MWS AI Meta-agent: AI Started Assembling AI Agents for Users

Danila Katalshov, technical lead of the MWS AI prompt engineering team, created a meta-agent — an intelligent system where AI acts as an architect, designing and assembling other AI agents based on a user's text description. The project won a competition and was soon integrated into the MWS AI Agents Platform.

Low barrier to entry, but still high

MWS AI Agents Platform is a low-code solution for building AI agents and multi-agent systems. Instead of writing code in Python or JavaScript, the user simply drags and drops blocks in a visual builder, selects language models, configures connections between components, defines conditions and triggers. No programming required.

But there's a catch. Even a visual builder requires engineering thinking — architectural thinking. You need to understand the development cycles of AI solutions, data types, how components interact with each other, the logic of information processing flows.

The platform eliminates coding, but not the need to design the architecture of an entire system. This blocks access for non-programmers and slows down work even for experienced developers. Danila solved this problem directly: if the platform is already capable of assembling agents, why not entrust the design to AI itself?

An AI architect building AI agents

Within MWS AI's internal competition, a meta-agent emerged — a system where AI analyzes a user's text description and automatically designs the structure of an entire system of interacting AI agents. Imagine: instead of manually building a bot to track marketplace prices, send notifications, and analyze trends, the user simply writes a few sentences: "I need a bot that monitors prices and sends notifications if the price drops by more than 10 percent." The meta-agent parses this description and instantly designs the architecture: how many separate agents are needed, what roles they should perform, how they will exchange information. Typical process:

• User describes a business task in a few sentences in Russian or English
• Meta-agent analyzes the description and automatically designs the architecture, selecting the optimal number of agents and their specialization
• The system creates each agent with the necessary parameters, selects appropriate models and tools
• Automatically establishes connections between agents, determines data flows and activation conditions
• The ready multi-agent system is immediately deployed and ready to work

The project won the competition. This was not just praise for the code, but a signal to the company: this idea works and has enormous potential.

From competition to release

The MWS AI team didn't stop at the victory. They took the idea, refined it, and integrated the functionality directly into the platform. They deployed the result quite recently: now all users of the MWS AI Agents Platform have access to an embedded meta-agent, available in just a few clicks. Even a person with no experience in developing AI systems can assemble a working multi-agent system that solves real tasks in a few minutes. Development speed has increased tenfold. The barrier between idea and implementation has been erased: description → ready system. No intermediate steps, no manual design.

What does this mean

The boundary between no-code and full automation is disappearing. Now AI automation is accessible to everyone — not just engineers, but managers, analysts, entrepreneurs. If previously creating a multi-agent system required weeks of work, now it takes minutes. And this is just the beginning: meta-agents will be improved, new types of automation will emerge that we cannot imagine today.