Vibecode in Bitrix24: Building a Customer Profile with AI Agents — A Step-by-Step Experience

The Bitrix24 team published a detailed breakdown of working with the Vibekod platform: how, using AI agents, they assembled a "customer passport" — a consolidated company card with contacts, deals, leads, and all activities in a single screen of the business portal.

What is Vibekod and why is it needed

Vibekod is a development platform within the Bitrix24 ecosystem, built on the principle of vibecoding: you describe a task in natural language, an AI agent generates the interface and logic, you check the result and refine it through iterations. No manual code writing, no need to dive into API from scratch.

The platform's main goal is to give business analysts and product managers the ability to independently customize the portal without involving a developer. This is exactly what the authors decided to test on a specific practical case.

Customer passport: what was collected and how

In a typical CRM, company data is scattered across different sections: contacts, deals, call history, and emails live on different screens. A manager is forced to piece together the picture manually before each call or meeting.

The "customer passport" solves this problem: one page where you can see everything at once. The authors collected in the card:

• company contacts with positions and communication channels
• open and closed deals with amounts and statuses
• leads with sources and responsible managers
• activity chronology — calls, tasks, meetings, emails

The implementation was built through iterations: they described each block to the agent, looked at the result, and refined it. Some tasks — for example, basic contact list output — the platform handled without questions. Connecting all blocks with live data from the CRM turned out to be more complex.

Where the AI made mistakes and how they solved it

There were no shortage of errors. The agent periodically formed requests to the Bitrix24 API with incorrect parameters: fields had different names than in the actual structure, or the selection logic worked differently than expected. In such cases, the authors clarified the task conditions in the dialog and restarted the iteration.

Some problems were solved by reformulating the request to the agent. Others required direct manual fixes — when the agent reproduced the same error or when it was necessary to precisely specify the request structure independently.

"We're telling about our experience step by step — what was simple,

how we built the work, where the AI made mistakes and whether we had to fix something manually," the authors write.

This is a standard vibecoding pattern: the AI takes on most of the routine work, but precise integrations with real data and final debugging remain with the human for now.

Why this matters for business

Vibekod is a telling example of how corporate platforms are embedding AI development directly into their ecosystem. Previously, any Bitrix24 customization required a developer with API knowledge. Now a business analyst can describe a task in Russian and get a working prototype in a few hours.

The main current limitation is the same as with most AI development tools: the more complex the integration with real data and the more specific the business logic, the more often the agent misses the mark and the more manual refinement is required. But even with this caveat, the barrier to entry for customization drops significantly.

What this means

The customer passport case shows both sides of AI development in corporate systems: real acceleration on typical tasks and persisting dependence on manual control where data is complex. The more such honest breakdowns appear — the more accurately business understands where vibecoding is justified and where it's better not to cut costs on developers.