The Vibe Coder's Cookbook: One LLM as Architect, Reviewer and Tester

In August 2026, a Russian developer published a detailed guide on Habr to multi-agent development with a single language model, with no extra infrastructure to buy and no months of setup.

What Is "LLM Schizophrenia"?

The whole method rests on one basic effect: the same model that has just written a feature and is satisfied with it will, in the reviewer role, find three real bugs in it, complete with failure scenarios. One set of weights produces two diametrically opposed opinions about the same code.

Key facts about the method:

• The troupe consists of four roles: architect, developer, reviewer, tester
• Switching between roles happens through role prompts ("skills")
• The author runs a real production project on this principle, written by agents with almost no manual code
• The first step requires not a single line of hand-written code: the troupe builds itself

The split is not a bug but an architectural technique: a reviewer that doesn't know it is looking at its own code criticizes more honestly than any human colleague.

Why the Process Must Live in a File

Architectural decisions the model made in the architect role vanish from context by the time the same model switches to the developer role. Without a written plan, the agent reinvents the wheel every time and wastes tokens repeating work that is already done.

The author insists that any decision that must outlive a single session has to exist in a file, not in the dialogue and not in the model's memory. Skills, the sets of role instructions for each of the four roles, are stored there too and loaded at every switch.

"The process must live in a file, not in your head," is the key thesis of the guide published on

Habr.

Changing context between roles takes seconds, and accumulated agreements and decisions are not lost between sessions.

How to Launch Your First Multi-Agent Project

The troupe can build itself, and that is the article's main practical takeaway. All it takes is to give the agent in the architect role a description of the task: it will draw up the project structure, distribute tasks among the roles and create files with plans.

According to the author's description on Habr, the entire cycle, from an empty repository to working code with review and tests, runs without a single line written by hand. The workflow looks like this:

• Skills: role instructions that define the agent's "personality" in each role
• Plan files: recorded decisions that survive changes of session and context
• Role cycle: architect → developer → reviewer → tester, repeating when bugs are found
• Self-assembly: the architect agent builds the project structure, not a human

"Accessible, no encyclopedia, done in one cup of coffee," is how the author describes the format of his guide on

Habr.

What This Means

Multi-agent development has moved beyond the lab experiment: one person with one language model builds an in-house team that criticizes itself more honestly than most human reviewers. The main barrier to entry is not infrastructure but the willingness to commit the process to files.