Александр Шапошников - Как начать работу с ИИ в 1С

Summary

Alexander Shaposhnikov demonstrates how to use an AI coding agent with a 1C configuration, then discusses the workflow’s benefits, risks, and limitations.

Setup and workflow

• Shaposhnikov starts with a standard 1C demo database based on BSP (the standard subsystem library), exports the configuration to source files, and connects the project folder to an AI agent.
• The agent works with the files through Unika, an open-source toolkit for 1C development. Unika combines tools such as MCP servers and skills, and includes V8 Runner for tasks such as building and loading a configuration.
• The project uses Git to track and commit changes. A small demo configuration took a few minutes to export; a large ERP could take much longer, depending on its size and the computer.
• Shaposhnikov shows three levels of agent permission: automatic execution, approval before file edits, and approval before any action. He uses automatic mode on a personal, backed-up computer, but warns that it carries risks, including destructive actions.

Live 1C task

The audience proposes adding a command to create a customer invoice from a counterparty in the BSP demo. The agent:

1. Inspects the configuration and locates the relevant document, directory, and existing “create based on” mechanisms.
2. Changes four configuration objects, including the counterparty directory, the invoice document, and BSP-related modules.
3. Encounters a problem editing XML and a build issue involving locks. After retrying the build, the configuration is loaded and tested.
4. Creates an invoice from a counterparty.

The demonstrated workflow takes about 20 minutes.

A subsequent request adds an HTML field to the counterparty directory to display a website. The page loads, though the 1C HTML window does not fully handle the site’s cookie controls. Shaposhnikov presents this as another short demonstration, not a guarantee that every task will work on the first try.

Tools, models, and practical guidance

• Unika is presented as a free, open-source starting point for 1C developers that covers many development tasks. It does not handle working with data by itself; additional MCP tools may be needed.
• Unika Arch, a separate commercial tool from Unika’s creator, is described as supporting more complex task design: defining requirements and testable conditions, creating technical specifications, and producing process or BPMN-style models.
• The agent still needs suitable tools, skills, and a well-specified task. If it cannot complete something, possible remedies include improving the tool, adding a skill, or giving the model examples from the existing project.
• Shaposhnikov recommends reporting tool bugs to their GitHub projects, while checking that reports do not expose confidential data, credentials, or sensitive source code. For testing, he also mentions masking data in a test database.
• He uses Claude and refers to other agent environments and models. Model and subscription costs depend heavily on usage: multiple agents and long-running tasks can consume limits quickly. In his experience, a $20 plan was enough for occasional assistance, while heavier parallel work required more capacity.

Analysis and limitations

• Shaposhnikov gives an example in which Claude completed a customization in about 15 minutes that a contractor had estimated at 60 hours. He presents this as a specific case, not a universal productivity guarantee, and notes that his own estimate for doing it manually would have been lower than the contractor’s.
• AI can reduce routine coding and help developers understand existing systems, but task analysis, design, review, testing, and communication with the customer remain important. The agent’s output should not be accepted without verification.
• Adoption in large companies may be constrained by closed environments and restrictions on external cloud services. Shaposhnikov argues that locally available models can lag behind leading models on programming tasks, making the business case harder to establish.
• He expects developers’ work to shift toward higher-level problem-solving rather than disappear. For 1C systems specifically, he argues that AI-assisted coding need not mean ever-larger monolithic ERPs: some new, narrower functions may be delivered as separate services that integrate with core systems.

Main speakers and sources

• Alexander Shaposhnikov — primary speaker and demonstrator; 1C systems architect.
• Igor Apressov — Unika creator; briefly explains Unika and Unika Arch.
• Dima — host or presenter referenced during the talk.
• Audience members — suggest demo tasks and ask questions about safety, cost, and adoption.