Earlier this year I was the sole developer on a new WordPress site build. AI tools moved the work along, but only because I spent the project teaching them how to work in my way – correcting mistakes, capturing what I learned, and feeding it back so they wouldn’t repeat them over and over again.
This talk is about that practice: teaching your AI tools your process as you go, so they get better instead of just faster.
Two actual instances from the project as examples:
- A mega-menu nav with a dead click zone where the AI’s first move was a complicated JavaScript event handler – the actual fix was a few CSS utility classes on the link.
- A tangled Tailwind class string that smelled wrong before I could name why – four AI agents running in parallel helped me diagnose it, clean it, and codify a rule that’s held up since.
What I’ve learned and my process improvements go beyond just writing code:
- Asking for HTML output instead of markdown when richer visualizations are helpful.
- Dispatching multiple models to get different viewpoints for comparison and brainstorming solutions.
- Keeping the lessons I’ve taught the tools portable so I don’t lose the learning when I migrate to a new model.
My processes are still evolving and improving. This talk gives a snapshot of where I’m at now and attendees will leave with a new set of tools and perspectives to improve their own AI interactions, grounded in actual project work on a shipped site.
