Architecture as Code: How LLMs Speed Up Systems Design

Architecture documentation in large companies is almost always a pain. Weeks of approvals, endless diagram edits in Draw.io, and versioning through Confluence that feels more like archaeological digging than an engineering process. The AI department of the BCS group of companies decided it was time to end this, and showed how an Architecture as Code approach combined with LLM assistants can radically change the rules of the game.

Aleksey Pronsky, who is responsible for architecture in BCS's AI division, described a problem familiar to every corporate architect. His team builds agentic systems, AI assistants, OCR solutions, speech analytics and classic ML models. Behind each such project stands an architecture decision: a document that has to pass a multi-stage approval with the business, corporate architecture, the information security service and the owners of adjacent systems. On average, the path from receiving business requirements to final sign-off takes two to three weeks. For an industry where iteration speed determines competitiveness, that is far too long.

The core of the proposed approach is moving architecture documentation out of visual editors and wiki systems and into code. The BCS team chose Structurizr, a tool that describes the architecture of software systems in a dedicated DSL. Instead of dragging blocks around the Draw.io canvas by hand, the architect describes components, relationships and contexts in text. This brings all the benefits developers have long enjoyed from Infrastructure as Code: versioning through Git, code review, automatic diagram generation and, critically, the ability to plug in an LLM assistant.

This is where it gets interesting. Pronsky shows how Claude Code, acting as the LLM assistant, can take over a significant share of an architect's routine work. When the architecture is described in code rather than pictures, a language model can analyze the existing structure, propose changes, generate new components and even help prepare documentation for approval. In essence, the same thing is happening that happened to coding over the past year: LLM assistants do not replace the specialist, but they speed up the work many times over by taking the routine off their hands.

It is important to understand the context in which this practice emerged. Architecture as Code is not a new idea. Tools such as Structurizr, PlantUML and Mermaid have existed for a long time. But before powerful language models, text-based architecture description remained a niche approach: the entry barrier was high, and the gain over visual editors was not obvious. LLMs changed that equation. A model that works fluently with text and code turns Architecture as Code from an elegant but labor-intensive practice into a genuinely effective workflow. The architect states the requirements, the assistant generates a draft in the DSL, the human reviews and corrects it, and the cycle shrinks from weeks to days.

For the enterprise environment, this approach brings additional advantages. When the architecture lives in a Git repository, every change is transparent and traceable. Code review of architecture decisions becomes as natural a process as reviewing software code. The security team can automate part of its checks. And most importantly, dependence on any particular visualization tool drops. Diagrams are generated automatically from code and can be rendered in any compatible renderer.

The BCS experience is telling also because this is not a startup experimenting with new approaches, but a large financial group with strict requirements for documentation and approval. If Architecture as Code with LLM support works in such a regulated context, the approach is mature enough for broad corporate use. Most likely, over the next year we will see a wave of similar adoptions, especially at companies that already use LLM assistants for development and want to extend the practice to adjacent engineering disciplines.