In recent years, if you're a software engineer or aspiring developer, you've probably heard advice like:
- "Grind LeetCode every day"
- "Master algorithms to pass tech interviews"
- "The more code you write, the better"
There’s truth to all of that. But here's the catch: much of what we practice — AI can already do quite well. In fact, sometimes it does it faster and with fewer errors.
As developers, our real value doesn't lie in writing thousands of lines of code. It lies in understanding the problem, designing scalable systems, organizing logic, and — increasingly — knowing how to control and orchestrate AI tools effectively.
Real-World Experience: The AI Comic Studio Project
I'm currently building an open-source project called AI Comic Studio — a lab-stage system that explores how multiple AI agents can collaborate to assist creators in producing comics, from rough ideas to detailed scripts.
Instead of just "using AI to generate content," I designed a system that manages how AI works across multiple steps. I treat AI as part of the system architecture — not just a plugin, but as a worker that needs coordination.
Key Technical Highlights
Layered architecture: From frontend surfaces to an agent orchestration engine and backend APIs, each part is modular and loosely coupled.
State Machine Design Pattern: The AI workflow is managed by a state machine. Agents are only triggered in specific workflow states like
RESEARCHING,SCRIPTING,REVIEWING, ensuring clarity and consistency.Offline-first with cloud sync: Data is stored both in Neon Postgres (cloud) and IndexedDB (local), enabling seamless offline editing and later synchronization.
AI orchestration layer: All AI calls go through an abstraction layer with retry logic, queuing, error handling, and fallback strategies.
Human-in-the-loop design: At key decision points, the system pauses and asks for human approval. This is intentional — AI assists, but doesn’t override human creativity or authority.
Why System Design Still Matters in the AI Era
The more you use AI, the more you realize this: you need solid system design skills.
AI is powerful, but without clear boundaries and orchestration, it turns your codebase into chaos. You need:
- Modular architecture
- Clear API contracts
- Fault-tolerant error handling
- Separation of concerns
- Observability and logging
AI can help implement your logic, but only you can define what that logic is, and how everything fits together.
So... Is LeetCode Useless?
Not at all. But it's just one part of becoming a well-rounded software engineer. If your goal is to lead projects, solve real-world problems, and build production-grade systems, you should invest in:
- System design and architecture
- Understanding async workflows, idempotency, caching, etc.
- Full-stack development pipelines: design, build, test, deploy, operate
- Knowing when and how to use AI effectively — and when not to
Final Thoughts
We’re entering a new era: AI can write code, generate designs, debug issues, and even document features. But that only raises the bar for what it means to be a developer.
We're no longer just "code writers." We are system architects and AI orchestrators.
Learn how to guide AI — don’t just ask it for answers.
And if you're curious about how I designed the AI Comic Studio architecture, state machine, and AI agent collaboration model, I’ll be sharing deeper technical breakdowns in upcoming posts.
AI won’t replace you. But someone who uses AI better than you will.
Stay tuned.
Top comments (2)
This is such a relevant discussion! I completely agree that practicing code remains a cornerstone of personal development for any developer.
While AI tools are incredible for speed and handling boilerplate, the act of manually solving problems and writing code is what builds the mental models and technical intuition needed to use those tools effectively. It is similar to learning math, just because we have calculators does not mean we should not understand the underlying logic.
Practicing code is what allows us to move from vibe coding to actually understanding the ground truth of a system. Without that foundation, it is hard to judge if the AI-generated code is truly safe or performant in the real world. Ultimately, mastering the code is what turns us into production-aware partners with the AI, rather than just passive users.
Yep—AI is a power tool, not a replacement for skill. If you don’t practice, you can ship faster… but you won’t know what you shipped