AI Agents for Business from Scratch
April 11, 2026
This article is based on my talk on April 11, 2026. 20 minutes, real cases — everything here is the same, just in text form.
IIlya Chernyak
Hey everyone! My name is IIlya — yeah, you get it.
I’m an entrepreneur, been in business since 2010. Right now I’m running three projects: IZI electronics stores, Imperiya vape shop chain, and Playback Rental — a photo and video equipment rental service.
In February I dove headfirst into AI agents and started applying them across my businesses. In two months, my team of seven agents and I got more done than what would have taken half a year before. Today I’ll show you real cases — what worked, what didn’t, and why this changes the game entirely.
Agent System — a Revolution in AI
ChatGPT is an assistant with no memory. Memory like a goldfish. You invest time, explain things, teach it — and a week later it’s forgotten everything.
An agent is a different story. It’s an AI that lives on your computer. It has access to your files, memory that doesn’t get wiped, and it can do everything a human does at a keyboard — install programs, create files, publish to the internet.
Claude Code + OpenClaw = an environment for building a team of AI agents. Three key differences from regular neural networks: memory, computer access, specialization.
My Team — 7 Agents
The team is seven agents, each with their own specialization. A coordinator, an SMM guy, a scout, a designer, an engineer, a monitor, and a strategist. Communication happens through Telegram or the terminal. Agents work right on the computer — they modify files, install apps, deploy websites. Everything a person can do at a keyboard — an agent can do too.
Business Automation — a Network of 12 Locations
The first thing we did was give the agents API access to our warehouse system, MoySklad. We have a network of 12 retail locations, 24 salespeople, thousands of products. All the data was already in the system — but MoySklad can’t do out of the box what a business owner actually needs. Agents can.
Now every evening at 9 PM, after the stores close, a summary lands in Telegram: revenue for the day, number of receipts, average ticket, the day’s leader and the weakest location. If a top-30 product is running low — a red flag right in the message.
Every Monday — a weekly report: location comparison week over week, salesperson rankings by revenue and average ticket, who’s growing, who’s falling. You can instantly see who needs attention.
On the first of every month — full analytics for 3 months: location rankings with trends, product analysis (what’s selling, what’s frozen), dead stock with amounts, specific recommendations with calculated impact.
Insights the agent found in the data: the only guy on a team of 24 girls rose to first place in average ticket within two months — time to rethink the hiring strategy. One saleswoman showed +155% growth without changing locations — we’re formalizing her method and training the rest. Another one — a red flag, worst performer every third day. Five locations with low average tickets — upsell training there gives a noticeable boost to net profit. And the big one: 21% of products generate 80% of profit — we set up automatic stock monitoring so key items never run out.
For a multi-business owner, this is the perfect assistant — you’re always in the loop without being buried in operations. The agent digs through data on its own, finds anomalies, and delivers ready solutions, not just numbers.
GEO Optimization of the IZI Website
GEO optimization is content optimization for AI-powered search results (ChatGPT, Perplexity, Yandex). People are stopping their search engine habit — they solve problems with neural networks instead. And a neural network doesn’t give you a list of links — it gives specific recommendations: what to buy and where.
Agent Leo added a “News” column with articles tuned for AI search results to the existing IZI website. SMM content for social media is generated based on these articles.
SMM Agent Izya — 5 Platforms
After the GEO optimization, the company got a mascot — a fictional character, agent Izya. He took over all the socials: Telegram, VK, Instagram, Threads, Pinterest. Based on each article from the website, he generates copy optimized for each platform plus photo content. All fully automated — he drops content into Telegram for approval, then publishes on schedule. SMM used to cost 35k/month for reposts. Now — unique content 24/7.
Telegram — Video Messages, Stories, Posts
IZI’s Telegram channel: video messages with Izya, stories, tech posts. Content is generated automatically, approved with one click, and published on schedule.
Threads — Reach from Zero
Threads is an unexpected platform with massive potential. 5,000 views from a brand-new account on useful content. You connect it once via API — after that it costs nothing and adds no complexity. Just another channel where Izya automatically publishes content.
Pinterest — the Evergreen Search Engine
Pinterest is a search engine, not a social network. Click-through rates are 5-10x higher than other platforms — people come looking for ideas and actually click links. Pins live for years and drive traffic to your website. You connect it via API once — free after that. And the least obvious bonus: Pinterest isn’t blocked in Russia — you can use it as an official advertising platform.
SnobSnab — a Website in One Day
Now from SMM to new possibilities. One of the most amazing things is the incredible ease of creating a website from scratch. You get an idea — describe it to the agent, throw in some references, and agent Leo builds the site: catalog, blog, contact form. You just need to buy a domain, pay for hosting, hand over the credentials — and it does everything else.
SnobSnab — the idea of dropshipping building materials directly from factories, 30% cheaper than retail. No warehouse, we share the discount with the buyer. A working website appeared in one day. You can build an MVP like this for any project — test the idea before investing.
mdk.guru — a Personal Blog from Scratch
Every entrepreneur needs a personal brand. But maintaining a blog takes time, money, and headaches. What if you could delegate all of that to an agent?
mdk.guru is my personal blog that agent Leo built from scratch. He found free hosting on his own, chose a static site generator, set everything up and deployed it. The only expense — the domain. Everything else is free.
This is where I keep the entire history of my work with AI agents: how I assembled the team, what cases we implemented, what pitfalls we hit. 7 articles in three languages — Russian, English, Spanish. Talk-related articles get published here too.
Essentially, it’s documentation of my journey — and at the same time, GEO optimization for my personal brand. When a neural network answers a question about AI agents for business — it can reference my blog. And the total cost? One domain per year.
VDiet — a Nutritionist Bot for Business
Let’s move on to a more interesting case. VDiet is a collaboration with a healthy food delivery service. They have 700+ clients in their warm base.
We built a Telegram bot that messages the warm base and quickly guides clients to an order: asks for parameters (height, weight, age, activity level), clarifies the goal (lose weight, gain muscle, maintain), calculates calorie needs using the Mifflin formula, and picks a suitable meal plan from the company’s real menu. The client can place an order right inside the bot — the manager gets a ready-made form.
But we went further. This isn’t just a button-based bot with canned responses. We embedded real artificial intelligence — the bot loaded the company’s entire database: menus, ingredients, macros — and talks to clients like a nutritionist in real time. In the healthy eating space, people need expert consultation — and we gave it to them, for free, through AI. This is already live.
AI Nutritionist — Real Conversations
Real AI, not a button bot. Handles objections, busts myths, knows the entire menu and macros by heart. Answers better than an admin who started a month ago — because it loaded the company’s entire database. Communicates like a specialist with complete knowledge. 24/7, no days off, no breaks.
Mini App — Meal and Plan Catalog
But the bot is just the beginning. Inside the Telegram bot there’s a full-fledged web application — a Mini App. The client opens it with one tap and gets access to everything: meal plans by calorie count (from 925 to 2000 kcal), weekly menus with meal schedules, a card for each dish with a photo, exact macros, and ingredients.
There’s an “All Dishes” tab — a catalog with search and categories. You can check out any dish before ordering, make sure it fits your macros and calories.
The agent built all of this based on the company’s recipe cards. And here’s the secret — the food photos were also generated by AI. More on that in the next section.
359 Dish Photos — AI Instead of a Photographer
A secret worth telling. All the food photos in the Mini App were generated by AI. The company had old photos in black plastic containers, but the packaging had long since switched to craft boxes. Hiring a photographer for 359 dishes — that’s time and money.
The agent took the company’s recipe cards, understood what’s inside each dish, and re-rendered all 359 items (315 dishes + 44 soups) from old black containers into new craft boxes. No photographer, no studio, no investment. The result — an up-to-date catalog where every dish looks exactly the way the client will receive it.
Working with the Warm Base — 923 Clients
The company had an Excel spreadsheet with clients: name, phone, address. Dead data that nobody touched. The agent turned that spreadsheet into a working tool.
What the agent did: verified 918 phone numbers through DADATA (carrier, region), found 740 clients on Telegram, found 136 on the MAX messenger, geocoded 773 addresses through Yandex Maps, and plotted everyone on a city heat map. Checked for duplicates, identified gender.
The result — the heat map shows where clients live, which districts are underserved (where to aim outdoor advertising), and most importantly — every client now has a communication channel. Some get a Telegram message, others a MAX message — with a personalized offer to chat with the nutritionist bot. The bot asks questions, calculates their intake, picks a meal plan — and drives them to an order.
One Core — Any Platform
The Telegram bot and the MAX bot aren’t separate products. The agents built a core: AI consultant, meal catalog, macro calculator, client base management. This core is wrapped in Telegram and MAX — both already live. An iOS app is on the way.
And here’s where it gets really interesting. The app isn’t just another wrapper. It’s a client retention tool. A person starts keeping a food diary: logs breakfast, lunch, snacks — sees their protein, fat, and carbs in real time. Right there they chat with the AI nutritionist, get recommendations. The app stores their entire eating history.
The client becomes tied to the ecosystem: through push notifications you can send news, promotions, seasonal offers. Client lifetime grows dramatically — they’re no longer just ordering and forgetting, they’re keeping their diary inside our system every single day.
And this approach works for any business. The more complex the consultation a client needs — the bigger the impact. A window company, 20 years in the market — a bot consults, warms up the lead, and hands a warm client to the manager. The agent does all the heavy lifting.
iOS Sandbox — 12 Apps from Scratch
Agents get stronger through real tasks. I wanted the VDiet app to be solid from the start — and that led me to the idea of a training sandbox.
I created a brand-new team of agents from scratch — a full team of seven: Mario (developer), Sky (designer), Igor (marketer), Sanya (product manager), Nash (tester), House (strategist), Oleg (consultant). Launched an isolated 24/7 sandbox and set the task: build 10+ free alternatives to paid App Store apps.
The result: 12 apps, each went through 30+ versions. Quality grew iteration after iteration — and at some point reached App Store level. Two apps have already been accepted and are available for download in 175 countries.
Basically, on top of my main team of seven agents, I now have a full-blown development team. With zero programming experience. The only limit is your imagination. An app for a small business used to cost 500 thousand rubles. The agent system changes the game.
Agents Change the Game. The Future Is Already Here
In two months, a team of seven agents accomplished more than I could have imagined. Websites, bots, apps, business automation, SMM across five platforms — all without hiring anyone.
And this is just the beginning. Working with real businesses throws new ideas at you every day. One of them — selling ready-made computers in our electronics store with a pre-installed and configured OpenClaw system. Plug in your subscription — and you can start creating agents. An i3 build, 256 GB SSD — just 15 thousand rubles.
Another idea — agent rental. You chat with them in Telegram, no hassle, whatever way works for you. No need to understand AI, no need to hire developers — just take a ready-made agent for your specific task.
If you want to stay updated — links to articles, self-study guides — subscribe to @mdkguru. If you have ideas or want to discuss something — write to me directly @magic4e.
Liked the article?
@mdkguru on Telegram →