Top 5 ways to use AI in Software Architecture
Automatically Reviewing and Improving Diagrams
AI can take existing software diagrams—such as component, sequence, or deployment diagrams—and check them for clarity, best practices, and consistency. It identifies missing pieces (for instance, a missing database icon or an unclear communication path) and suggests improvements based on widely accepted guidelines like UML or C4. This capability offers near-instant feedback, helps maintain consistent styles and structures, and speeds up the review process, reducing the time spent on manual checks.
ChatGPT for example can help you drastically improve the diagrams. Consider this diagram. Is it aweful? Indeed. Everything is one-color, there are no connections between the components, the purpose of some services are copy-pasted - looks like a real-life example, right?
Well, with some prompts it can look at the diagram, spot the problems and create a new one based on the UML. It will look like this:
Now it looks much better: color coding is in the place, the connections between the components are outlined, and it's straightforward to reason about the layers and components responsibilities.
Improving Architectural Decision Records (ADRs)
Architectural Decision Records store the critical design choices you make, along with the context and alternatives considered. AI tools can improve these records by suggesting more precise language, highlighting overlooked details, and structuring each decision’s pros and cons. This ensures that anyone revisiting the ADR later can quickly understand why a particular decision was made, how it was made, and what potential trade-offs were involved, all in a clearly documented format.
Imagine your team chooses the database for your new Food Delivery project. They outline 3 options, make a call and proceed with MongoDB. A solid choice, one would think, but why exactly 3 options? And what is the business rational behind MongoDB in particular? You can feed your requirements into an LLM, also give it your ADR and receive something like this in response: "Decision: Based on initial research, we will consider three primary database options: MySQL, PostgreSQL, and MongoDB (NoSQL). After weighing the pros and cons of each, we have decided to use PostgreSQL as our primary database. While it may not be the most cutting-edge option, its robustness, scalability, and community support make it a suitable choice for this project."
Creating Request for Comments (RFC) Templates
When teams propose new features or major system changes, they often issue a formal RFC. RFC is a format of Request-for-Comments: the team creates a document which covers functional and non-functional requirements, diagrams, architecture decisions, cost analysis, calculations, reliability tactics and many more. This is a hard document to write.
AI can generate robust templates that prompt authors to include all the vital information—objectives, design overview, data flow, testing strategy, and risk assessment. By making sure every important topic is covered up front, it becomes easier for team members to evaluate proposals, contribute useful insights, and arrive at well-informed conclusions.
Pros and Cons Analysis
Selecting technology or deciding how to structure a system can involve many competing factors. AI-driven assistants can process a variety of inputs—like your system’s scale, anticipated workload, performance criteria, and budget—to produce a concise analysis of each option’s advantages and drawbacks. This helps reduce bias toward familiar tools or technologies and provides a broader, more objective perspective on possible solutions, ultimately leading to choices that better align with your project’s specific needs.
Indeed, LLMs tend to agree with the user in favor of providing precise information, however knowing the downside can play in your favor. Ask LLM to act as independent technology consultant and provide you with a comprehensive list of pos and cons of each option.
Automated Code and Architecture Prototypes
AI can do more than just documentation or textual analysis—it can also help build tangible prototypes. These tools can generate initial code scaffolding, outline module structures, draft data models, and even estimate performance needs. By giving architects and developers a functional starting point, AI frees up more time for deeper innovation and customization, rather than spending hours manually setting up new projects or guessing at optimal cloud configurations.
The tools like Bolt.new, Loveable.dev, Replit AI surely can quickly prototype the solutions for you which can be initially tested for further feedback and learnings. This PoC can be thrown away to build a more robust solution.
Final Thoughts
With AI rapidly advancing, these applications will become even more powerful, providing faster feedback, better consistency, and richer insights. Rather than replacing human architects, AI acts as a critical companion, ensuring that design processes remain efficient, well-documented, and closely aligned with both technical requirements and business goals.
System Design Course Cohort #5 is open!
Do you want to design software solutions which fit business needs and advance your career? This course will emphasize the importance of recognizing and addressing business priorities with technical approaches and show how to leverage modern AI-driven tools to bring the architecture game to a new level. Learn to go beyond load balancing options and performance tactics by focusing on solving real business challenges. Course is completed by 50+ engineers with great feedback!