Last updated: August 19, 2026
I use AI a lot. I also think most public “AI policies” are either legal cover written by someone who’s never used the tools, or vague reassurance that doesn’t actually say anything. So here’s mine, plainly, the way I’d explain it if you asked me over coffee.
I don’t scrape sites. I look at them.
There’s a difference, and a lot of people selling AI courses right now have blurred it on purpose, because “scrape a competitor’s site and have AI spit out something similar” is an easier sell than “spend twenty years building an eye for what actually works.” I don’t do that. What I do is look, at type, at layout, at the way someone solved a problem I haven’t hit yet, and keep a record of it. You can see the actual list, tools, foundries, real sites I reference, in this post. That’s the whole method, out in the open. Looking at something and being inspired by it is not the same as taking it. If you want to know more about where that line sits for me, ask, I have a lot to say about it.
I don’t steal code.
How I actually write with AI.
I never take someone else’s writing, or a competitor’s post, and run it through AI to come out the other side as mine. That’s not writing, that’s stealing. My process starts with me: a brain dump of what I actually want to say, in my own words, before AI touches any of it. Then I run that through AI to refine it, tighten it, find the sentence I was reaching for. The thinking is always mine first. AI never originates the idea, it helps me say it better.
Graphics and images.
I don’t copy graphics either. When I use stock imagery, some of what’s on sites like Adobe Stock and Unsplash now includes AI-generated content, and I follow their licensing and attribution policies for that content directly, rather than inventing my own rules for images I didn’t make and don’t control the terms of.
Where AI actually saves me time.
I test AI against my own design systems constantly, tokens, components, voice rules, feeding them in and checking whether the output actually holds the brand or falls apart the second nobody’s watching.
The real goal behind all of it: I never want to manually style a Google Slide or a Google Doc again in my life. If a design system is built properly, AI should be able to pick it up and apply it without me sitting there nudging fonts and colours by hand every time. That’s not a hypothetical, it’s the actual thing I’m building toward.
How I use AI inside the site itself, and why it matters.
Every site I build gets structured content underneath it, named fields, defined post types, a real design system, not just a blob of text sitting in a page builder. That structure is what lets AI actually help without guessing. I use a tool called Novamira as the bridge: it gives AI access to a specific field by name, the testimonial’s quote, the pull quote, a page’s title, not blanket access to a raw database it has to interpret on the fly.
That distinction is the whole point. An AI editing “the testimonial field” on a defined post type is a fundamentally different, safer thing than an AI editing “the page” and hoping it lands in the right spot. One is checkable before it ever goes live. The other is a guess.
This is also why building a site properly matters more now, not less. A site with real structure underneath it is a site AI can actually help maintain. A site held together with page-builder markup and no defined fields is a site where AI has nothing solid to grab onto, which means more guessing, not less.
The tools I actually pay for.
I use paid, curated reference tools like Mobbin (for real app and product UI) and Codepen (for how people are actually building motion and interaction), not because they’re free shortcuts, but because they’re the professional version of looking, curated, current, and worth paying for. The full list of what I use and why is public: the bookmarks post.
None of this is a marketing claim. It’s checkable, and I’d rather you ask a follow-up question than take my word for it.
Last updated: August 19, 2026
