AI Programming Keeps Advancing — How Can You Improve Your System Architecture Skills?

As nowadays, AI code writing is really getting better and better. Some small modules are used to write AI, which is fast and good. Gradually, the requirements for programmers' ability to write code will be reduced, but the requirements for system architecture will be higher. Moreover, this part of the capabilities will not be easily replaced by AI, because AI cannot see the full picture of a complex project like humans, and cannot fully understand the business context and constraints of the project, so people need to split the tasks and perform only one small task at a time.

So the question is, how can newcomers improve their system design capabilities?

As a veteran, I think it may be similar to whether there is AI to improve the system design capabilities. It is inseparable from watching, practicing and reviewing more. The advantage of AI is that it can help you check information more easily and understand projects more easily, but it cannot be used as a hands-on practice on your behalf.

What is system design?

Before discussing how to improve system architecture capabilities, you need to understand: What is system design?

Think about it, if you make a simple web page, or a small script, do you need a system architecture? Basically it is not needed because it is simple enough to write whether there is a framework or not. What if you want to create an application like ChatGPT? That must have a good system design because it is too complicated! Moreover, one person cannot do it, and it requires many people and even a lot of teamwork to complete it.

But even with such a complex system, everyone working on the development team has not very complicated job responsibilities. For example, some people are only responsible for the API encapsulation and deployment of the model, some are only responsible for text editing components such as Canvas, and some are only responsible for a certain function on the iOS client. A complex system is designed to separate the complex systems layer by layer and turn them into small modules. Finally, these modules can be run together, allowing users to use ChatGPT's services stably.

So the system design is Disassemble a complex system into small modules that are easy to understand, implement, and maintain, and then clearly define the process of how these modules collaborate and communicate with each other .

In layman's terms, it's like you build a building, and the system design is the design drawings. The drawings describe the structure of the building in detail, such as foundations, steel frames, walls, pipes and wires, and clearly tell the construction personnel how to cover each part, which parts are first, which covers the back, and which parts are connected.

How to do system design?

System design seems mysterious. System architects seem to be legendary, but system design is not as complicated as you think, because most system designs have mature solutions to refer to. As an architect, you can flexibly apply mature architectural models, and generally you don’t need to invent new architectural models.

But the system design is not that easy, otherwise everyone will be an architect. What is the difficulty in system design?

• • Full understand the requirements and the team : I put understanding the requirements and the team first, because the system design serves the requirements, and the design you do not understand the requirements may be wrong; at the same time, the system design must match the organizational structure. After all, your system design still needs to be implemented by the team. If you only have two or three people to complete hundreds of microservices, it is obviously a problem, but your team or single application may also be problematic.
• • Faced with uncertainty : System design is not like writing code, which is a very certain thing. The input and output of the code module are fixed, just implement the algorithm. However, system design faces uncertainty: uncertain demand, uncertain future development, and uncertain impact of technology selection. You need to make relatively robust decisions in this uncertainty.
• • Fight trade-offs : In system design, it is almost impossible to have an absolutely perfect solution, usually weighing multiple options. For example, it needs to take into account performance and cost, scalability and development complexity, security and user experience, etc. This requires designers to make reasonable decisions and understand the gains and losses behind each choice. For example, choosing a microservice architecture may lead to better scalability, but it will also increase operational complexity and network overhead.
• • Communicate and persuasion : The system architecture is not just for yourself, but more importantly, it is to make all members of the team understand your design, and even non-technical personnel can roughly understand the overall structure. This requires architects to have good document writing and expression skills, and use clear language, diagrams, and processes to convey design intentions. Not only that, when designing a system, it is not something that others can recognize when you make a design. You need to convince others to agree with your design, and sometimes you have to make some compromises.

Of course, in the AI ​​era, some places have improved. For example, if you don’t have enough resources, you can use AI to make up. If you are not good at expressing, let AI help you write prototypes and documents. If you don’t know how to choose, use AI to implement a set of POC (proof of concept) separately, and then compare and see the effect.

How do newcomers improve their system design capabilities?

The system design does not have a fast track, but is more of a long-term accumulation process, which is usually calculated in years. To sum up, there are three key points: "Look more", "Practice more" and "Review more".

"Look more": Accumulation architecture pattern

is to learn classic architecture cases and know what architectural design methods are available. There are many systematic design interview questions and analysis online, which are all good learning materials. You can also look at open source projects to see how complex open source projects work, pay more attention to:

• • How do these systems split functional modules?
• • How do modules communicate?
• • Why did they choose a specific technology or architecture solution?
• • What challenges did you encounter in practice and how did you solve them? For example, how did Weibo 8 stars cheat?

Suggest some learning resources:

• • System Design Interview Alex Xu systematically introduces common architectural patterns
• • High Scalability Website, which contains a large number of real system architecture cases https://highscalability.com/
• • Open source projects Archive documentation such as Kubernetes, Kafka, and Elasticsearch

"Multiple practice": From imitation to innovation

In addition to theoretical study, it is more important to take the initiative to practice. System design is like system design, even if you train all public system design schemes like AI, you still won’t know what their advantages and disadvantages are in different application scenarios if you don’t try it. Just like microservices are never a technical issue, but an issue related to organizational structure.

For newcomers, it is best to start with imitation, first copy and then improve according to their own ideas.

Some ways you can practice:

1. 1. Architecture restoration exercise : Choose a system or product you are familiar with, try to split the architecture, draw the architecture diagram; write down the functional definition of each module, and clarify how to communicate between modules; and then find someone who understands this system for advice and comparison.
2. 2. Comparative Learning : Go and see open source systems with similar functions, see what the design schemes are different, and what their respective advantages and disadvantages are. For example, compare the architectural differences between Redis and Memcached.
3. 3. Design first : When doing a relatively complex function, go to design first and then start writing code. Even for personal projects, you must ease the easel and write design documents.
4. 4. AI-assisted verification : Describe the modules to be implemented in natural language as much as possible, so that AI (recommended models such as Claude, Gemini, DeepSeek) can be implemented to see where the gap between the results and your expectations is. If it is not what you want, adjust your description repeatedly or continue to split until the AI ​​can achieve the effect you want. This process is actually training your modular thinking.
5. 5. Refactor the existing system : Defactor the existing system and improve in terms of stability, code reuse, and performance. The reconstruction process will give you a deep understanding of the original design issues.
6. 6. Side Project Practical : Go to do side project, first try secondary development based on open source projects, and then try to build a relatively more complex system from scratch. For example, make a simplified version of distributed message queue and a mini version of container orchestration system.

Only by stepping on pits and solving problems in real projects can you truly understand the meaning of various choices in system design.

"Multiple review": Refining wisdom from experience

Every time you finish a project or complete a design, you must review and summarize the process:

• • What was the basis for making decisions? How effective are these decisions later verified?
• • If you do it again, where will you make different choices?
• • What are the difficulties and challenges encountered? How are they overcome?
• • Is there a better technical or architectural choice ignored at the time?
• • What is the feedback from team members? Did they understand and accept your design? If you don't accept it, what is the reason?

Review is an important part of allowing you to extract general methodology from specific practices. Only by repeated summary and continuous optimization can your system design capabilities gradually improve.

AI era should make better use of AI to help you improve your system design level

In the process of learning system design, although AI cannot replace you in the complete design of the architecture, it can become an important tool for you to improve your abilities and help you grow rapidly:

• • Quick data and cases : Let AI help you quickly collect and organize cases and best practices for system design, greatly saving learning costs. For example, you can ask AI: "Explain the application scenarios of the Circuit Breaker mode in microservices."
• • Assisted Practice : You can first let the modules you designed by AI to try to implement to test whether your design ideas are clear and reasonable, and continuously optimize the design from feedback. It's like having a pairing programming partner that's always online.
• • Assisted communication : You can first tell AI about complex designs and ideas. AI will help you simplify expression and generate documents that are easy for team communication and understanding. It can even let AI help you generate description text of the architecture diagram.
• • Assisted Decision Making : When you hesitate between different architectural solutions, let AI help you analyze the advantages and disadvantages of each solution and applicable scenarios, helping you make choices quickly. AI can quickly list trade-offs for various solutions to you.
• • Simulation Scenario : You can let AI simulate different fault scenarios and help you think about the fault-tolerant design of the system. For example, "What will happen to my system if the database hangs?"

The key to improving architecture design capabilities is still people, and AI can help you learn more efficiently, trial and error and iterate more quickly, and accelerate your pace of becoming a qualified system architect.

Written at the end

I was originally planning to write it simply, but I accidentally got too much. There is no way, there is really no shortcut to system design, you can only persist in learning. Of course, it can still speed up if the method is right.

In the AI ​​era, the value of architects was not weakened, but magnified. Because AI makes implementation easier, the quality of the design will more clearly affect the final result. A good architectural design, combined with the implementation capabilities of AI, can increase your productivity by ten or even hundreds of times.

I hope you can master the architectural design as soon as possible, and don't worry about being replaced by AI. Instead, you can use the system design capabilities to make AI work better for you. After all, no matter how powerful AI is, someone needs to tell it what kind of house it wants to build, right?