How AI Agents Are Changing the Way Businesses Work with 1C Data

Roman Korotaev Head of Enterprise Automation at Cinimex In 2025, the AI market saw two notable developments. First, companies began actively discussing how AI models could connect to data and functions of enterprise systems through the so-called context extension layer, or Model Context Protocol (MCP). Then, agent-based scenarios came into focus, where AI does not simply answer a question but performs a sequence of steps, uses tools available in the real-world environment, and builds results based on actual system data. In 2026, these approaches are making their way into the world of 1C systems. This is particularly important because, in large companies, 1C often forms the foundation of the entire information infrastructure.

How Can AI Access Data?

Until recently, discussions about AI in 1C often revolved around a simplified idea: connecting a chat interface to the system so that users could attach some information from the system and ask questions in natural language. This may look convincing in a demo, but in practice, a chat interface alone does not solve much. If a model does not have reliable and controlled access to the system's data and functions, it will compensate for the lack of information with assumptions. For management analytics, commercial information, financial data, and personal data, such a scenario is unacceptable.

That is why discussions about AI in 1C quickly move beyond the interface itself and focus on more practical questions: how to provide access to actual system data, how to ensure security, and how to make the results genuinely useful for the business.

This approach has been formalized as a standardized layer for AI access to system data and functions. It later evolved into an industry standard known as MCP (Model Context Protocol). The key change is that the AI model no longer operates blindly. Instead, it can access predefined tools to retrieve data, invoke business functions, extract relevant entities from multiple systems, and return verifiable results.

AI Agents

The next stage in the evolution of chat-based interaction with enterprise systems is an AI agent combined with an access layer for tools and data, such as MCP.

Unlike a conventional AI model, an agent performs a sequence of actions: it understands the task, selects the necessary tools, collects intermediate results, compares them, and only then provides the final answer to the user. This is particularly important for businesses because real-world requests rarely consist of a single question in a chat.

Consider the following request: “Show me the top 10 customers by shipment volume for the previous quarter across all 1C databases.”

To answer it, the system needs to identify the relevant period, collect data from multiple databases, and aggregate the results. From the user's perspective, this looks like a simple chat interaction. Behind the scenes, however, the agent is performing a sequence of multiple steps.

Why is an AI agent combined with MCP business functions and an AI model deployed within the company's local environment particularly relevant for 1C?

First, 1C contains large volumes of data, often distributed across multiple databases.

Second, the cost of an inaccurate answer can be high. An error in analytics based on accounting data can have significant consequences.

Third, security is critical in enterprise accounting systems. This includes not only commercial confidentiality, but also personal data, financial information, role-based access control, and internal auditing. As a result, enterprise companies are increasingly interested in local AI models deployed within their own infrastructure and architectures in which sensitive data does not leave the organization's perimeter.

However, a local AI model alone does not solve all these challenges. In practice, the quality of the solution depends not only on the model itself but also on how well the tools it uses are designed. Simply giving an agent access to the technical structure of a database and expecting it to figure everything out on its own is unlikely to produce reliable results. A much more robust approach is to design tools as business functions with clearly defined purposes, limitations, and verifiable outputs.

This is precisely the approach that currently appears most promising for 1C: a set of business tools through which the system can obtain the required analytics.

For example, such tools can include functions for identifying top customers by shipment volume, key suppliers by incoming deliveries, analyzing product categories, selecting goods within specific categories, or aggregating data across multiple databases. The more precisely these tools are described, the smaller the room for errors and the higher the explainability of the results.

In other words, the key architecture consists of several interconnected elements: the AI model, the data access layer, a set of business tools, security rules, and agent logic.

AI Platform for 1C

We use this architecture in our own projects. In practice, it can be implemented as follows.

1C Bus can serve as the integration layer. On top of it, a data and function access server can be implemented using 1C.Element technology as an MCP server. This architecture makes it possible to connect 1C databases seamlessly, add new analytical tools, and manage them within a unified architecture.

At the same time, a local AI model can be deployed within the company's infrastructure, ensuring that sensitive data remains secure. The user interface — the AI agent — can either be provided by a third-party solution or implemented as a chat interface embedded directly into the 1C database.

Such a solution can, for example, be specialized in management analytics across multiple 1C databases through a single AI interface. The user asks a question in natural language, after which the system collects data from different sources and returns the result.

It is important to emphasize that, at the current stage, we see the most realistic business scenario as a read-only approach. In this model, AI helps users quickly collect and interpret analytics based on real data but does not post documents, modify data, or initiate high-risk operations.

This approach reduces risks, simplifies discussions with the IT department, and allows the business to gain practical value from the technology faster. It also makes the discussion around implementation much more concrete.

Accuracy deserves particular attention. “Hallucinations” are often perceived as an inherent problem of LLMs. In practice, their frequency can be significantly reduced when the model no longer operates independently of the underlying data and instead uses predefined tools.

The combination of a data access layer, business functions, and agent logic fundamentally improves the reliability of the results.

This is especially important in the 1C environment because trust in the answer is more important than any impressive technological effect. Businesses do not simply need an “intelligent chat partner.” They need a tool that helps them answer specific management questions faster, without compromising security requirements or forcing employees to manually collect information from multiple databases.

Conclusion

Today, we can already see a broader market shift. AI-agent technologies, implemented both within the 1C ecosystem and through external software, are gradually entering even such a closed environment as 1C.

This is not yet a widespread practice. However, business interest is clearly growing: companies want to integrate AI into their corporate environments in a way that makes it accurate, secure, and genuinely useful.

If 2025 was the year when the market learned how to connect AI models to data and system functions, then 2026 is increasingly becoming the year when businesses begin to view agent-based scenarios as the next logical step — including within 1C.