problem. It is a human behavioral problem. If you want a traditional workforce to actually use your AI, you need to stop building dashboards and start following a very specific framework. Here is the exact 3 step adoption framework we used to capture 33 percent of the Mexican automotive market. 1. Hijack an existing habit If you ask a 50 year old salesman to open a laptop, navigate to a URL, and authenticate a login, your churn will be 100 percent. Every new step is a drop off point. We completely killed our web portal and moved the entire AI operating system into WhatsApp. That is where they already spend 12 hours a day talking to clients and family. No downloads. No passwords. No new habits. We just became another contact in their phone. 2. Unblock the money workflow Frontline workers do not want to chat with an LLM. They do not want a general assistant. They want to get their commission checks. We found the exact bottleneck that was losing them money: searching through complex pricing Excels and technical PDFs while a buyer was waiting. We didn't give them a tool to play with. We gave them a tool that gave them the exact pricing and photos they needed in 5 seconds so they could close the deal. Make the AI directly responsible for them making more money. 3. Zero learning curve If your software requires a 10 minute onboarding video or a corporate training seminar, it is already dead. We made the interface natural language. If they know how to text their wife, they know how to use our software. They literally just text the bot: "Send me the interior photo of the blue SUV and the 48 month financing quote." The bot replies instantly. Zero training required. Most AI products fail because they are built for impressive Twitter demos, not for the messy reality of a daily workflow. Adult users do not hate technology. They just hate software that forces them to change how they work. Meet them where they live, unblock their workflow, and make it invisible.

I spent hundreds of hours sitting inside car dealerships before we ever reached product market fit. Just watching how the floor actually operates. When you spend that much time observing, you realize that a lot of assumptions we make in tech are just wrong. I used to think closing a deal was about having the perfect pitch or being a master negotiator. It's really not. A salesperson does not lose the sale because they cannot sell. They lose it because the information they need is scattered across five different places. Picture this. A client walks in. They want a specific model, in a specific color, with a specific interior. But you don't have that exact car on the lot today. The client asks a technical question about the motor, and wants to know what the monthly payment looks like. Here is what the salesperson actually has to do to answer that: First, they open a heavy, complex PDF from corporate on their phone and squint to find the exact motor specs for that specific trim. Then, they spend 10 minutes scrolling through random automotive Facebook groups trying to find a real life photo of that specific color and interior to show the client. If they can't find it, they drop a message into a chaotic, 100 person dealership WhatsApp group asking if anyone has a video of the blue version, and just hope someone replies. Finally, they navigate a massive, cluttered Excel spreadsheet to figure out the exact pricing and this month's financing promos. By the time they gather all this information, the momentum is completely dead. The client lost the emotional high of the purchase. They say they need to think about it, and they walk off the lot. Salespeople are literally losing commissions because of fragmented data. The salesperson who wins the deal isn't necessarily the smoothest talker. It is the one who gets the right answer to the client the fastest. That was the lightbulb moment for us. They didn't need a new CRM to log their calls. They just needed immediate answers. So we took the technical PDFs, the pricing Excels, the car photos, and the walkaround videos, and we put all of it into a single AI contact inside WhatsApp. Now, when the client asks about the sand interior or the 48 month financing, the salesperson just texts the bot and gets the exact photo and the exact quote in 5 seconds. The momentum never drops. If you want to help frontline workers perform better, stop trying to teach them how to do their jobs. Just unblock them. Put all their scattered information in one place, exactly where they already work.

@RealHoldenGR @aaron_epstein @ycombinator bruh

@maksliamin @aaron_epstein @ycombinator vibecodestats.dev - obviously tracking your claude code tokens will have a chokehold on society, you just wait.. .

Every generation has a company that seems inevitable. Microsoft in the 90s. Google in the 2000s. Facebook in the 2010s. Anthropic/OpenAI now. It always feels like it's different this time. It never is. Startups always find a way.

THE VERTICAL BLUEPRINT Now here is what I would do right now if I were starting from zero and I want yall to understand the energy I am bringing to this. I will find the most chaotic, traditional, fragmented enterprise vertical where the data is unstructured, the workflow is a mess of PDFs and Excel files, and the stakes are tied directly to revenue. I will ignore the generic "AI assistant" play entirely. I will pick the single most painful workflow that directly impacts how much money the business makes and I will automate exactly that. I will ingest their terrible, unstructured data and turn it into the proprietary training signal that becomes my permanent moat. I will make sure the software requires zero behavior change. I will integrate into their existing practice management systems so the value is frictionless. I will expand from that initial wedge into every adjacent workflow until my software touches the entire lifecycle of their client. I will embed the software so deeply into their daily habits and firm specific AI training that leaving the platform would mean rebuilding years of institutional memory from scratch. I will eventually offer to just run the operations for them, turning my SaaS into a systemic infrastructure dependency. I will do this while every other 24 year old founder is building a generic AI wrapper for tech workers and complaining about Claude on Twitter. And in five years I will be sitting on a platform worth billions serving an industry that most VCs think is too boring to touch. The playbook exists. It is sitting right there in the open. The only thing missing is someone willing to do the heavy lifting to execute it.

FROM SAAS TO INFRASTRUCTURE They added a feature called Mirror Mode. The AI learns the exact drafting style, structure, and language of the specific law firm's winning documents. It replicates the senior partner's brain. Now look at the retention. How do you churn from a software that knows exactly how your firm writes, manages all your medical data, and directly increases your settlement payouts? You do not. You pay them forever. And now they are going for the throat. They just launched PLAAS. Pre Litigation as a Service. They are combining their AI with US based case managers to handle the entire claim setup, care management, record retrieval, and negotiation for the firm. They are no longer a SaaS company. They are an AI enabled operating system running the actual business. And not a single piece of this was built by someone trying to build "the ultimate AI legal assistant" for everyone. They picked the messiest, most overlooked, paper heavy vertical imaginable, understood the exact bottleneck that makes those lawyers money, and built a monopoly on top of it.

Day 1 of dropping actual game that most people on this app have never heard of. EvenUp. Most of yall have never heard this name. It does not trend. It is not on any podcast you listen to. There is not a single influencer thread about it. 🧵🧵🧵

@emonuxui Data and user adoption I’d say

@maksliamin If it still works because of data, workflow, distribution, or system design → it’s a real product.

how do you distinguish an actual AI product from a GPT-wrapper?

@ChiefSnack I'm really curious about the Telegram acquisition. Can you talk about it?

Two years ago I moved to San Francisco with zero connections in tech. Today I'm running Sirius through a16z Speedrun with a team of 15. 2 companies built. A Telegram acquisition. An a16z cohort. This city rewards builders. Everything else is noise.

@battleangelviv in which hotel was that?

Airbnb > hotels Solely for the reason that an Airbnb won’t have housekeeping showing up at 7am