An AI coding workflow that keeps you doing the thinking

An AI assistant can generate a whole module in a few seconds. The question is no longer "should I use it?" but "how do I use it and still be the person who understands the system best?"

Write the "why" yourself, let the AI write the "how"

Before asking the AI, write down the goal, the constraints and what "done" means. That part is your thinking, and it's also the best prompt you can give.

Goal: add accent-insensitive search to the blog.
Constraints: no paid services, runs in the browser.
Done when: typing "lap trinh" finds "lập trình"; there are unit tests.

The blog in this example is in Vietnamese, where *lập trình* means "programming": readers often type without the accent marks and still expect to find the post.

Read every line before you commit

• Can you explain each function to a colleague?
• Is there a test that proves the behaviour is right?
• Did it add a new library you haven't evaluated?

Use AI to learn, not just to get things done

Once you have a solution, keep asking: "Is there another way? What are the trade-offs?" That's when the AI becomes a mentor instead of just a typing machine.