Oleg Melnikov

should we eliminate all entry level jobs? i can remember some of the skills that i became good at: - chess (played for 8 years, started at the age of 5) - maths problem solving (was top-12 in russia under the age of 18) - coding competitions (codeforces international master) - trading (worked at a $1B high-frequency trading fund as a quant) - content marketing (20k+ youtube subscribers) every time it started with playing, bullsh*tting around before I started understanding stuff and delivering value and if before it was necessary, now it looks like a wasted time and money for the employer cause you can just shortcut a lot with ai as sad as it gets, i think most entry jobs are lost and we have to find a new paradigm instead of trying to bring the past back maybe it's shifting towards new professions and redefining existent professions for instance, you can be an expert in time management, that teaches specifically how to manage your time when orchestrating multiple ai agents this job never existed before, but has high demand and roi now wdyt?

AI made shiny object syndrome 10x worse now every new idea looks like a one-day build each one walks by like the woman in the red dress from The Matrix you keep jumping to the next new idea and nothing gets finished. it feels like you can build anything now, so the moment the thing you're building gets hard you think AI could knock out that other idea in a day, so why not switch? I'm building an AI media company and I get the same urges a while ago me and the guys built a Chrome extension that reads any terms and conditions you're about to accept and tells you in simple words what they really say. we built it and never launched it once the first dopamine rush wore off it needed marketing and a lot of content, and everyone had more important stuff in the main business. so it's on pause so now when the main thing gets boring I want to find another problem inside my own business and chase that one

just dropped my collab with Higgsfield showing how to build viral Instagram pages with Claude Code for the page management, coming up with ideas, finding trending music and controlling Higgs through MCP I wasn't expecting this video being so hard to produce tbh figuring out the cinematic image generation prompting & realistic video prompting ended up with this stack: FLUX Pro 2.0 for cinematic images Seedance 2.5 for videos (obviously) and I made the whole system completely open-source full guide here, enjoy: youtu.be/rXb8ThUyuFc

i met the founder of Revolut by accident, in a breakfast line in Dubai a month later he told an interviewer why 138 of 140 startups in his first office died it was a saturday in september. Brix on Jumeirah, the place is always packed on weekends midday, no reservations. i waited about 45 minutes and ended up sharing a table with three Russian guys from VC, 10+ years in the industry then Nik Storonsky walked in. with his wife, pregnant with their fifth kid i just went up to him and shook his hand. told him i watched his interviews when i was 16-18 and they were a part of why i became an entrepreneur at all. him and a few other founders who built their companies in public back then in the interview he talked about his first co-working space, 140 companies. after 5 years 3 were alive, after 10 only two, and all of them worked weekends. so he says weekends are a must, "purely statistically" i work 6 days a week. tuesday to sunday. monday is a day off for me. i made it that way because Dubai is too busy during weekends, not a fan of that. i walked out of that breakfast realizing i'm not thinking big enough. the problem my business solves is too small. this guy picked finance, one of the biggest problems there is, and has been solving it for 15 years now so i'm changing the problem i'm solving. still marketing, just much bigger (will share soon) P.S. he indeed has huge bags under his eyes

264 AI models have come out since 2022. I bet you remember about 10. Mine: ChatGPT, 1 million users in 5 days. DALL·E 2 and that astronaut riding a horse. Sora and that woman walking through Tokyo. Absolutely surreal. Nano Banana and the 3D figurine trend, the first truly insane images. Claude Opus 4.8, a holy f*ck moment in Claude Code. Claude Fable 5. What the f*ck is happening? (3 days later a US order switched it off.) My bachelor's thesis at SPbU was built around BERT, Google's model from 2018. We did it together with JetBrains. And in July 2025 an OpenAI model reached gold-medal level at the International Math Olympiad. In high school I was one of Russia's top 12 math students under 18. I got to about silver level and was trying to get that gold myself. The video has all 264 that OpenAI, Google, Anthropic and xAI released for public use, including ones I don't remember at all, like GPT-Realtime-Translate, Grok Voice 1.0, o3 Operator and Gemini 2.5 Flash TTS. Every model I still remember comes with a moment I can picture, like the astronaut on a horse or me in high school chasing that gold. I think founders get remembered like that too. I help B2B founders tell stories people can picture. Need a hand? 👇 boldane.com/?utm_source=x&ut…

AI agents will prefer Linux over any other OS. Just like Claude Code preferred terminal over UIs. I've started noticing this idea flowing over X. From Tibo, the Codex lead at OpenAI and others. Here is why this is happening, why Linux is better than Windows or MacOS for agents: 1. No locked doors. Mac and Windows ask "are you sure?" all day. On Linux the agent can touch anything. 2. Everything is text. Settings are plain files and every job is a typed command. Agents read and type. They are bad at clicking buttons. 3. Free copies. One agent or 1,000 agents, no license to pay. Each gets its own fresh computer in seconds. 4. It is already their home. Almost every server runs Linux, and agents work on servers. They also learned from mountains of Linux help pages. 5. It can be tiny. Cut the screen, the apps, all extras. Small means fast and cheap. Terminal was something "for devs" and has became a love for normal people. I remember using Cursor for 2 months in the late 2025, then switching to Claude Code via terminal and falling in love with it. It's been a pleasure to build all the AI software for marketing that we've built using this stack ever since. Linux is probably next.

I think AI labs have a simple goal: Make sure their product solves any problem you face, with the least effort. Claude Chat and Cowork have just merged yesterday. OpenAI just launched their specialized version of Astra that solves Law and Legal tasks for you. The trend is clear: AI labs want to simplify stuff and cover more use cases (all of them eventually). I see the future where they become super apps. Whereby their app includes finance, taxi, travel, delivery, and other lifestyle products. For the majority of the year I live in Dubai. In Dubai we have a super app called Careem. I use it everyday for these needs: - taxi - food delivery - home cleaning - ecom shopping (instead of amazon) - even flowers delivery and probably a few other use cases that I already forgot 😁 It's already convenient. But if Anthropics and OpenAIs of the world handle it, the quality will be even better, I guess. Because these guys understand you on a way deeper level. They have the data to know what you want and when. And apparently it can give them 10x more data, and 10x more profits. So I think the future is headed towards one AI super app. P.S. It can get scary at times, observing the rapid expansion of AI labs. Like, is my business going to survive at all?... Honest answer: I don't know. P.P.S. I'm still going after the AI software for media. I think the media market (organic content and ads) is very inefficient.

ElevenLabs can now dub my voice into flawless English. I'm Russian, built over 20k YouTube subs without it, and I still wouldn't use it. Ninety plus languages, in your own voice, straight from their API. I was born and raised in Russia. Lived there for the first 21 years of my life. English is not my first language. Learning it on an actual communicational level, with an accent that sounds adequate, felt like learning to breathe. Without it I felt disabled. I felt like I could not do international business. If I had dubbed my way into an English channel, I would have skipped the reps that now let me sit on a call and close a B2B deal with someone in the US. I met Thomas Kralow once, at his villa in Dubai. Crypto guy, three million plus followers, originally from Latvia. Speaks perfect Russian. Posts everywhere in perfect American English. I asked him how he trained the accent. He said it came from making thousands of videos, just pure reps 😁 So I did the math on my own reps. Roughly 170 pieces of video so far. About 50 long form, 120 short form. Every single one of them is my own voice, my own mistakes, my own accent, getting a little better each time. P.S. Kralow lived in New York for 4 years too. He told me the videos did more for his English than the city did 🙌🏻

This table shows how desperate Claude and ChatGPT are for new users... Claude's $200/mo plan buys you $8,000/mo in tokens 🤣 I found this stat on X, a few guys run an actual test, comparing the subscription price to the actual API cost of the same amount of credits. (in case where you're maxing out the plan) With these numbers, how could the AI labs be profitable at all? 🫧🫧🫧 (this is supposed to illustrate a bubble 😁) I get the argument that they are startups, and they have to capture the market faster than another guy. And that they need to win the "AI race". But it's not like with Uber, where building a monopoly yields a better business for everyone. In case of taxi apps, more drivers + more buyers = more money / less cost for everyone. And aggressive market expansion justified the cash burn. With AI there is no victory for everyone in one company being a monopoly. If anything, it's a loss. This prices are so ridiculously low, that you can be literally running your whole service business, serving dozens of clients with just one $200/mo Claude subscription. (if you have at least minimal understanding of how to manage the context window and model selection efficiently)

Creator of Claude Code ( @bcherny ) just shared 5 levels of AI users: from using 0 AI agents, all the way up to using 1000+ agents at once. The majority of engineers at Anthropic are now at level 3 (100+ agents), while he just stepped into level 4(1000+ agents). At first, it blew my mind and I literally couldn't digest the reality we are entering now... Let me walk you through the levels, how to grow from one to the next, and where I'm at: Level 0: The Chatter You are using ChatGPT, Claude inside the chat. ...to upgrade: start using Claude Code... Level 1: The Micromanager You are using 1 agent. That's decent, but you are checking every single part of its work and it's tedious. The thing that keeps you stuck is trust. ...to upgrade: start using --dangerously-skip-permissions command and /code-review... Level 2: The Orchestrator You are using 10 agents. Not bad, you have some leverage. You are only validating the final results of the agents' work. Now the thing that keeps you stuck is that reviewing the results takes a lot of time. ...to upgrade: start using /goal, ultracode, /schedule... Level 3: The Manager You are using 100 agents. Aweeeesome. You are a high-producer. Now the thing that keeps you stuck has nothing to do with AI. It's all about the clarity of your business vision. A vague vision wastes an army. ...to upgrade: start using Claude Agent SDK... Level 4: The CEO 👑 You are using 1000+ agents. Now events launch agents, not you. Claude Agent SDK works like that: instead of you launching Claude Code inside your terminal, it now launches on the backend, when needed, without you. P.S. I was at Level 2 before reading about that, now I'm at Level 3. P.P.S. Funny that the guy from Anthropic shares that, because this basically encourages people to spend more tokens (and make Anthropic richer $). But I believe the ROI is worth it.

Anthropic accidentally faked their benchmarks for Opus 5. In agentic coding Fable 5 is better, but the table highlights Opus 5 as a winner 😁 Happy IPO guys!! P.S. I know it's a minor thing, and hype is hype. Still a huge fan of Claude Code.

In 2026 I wrote 7,377 prompts for AI. All via voice through WisprFlow. 633,225 words dictated. Sometimes using Russian curse words like "tuporiliy pid***s" or "b***t" 😁 It's crazy how much my life is different now compared to 6 months ago. No more typing, no more micromanaging AI agents. Just 10 minutes of dumping raw ideas and context inside a prompt via voice. And then just let Opus 4.8 or Fable 5 (on the days when I'm rich) do it's thing for 1 hour while I'm busy doing other things. I can do team communication, clients delivery, sales calls or YouTube videos recording. All while hundreds of agents (thanks to /goal and ultracode commands inside CC) do software heavy lifting. This is a huge reason why we were able to improve the service at boldane.com with rapid speed. We live in a super accelerated AI world. To be fair though, it's not evenly distributed and a major piece of population is still lagging. Anyways, huge fan of dictating and shouting at my agents. IDK how people who work in open spaces do that. Do you even chat with AI via voice? I think WisprFlow made something to make whispering enough. But I'm still not sure how comfortable that is. Sending a ton of love to @WisprFlow 's team 🖤

I'm 25, and as a founder here are 5 simplest things that make me happy every day: - deep focused work (6+ hours) - workout (30 min) - meditation (3 times 5 min each) - offline by 9 pm - last meal before 7 pm, no coffee after 12 pm Focusing on fundamentals.

how i quit my 6-figure job in amsterdam and moved to dubai, in the pursuit of my dream it all started 3.5 years ago - i began my career as a researcher in a trading fund in amsterdam the job was hard, competitive, and addictive. it was pure science, but directly applied to finances and the goal was to predict assets’ prices in the future (like in any type of trading lol) and to be honest, to me it was about making money and becoming financially independent from the day one (money meant freedom to live the life i wanted, not to buy expensive stuff) i knew it wouldn’t be the industry where i will spend the rest of my life, i knew i wanted to build something that people use, i knew i wanted to inspire people at scale and that does not happen in trading so after i made enough money (to comfortably live for 2 years) I quit and moved to dubai the very next day (wondering why dubai? - that’s another story) what was on my mind at the moment? - i wanted to build an audience on youtube, talk about ai automation and build some sort of business around it long story short: after 8 months of living in dubai and making absolute ZERO money i closed my first b2b deal with a business from the US - the client came completely organically from my youtube and this summer my international youtube channel has crossed 10k subs all while i was shy to make content in english, not confident of my accent and feeling the imposter syndrome but i did it anyways i didn’t earn anything close to what i was making in trading (yet), and i’m still a rookie when it comes to business but i take risks, and live my life on my own terms which no money will ever buy

i've learned that growing an audience doesn't require fancy production. you need undeniable value (i've grown my YouTube to 13k subs) here's the story about my single most viral video. i made a free database, a simple searchable list, of AI automation templates from top YouTubers for businesses. templates are plug and play workflows that complete tasks using AI tools. type a query like, AI automation for CRM. you instantly get a workflow you can deploy in 3 minutes. the video script was simple. here is the database. here is how I built it. here is free access. no complicated production. just pure value, right in your face. it did over 87,000 views and added 5,600 subscribers. i was at 2,500 subscribers. one video exploded the channel. people were freaking out. just give value, bro.

Stop being politely vague. If someone invites me to an event a week or more away, I answer immediately. I say, no, I won't go, and here's why. No more 'I'll think about it' or 'let's decide later'. Directness earns trust, especially with people you admire.

i collected 37 best n8n templates for marketing and giving it away for free 🙌🏻 they cover every marketing use case: - content creation - SEO - ads - email marketing - cold outreach these templates are coming from top n8n builders like Nate Herk, enjoy evolva.notion.site/best-mark…