04. The Comeback of the Full Stack Engineer: Reevaluating Skills in the AI Era

The Former Chain of Contempt

For a long time, there was an invisible chain of contempt in the tech circle: Those working on underlying kernels looked down on backend developers, backend developers looked down on frontend developers, and all engineers specializing in a specific field tended to look down on “full stack”.

I once witnessed a senior backend engineer in a cross-departmental technical review meeting end a potentially valuable discussion with the phrase “Frontends don’t understand system complexity”.

The reason sounded sufficient: Human energy is limited. “Full stack” often means “jack of all trades, master of none”. In an era where you needed to read the MySQL source code three times to solve deadlock problems, depth was indeed the only way.

But today, the rules have changed. I recently noticed an interesting phenomenon in interviews: candidates whose resumes claimed mastery of a specific framework (such as “Spring Boot Expert” or “React Senior Developer”) were often less competitive when facing complex system design problems than those candidates who “have written a bit of everything”.

This is not because depth is unimportant, but because AI has redefined the cost of “acquiring depth”.

New Moat: Breadth × Judgment

In the AI era, the way technical depth is acquired has undergone a fundamental change.

• Before: You needed to spend 3 years immersed in the Java ecosystem to know the pitfalls of a configuration item. Those 3 years of experience were your moat.
• Now: If you encounter a configuration item you don’t understand, throw it to AI, and it can immediately explain the principle, give examples, and even list best practices.

This means that the value of simply “holding” certain knowledge is depreciating because the cost of retrieving knowledge approaches zero. Conversely, knowing “where to retrieve” and “how to combine knowledge from different fields” has become the new scarce ability.

A modern high-value engineer’s capability model should be:

1. Breadth (Vision): Knowing there are 10 solutions to a problem (Serverless, Containerization, Edge Computing…), not just the hammer in hand.
2. Judgment (Judgment): Choosing the most suitable one among the 10 solutions based on cost, schedule, and team level.
3. AI Steering Ability (Steering): Using AI to quickly fill in the technical details of the chosen solution.

“Jack of all trades” no longer means “master of none”, but means possessing a broader “possibility space”.

Ability to Acquire Depth > Time Holding Depth

Here is a counter-intuitive conclusion: Do not try to compete with AI in knowledge reserves; compete in the “dynamic loading speed” of knowledge.

I have an excellent engineer in my team who originally wrote Python. Last month, a project required refactoring a core module using Go. He didn’t complain “I don’t know Go”, but used AI to get started with Go’s concurrency model in two days, and completed the refactoring in two weeks, with code quality passing the review of senior Go developers.

If you ask him: “Are you proficient in Go?” He might say no. But he possesses extremely powerful “Meta-Capabilities”: understanding the general principles of computer systems (IO models, memory management, network protocols).

Language is just a dialect, principle is the universal language.

The AI era rewards those who master the universal language and can translate it into any dialect via AI at any time.

Risk Warning: The Danger of Knowing Only One Skill

For those engineers still clinging to “I am a backend” or “I am an Android developer”, the alarm bell has rung.

When AI can write code in any language at the level of an intermediate engineer, the irreplaceability of a single skill is declining sharply. When managers are laying off staff, the priority to keep is definitely the person who “can fix frontend bugs, repair databases, and optimize CI/CD processes along the way”.

This is not a value judgment of “how it should be”, but a realistic choice made by many organizations under huge pressure.

This doesn’t mean you have to be Superman. It means you need to overcome the fear of unfamiliar domains. What stops you from fixing frontend code is often not a technical barrier, but a psychological one: “This is not my job”.

Conclusion: The Victory of the Generalist

During the Renaissance, Da Vinci was a painter, an anatomist, and an engineer. People then admired generalists. Later, with the Industrial Revolution, Adam Smith proposed the theory of division of labor, and we were trained to be specialists on assembly lines. Now, AI, as a super tool, is bringing us back to the “New Renaissance Era”.

In your future career, please try to be a “T-shaped Talent”: Still keep a deep “vertical line” (your core expertise), but make sure to extend that “horizontal line” (your breadth) infinitely.

Because in the AI era, that horizontal line determines how high you can fly; while that vertical line only determines how steady you are when you land.