At AWS Summit Warsaw 2026, retailer COMFY presented its internal AI Workflow system. It is an orchestrator that manages a group of AI agents and fully automates the software development cycle — from business analysis to deployment. Thanks to it, the company rewrote the old legacy admin that had been in operation for over 10 years.
At AWS Summit Warsaw 2026, retailer COMFY presented its internal AI Workflow system. It is an orchestrator that manages a group of AI agents and fully automates the software development cycle — from business analysis to deployment. Thanks to it, the company rewrote the old legacy admin that had been in operation for over 10 years.
The main insight: now a business user can make changes to the site interface without involving developers. Mykhailo Berelet, Head of the e-commerce systems department at COMFY, told dev.ua about the technical essence of the project.
Conventional «vibe coding» (chaotic code generation via ChatGPT) creates unmaintainable garbage. Instead, COMFY built a system of specialized AI agents integrated into a familiar Jira-like interface:
Analysis: The analyst agent accepts a request from the manager in human language and generates a specification.
Development: Agents generate code. The system is provider-agnostic and supports various LLM and CLI systems, including Kiro CLI, Claude Code, and Codex CLI.
Quality Gates: Separate review agents test the code for security, performance, and conduct API tests.
The new admin panel is already working in production. According to the company, the average task (from idea to merged pull request) takes about an hour, and the acceleration of processes in certain areas is from 4 to 20 times. Developers have been retrained from routine performers to architects of the AI environment.
However, there is no complete autonomy: AI still hallucinates and makes mistakes if the task is vague. Therefore, blind trust is excluded — human expertise and architectural supervision remain mandatory.
