Someone Still Has to Look
AI writes a lot of my first drafts now. Here is how I keep the decisions, the quality and the responsibility with me.
- Published
- Read time
- 7 min read
I use AI coding tools every day. They make me much faster, but only because I have been strict about what they are for. The tool drafts. I decide, I check, and I am responsible for what ships.
I own the decisions
Before I ask for any code, I write down what the screen is for, who uses it and what it must never do. The AI is good at producing options and very bad at knowing which one your client actually needs. That judgement is the job.
Trust the browser, not the summary
An AI tool will tell you a feature works. Sometimes it is right. I check every change in a real browser, at phone and desktop sizes, and I look at screenshots, because automated checks pass happily while a button sits in the wrong place or a picture never loads. Many of the problems I catch are only visible to a person looking.
- Write the goal and the limits before asking for code
- Review every change as if a junior developer wrote it
- Test in a real browser, at real phone sizes, and look at it
- When one bug is fixed, search for every other place it could be hiding
The tool can write the code. It cannot take the blame.
Write the lessons down
Every time a client corrects something, I write down the reason, not just the fix, in a playbook I keep for myself and my tools. Over time that playbook has become the most valuable thing I own: a record of how I make decisions, so every new project starts from everything the old ones taught me.
