What Engineers Should Consider in the Age of AI-Generated Code

Introduction

Recently, AI writing code itself has become commonplace.
For simple processes or sample code, you can get something that works reasonably well without a human writing a single line.

On the other hand, do you ever feel a slight sense of hesitation while looking at the code?

"It works, but I don't feel like I should use this as is."

There still seems to be a significant gap between "working" and "being suitable for professional use."

In this article, I will try to organize my own thoughts on what engineers should think about in an era where AI can write code.

AI Has Taken Over "Implementation," But Not "Design"

AI performs the "implementation" part—such as writing and refining code—extremely fast.
In fact, when it comes to the implementation task alone, it is already not uncommon for AI to be faster than humans.

However, the role of considering what to build, what constraints exist, and how it will be operated—and then translating that into a form that "delivers value safely"—still remains with the human.

"Code That Works" Is Different from "Code That Is Okay to Use"

It is not rare for AI-generated code to actually "work."
However, in professional practice, that alone is insufficient.

Can it withstand unexpected inputs? Can you trace the cause when a failure occurs? Can you safely roll it back?
Only when these points are included does code truly become "okay to use" and "operable."

What AI is Not Good At

AI is highly capable, but there are areas where it currently seems to struggle.

First is business-specific assumptions and constraints. Implicit rules and background contexts are not reflected in the code unless they are clearly specified.

Next is imagining the future of operations. Anticipating the impact of changes or behavior during failures is still a human role.

Finally, there is the location of responsibility. Deciding to use that code in production or responding when things go wrong cannot be entrusted to AI.

What Engineers Should Think About in the AI Era

In an era where AI writes code for us, I believe the points engineers should consider are as follows:

• What to leave to AI and what humans should decide.
• Where there is a possibility of things breaking.
• Who takes responsibility.

The weight of judgment and selection is definitely becoming greater than the act of writing code itself.

Summary

I feel that the value of "writing code" itself has decreased compared to before due to AI.

On the other hand, what to build, where it might break, and who takes responsibility.
The importance of considering these things is likely higher than ever before.

AI is very capable, but it is humans who must ultimately decide whether that code is suitable for use in production.

I would be very happy if there are others who feel a similar sense of unease or share these same observations.