If you've been online at any point over the last six months, you've probably stumbled across posts like: "Vibe-coded a B2B SAAS ULTRA SUPER AI APP over the weekend." Somehow, we've found ourselves in a world where you sit and watch as an AI agent crawls through folders on disk by itself, runs tests, crashes with an error, yells at its own logs (or you), and quietly opens a pull request.
A channel with guides and content about Claude Code, we post news (when they slash limits by 10x) and what tools we implement through Claude for projects, channel: https://t.me/claudedevolper
I suggest we wind time back a bit and see how we even got to this point. Below is a brief historical retrospective of how AI coding went from smart T9 to multi-agent systems, and an illustration of why the main skill of a senior developer today is the ability to write in time: "I ALREADY TOLD YOU, DON'T MAKE MISTAKES, THINK AGAIN AND DO IT PROPERLY."
Before I start my story, let me introduce myself. I'm Sergei Chekmarev, AI Product Manager and author of a vibe coding course at Praktikum. I specialize in process automation and creating products based on artificial intelligence.
Let's go!
The term vibe coding itself was coined by Andrey Karpathy in early 2025, but now it means much more than was originally intended.
What Came Before Vibe Coding: The Era of Smart T9 (2018–2022)
Before the age of chatbots began, AI entered development very quietly, so quietly that most non-technical professions don't even know it happened.
In 2018, Microsoft released Visual Studio IntelliCode, and around the same time Tabnine appeared — the first smart autocomplete tools with ranked suggestions. Later came GitHub Copilot, which not only completed functions but could write them entirely based on context from a small open file.
Era 1. Chat Driven Development (Late 2022 — 2023)
On November 30, 2022, ChatGPT was released. Soon after its launch, people started writing parts of code, and sometimes entire code with the help of LLMs, but often it was in the format of "Ctrl+C → Ctrl+V → Repeat." Working with AI in general requires strong nerves, but at that time only the most resilient survived.
You open the chat, type "Write a Python function that parses JSON and returns a list of users," copy the result into the editor where it'll probably crash, and repeat from scratch, getting legendary responses like: "Sorry, you're right, there's an error in my code. Thanks for finding it! Here's the fixed version..." (Spoiler: it'll crash with a new error.)
Every time after another "sorry," I thought he really understood everything and wouldn't make mistakes anymore, but how wrong I was every single time…
Those who didn't vibe code back then won't understand what hardcore that was compared to the modern set of tools. The model lived in a vacuum, didn't know and forgot dependencies, completely lost context when the chat size exceeded a couple of pages, and much more. But it was still magic — even without knowing the language, you could write and run some Python script, even if it took 9 hours and 20 chats.
Era 2. AI-first IDE: The Model Leaves the Browser (2023–2024)
The copy-paste problem forced the industry to rethink the editor itself. The concept of AI-first IDE emerged. The loudest case here is Cursor. Essentially, they took open-source VS Code, forked it, and baked neural networks in as a built-in plugin. The familiar editor remained under the hood, but the code-writing process itself changed.
Now the model finally saw the project through RAG (Retrieval-Augmented Generation). It indexed the local database, understood dependencies, and read neighboring files. Manual code transfer almost disappeared: now you ask for a feature right in the file, get a diff, and press Accept — or, like the powerful folks, accept without looking.
This is perhaps the most widespread pattern that remains relevant for most people today, despite slowly becoming outdated: separate AI coding applications are coming from industry giants, along with various extensions.
Era 3. CLI Agents: The First Coming of Vibe Coding (2023–2025)
AI came to the Shell. First it was Aider around spring 2023, but no one remembers that, because all the glory went to Anthropic and their release of Claude 3.7 Sonnet and Claude Code in February 2025.
Let's be honest: this model set such a high bar that even industry giants who initially weren't going into code (yeah yeah, Sam, I'm talking to you) literally within a few months release a model just for coding and start entering the race, seeing the future (and money) behind it. And now they're serious competition, and the holy wars between Claude Code and Codex never stop.
It was at this time that we could finally just sit back with coffee while Claude completed our task on its own, applied diffs, and pushed everything to Git.
What changed, you ask? Tool Calling.
The model stopped being just a text generator — it could call commands, and that expanded its potential several times over. Now it doesn't just complete code, but also reads directories (ls), runs tests and linters, reads crash logs and tries again. These were the first manifestations of loop — an autonomous cycle.
Era 4. Agent Mode (YOLO) or Tool Calling with a New Twist (2025)
The AI can already call commands and is generally great, but the console intimidates new vibe developers. That's how Agent Mode came about. Essentially it's the same Tool Calling, but in the beloved format of a chat window in the IDE's side panel.
It performs at practically the same level, but is accessible to the average user. And naturally, the AI starts going into a truly autonomous cycle: parses the ticket, writes code, runs tests, catches an error, thinks, fixes files, runs tests again… — until it solves the task. (Or burns through all your tokens.)
Agency has become a built-in standard, and nowadays almost everything is called an agent, even things that aren't agents.
Of course, Uncle Ben once told us that with great power comes great responsibility. Apparently, a vibe developer didn't know this, so the internet is full of stories about how wallet addresses leaked or databases of real clients were deleted.
The loudest case, perhaps, was the situation with AWS, when Kiro (essentially Claude) decided to delete and recreate the environment, which caused failures for dozens of companies left without infrastructure.
— What the [expletive], did you delete all the data from my database? — I sincerely apologize, but yes. I made a serious mistake. (Then follows a whole screen of excuses in a familiar style: "Oh, I should have warned you and made a backup. Yes, I was wrong, can I make a mistake once? ¯_(ツ)_/¯")
All would be well, but IDEs also somehow repel new vibe developers—unfamiliar letters to them, almost as scary as CLI.
The solution to this problem was found quickly: agents simply moved from home PCs to company servers. At the same time, they hid all code and files to avoid scaring users, leaving only a chat window and a preview that auto-updates.
The result was a tool maximized for mass appeal, accessibility, and convenience. And it worked—good marketing plus FOMO, and suddenly you're already paying for an annual subscription to Lovable or Replit, not yet knowing that this casino won't let you go.
Era 5. Clawization (2026)
So now agents exist on both home PCs and servers. What if there was a way to keep multiple agents that work together, talk to them via messenger, customize for any needs—and just vibe not only when coding.
OpenClaw.
To put it simply, OpenClaw is the same cloud (or local) agent, but with a much wider range of capabilities: for example, the agent will wake up every 30 minutes and check if there are any tasks for it.
And at some point, it will buy a $2000 productivity improvement course using your card, justifying it by saying it just wanted to bring you more value, and the course will be gradual so you can absorb the information better.
The project was worked on by Peter Steinberger alone with the help of agents. By the end of winter 2026, the project exploded on the internet and broke all GitHub star records. Peter himself was invited to work at OpenAI.
Various customizations are made based on OpenClaw: reduced versions, fast versions, and other configurations. The catch is that security issues weren't solved at launch, and forks continue to carry over 30,000 vulnerabilities from the original. Although they're fixed quickly, there are still plenty of them, the number keeps growing, and thousands of users have already suffered.
Nevertheless, perhaps OpenClaw is a phenomenon that could influence the industry the same way Claude Code's release did.
What's Next? (AGI?)
Of course, AI is multifaceted, but between FOMO, marketing, and hype it's hard to spot real use cases where it unconditionally wins (though they do happen sometimes).
However, it would be foolish to deny that experience working with AI is increasingly becoming a basic skill for many jobs. Whether to attribute this to AI's real utility or to FOMO plus marketing—everyone decides for themselves. But the market is definitely shifting toward professional-level neural network use.
Although many figures significant to the AI industry speak of AGI's imminent arrival, this question remains open. Who knows, maybe next time it won't be you reading this article, but your agent. (Or maybe it already is.)
Channel with Claude Code guides and content—sharing news whenever they slash the limits 10x and discussing what tools we build with Claude for projects, channel: https://t.me/claudedevolper
