AI 編碼工具加快寫碼，不等於程式碼已可交付

Originally published on Substack. I also write on Medium (profile) and Substack.

AI coding tools have changed how quickly I can get a first version of something on the screen. For a WordPress developer, that means a plugin skeleton, a block, or a tricky query can appear in a few seconds. But the faster the first draft arrives, the more obvious it becomes where the real work sits: checking it.

Sonar's 2026 State of Code Developer Survey puts numbers on this. Among developers who have tried AI coding tools, 72% now use them every day. Yet 96% say they do not fully trust that AI-generated code is functionally correct, and only 48% say they always check AI-assisted code before committing it. Sonar calls the result a verification bottleneck: writing code got faster, but the gains many people expected have not followed, because someone still has to confirm the code works.

That matches my own experience. The time I save typing, I spend reading.

Why review becomes the job

When I write code myself, I understand every line because I made every decision. When an AI agent writes it, the code can look clean and still be wrong in ways that are easy to miss: a missing capability check, an unescaped output, a query that works on a small site and falls apart on a large one. In WordPress, these small misses matter, because plugins run on other people's sites.

The gap between the 96% who doubt AI code and the 48% who always check it is the part worth thinking about. Most of us know we should verify. Fewer of us do it every time.

A simple routine that helps

Here is what I try to follow, and it is not complicated.

1. I keep changes small. A small diff is a diff I can actually understand. A giant AI-generated change invites a quick skim and a hopeful merge.
2. I read the code before I run it. If I cannot explain what a function does, it does not get committed.
3. I test the behavior, not just the syntax. Code that runs without errors is not the same as code that does the right thing.
4. I check security-sensitive spots by hand: input handling, output escaping, permissions, and database queries.
5. I keep the decisions myself. AI can suggest an approach, but I choose the architecture and I own the result.

The takeaway

AI did not remove the hard part of software development. It moved it. The skill that matters more now is not typing speed, it is judgment: knowing what to keep, what to change, and what to throw away.

If AI helps you write code faster, make sure the saved time goes into reviewing it properly. Faster code only counts once it is code you can trust.

I would like to hear how you handle this. Do you review every AI-generated change, or do you rely on tests to catch problems?

Source: Sonar, 2026 State of Code Developer Survey results (community.sonarsource.com).