Ruslan Ianberdin and the cloud it
runs on. Alone.
The fact that you can just write a prompt and get a working full-stack application - that’s just the tip of the iceberg.
Build me a full-stack site.
Everything it needs:
Playcode is like Lovable or Replit, only with its own cloud: the app runs on a real server with its own database. At our core we have cloud infrastructure.
“Cloud infrastructure” sounds so simple when you say it like that.
You see the tip of the iceberg, #
and I build everything underneath
so that what’s on top looks simple
to people.
Build me a full-stack site.
1 OF 12
A very strong harness, #
fully custom-built from scratch
A harness for working with AI, a harness for the agent. Four years of experience went into it, without LangChain or SDKs. That’s how it is.
I also built our own AI gateway. There are many AI providers, each with their own SDKs, but we don’t need any of that - I built our own AI gateway completely from scratch, with pricing variations, versions, and so on.
2 OF 12
We provide full persistent computers #
These are not sandboxes at all. So if you need Redis - boom, a machine with Redis. Need Postgres - a machine with Postgres. Need two servers - two servers.
Almost everything in our infrastructure is written in Rust.
3 OF 12
I wrote our own file system #
Every action in it is a transaction
It allows rolling transactions forward and backward within a second. This lets us preserve the entire server state, including databases and source code, and move between states, while barely using any storage at all.
3 OF 12 - FILE SYSTEM
Fall asleep in one data #
center and wake up in
another data center
a second later
Moving between data centers, moving between machines.
3 OF 12 - FILE SYSTEM
A machine can scale up in size within a second. #
Scaling, downscaling, moving between data centers, rolling forward and backward - our file system makes all of this possible.
4 OF 12
The agent writes tests #
It runs the browser itself, runs end-to-end tests, writes backend end-to-end tests, unit tests.
5 OF 12
A single “publish” button updates the production machine #
A powerful development machine, and a production machine with its own database, its own configurations. Or the agent does it itself: it takes it, updates it, checks the logs, verifies everything works.
I wrote our own proxy server in Rust, which can obtain and create SSL certificates itself, and routes traffic between our full-stack applications, between static sites, and between virtual machines.
We also have static hosting, also written in Rust. For static sites, I also built a browser-based runtime where essentially Node.js with npm packages can run and do a full build directly in the browser, without any servers involved.
6 OF 12
I wrote our own DNS servers, for a few reasons #
- First, so domain resolution works very fast.
- Second, so the agent has full control over DNS records.
- And third, we sell our own domains.
- You tell the agent to buy a domain
- it buys it
- configures the DNS itself
7 OF 12
We have our own email #
sending capability
If a user’s domain was bought on Playcode, they don’t even need to configure anything - emails just work on their custom domain. 8 OF 12
We have one card, one credit #
system, whether the agent
pays or the user pays, it’s all
in one place.
9 OF 12
We also allow creating a #
private network, a private VPC Basically, servers have networks, and they can communicate across different data centers and regions, while still communicating using the same address. They don’t even notice they’re in a different region.
Cool, right? Cool.
10 OF 12
Native mobile apps, both iOS and Android #
We also have the ability to create mobile applications out of the box, within the same project.
11 OF 12
We have a fully built #
multiplayer system
Many users can work on the same project at the same time. Everything in the system syncs in real time, offline-first, very similar to how Linear.app works. Almost everything works with zero latency in the interface. Even the chat is shared. I literally studied this for half a year before AI and built it.
We have a very flexible permissions system too - I put a lot of effort into that.
12 OF 12
Everything we build #
is agent-first
When we create one operation on the backend - say, sending an email - we create just a single operation, and it’s immediately available in the API for the UI so the client can use it, immediately available for the agent to use, immediately available via CLI, immediately available via the user’s API so their own application can use it, and immediately available via MCP so even ChatGPT can do the same thing.
I also built a knowledge-tracking system for everything the agent does. It’s like a climber: to climb up, you have to anchor the rope. To climb even higher, you have to anchor the rope. This lets the agent make changes to a project almost entirely on its own.
Replit does this with hundreds #
of people - I did it alone.
The cloud platform is the hardest thing I’ve ever done. For several years I was figuring it out. I built it, built it, built it, but didn’t launch it in production. I wanted to quit hundreds of times. I started without AI at all, and built most of it when AI was still very weak, so I figured everything out myself. The scope kept growing, the complexity kept growing.
It launched on May 24, 2026, and customers’ apps run on it now.
Andrei, an engineer I’d worked with before, joined full-time on October 1, 2026.
I got into professional programming on my own. After Python courses, I found my first paid work within a few months, while still studying mining electromechanics.
For my diploma, I proposed a distance-learning system for the whole college instead of the topic I was given. The teacher disliked me, nothing good came of it. And then I realized something: in college the one in charge isn’t the teacher, it’s the director. I went to the director, and it turned out that it’s easier to talk to the person who makes the decisions. And then I understood life. I explained the whole idea to the director, and, strangely enough, he agreed. He said, “Cool, let’s do it.” Then he called the teacher who didn’t like me, and everything was settled at once.
That’s how I understood how #
I can get what I want in this life.
Full-stack from the first day. I code even #
when a CTO. I’ve been writing software for 21 years.
- Fifth gradeWhen the other kids played football, I sat taking the computer apart and putting it back together, deleting files from the file system and waiting for it to break, and then reinstalling it. This was in fifth grade, if not earlier.
-
2013CTO at Bit-T, a crypto exchange.I hired a couple of developers and together we made an exchange and wallet for BTC and LTC. #
-
2013-2014Founder of Doffee.me, my first startup, a marketplace for home services.I made a team (designer, 2 developers, devops) and developed a marketplace together. #
-
2015-2016CTO at Trendever, a marketplace and bots for Instagram.I became CTO, hired a new team, and everything got moving fast. The idea was good: buying through comments under a photo. Later Instagram built it, and the business went under. #
-
2016Founder of Playcode.I started Playcode in 2016 as a JavaScript playground and built it alone as a side project while I worked as a lead engineer for startups. It grew to 1.1M registered users. We were number one on Google for “JavaScript playground” and had 19,000 visitors a day at the peak. People have created 2.6M+ projects on it. #
-
2017-2022Lead engineer of the Pairity marketplace, where banks sell debt portfolios.I was hired by a US outsourcing firm. They had 4 months left to make a huge marketplace for 200 screens. Only the wireframes were ready. They were confused that I was a Golang/Python developer, and had no Node.js experience. I figured out all the necessary technologies in 2 weeks. I finished the project in exactly 4 months. Then I stayed on the support and refinement of this project for a few years. #
- 2022My first commits with OpenAI models went in on December 18, 2022, 18 days after ChatGPT came out.
-
2023-2024Head of Development at Simplenight, a travel booking startup in Miami.I came in as an advisor. Again, I went straight to the founder. I let go of 60 IT staff and built a team of 7 people. It was a tough and stressful battle of one person against a big system. #
-
December 2025Playcode, full-time.I’ve worked on Playcode full-time since December 2025. We’re not just building websites anymore, right? We’re AI native. #
In my first startup, I expected people to make orders on the website, but people wanted a handyman and didn’t want to use the interface. In the end they phoned, and the whole business turned into finding workers by hand. I didn’t like that, so I closed it and sold the platform and its source code.
The interface of the future, and the computer #
of the future, is the AI worker, something that
takes a task and solves it from start to finish
with minimal or no human involvement.
When was the last time I used Google? I don’t even remember. I used to use it 30 to 100 times a day. Now I use it zero times a day. Why? Because there’s an AI worker, and its name is ChatGPT.
The entrepreneur understands a person’s need and problem, builds an AI worker for it, and sells it as a service, making money. In Lovable it’s the other way around: you build software, you build a website, and you can add AI to it as a feature. We turn that around. We’re building the platform where entrepreneurs create AI workers, and a marketplace where people hire them.
Every new app on our cloud already has an AI agent, right in the admin panel. The owner can say, “How many people bought today?” Or, for example, “Send these three customers an email saying the order is delayed, and mark them as awaiting.” That’s it. You don’t even need special functionality, some buttons. Buttons have already become secondary, a legacy interface. The owner approves each change.
- One of our clients is building an AI worker for selecting properties - for buying or selling real estate.
- Elderly people often have a problem: they tend to be very lonely. They’re lonely, bored, and have no one to talk to. So another client is building a voice companion so they can talk and not feel so bored.
- A third example: we have a social worker who realized that people don’t understand what benefits they could be getting but aren’t. The government sends them letters, and they don’t understand how to respond properly. So he’s building a voice assistant that helps them communicate with the government. And he’s not building 300 screens for how it should all look.
People don’t want one more app. #
They want the job done.
No customer has ever come from an ad. #
We’ve put a lot of effort into organic traffic, into SEO optimization - we’ve been investing in that for years. It’s actually part of the product. We also know our weak point: one channel, and it’s not ours. Now I’m building channels that depend on us. I want to learn to get a customer for money.
I’m open to consulting. #
I could help with architecture, help audit a team, check if there are any bad developers or scammers on it. I could jump in part-time as a consultant and auditor.
Write to me. #
Playcode: Playcode.ai. X: @ianberdin