AI does the heavy lifting. I make the calls.

What AI does

I work with AI assistants, mainly Claude and ChatGPT. They do the repetitive heavy lifting: writing and rewriting code, drafting copy for me to edit, building tests, researching, and running the checks that catch mistakes before anyone sees them.

That work used to take days. With AI, a first version of a small business site can be ready the same day, and changes take minutes rather than a round of meetings.

What I do

• Decide what we're building, and write the brief.
• Set the rules each project has to follow.
• Review the changes on a preview link before they go live.
• Decide what goes live. Code changes are published when I approve them.
• Look after the site once it's live, and stand behind the result.

The rules are written down

Most of my projects have a rule file that the AI reads before it starts. It covers the facts it must not make up, like prices, testimonials and contact details, plus the design, the security settings and how changes are made. This site's file includes the rule:

"Don't invent facts about Matt, his clients, his projects or any numbers. If copy needs a fact you don't have, ask."

So when a page needs a number, the number comes from somewhere I can point to. The figures on this page come from each project's history on GitHub.

Every change is checked

Code changes don't go straight onto a live site. They follow the same steps:

• It's made on its own branch, away from the live site.
• Automatic checks run on it: broken links, missing image descriptions, page titles, security settings and tests, depending on the project.
• Cloudflare builds a preview link, so I can see the change on a real phone.
• It goes live when I approve it, and it's recorded, so it can be undone.

Scheduled data updates, like LateOne's morning refresh, publish on their own. They have checks of their own, and tell me when a source breaks.

Fast, in numbers

• mattg.com.au: live 1 hour 26 minutes after the repository was created.
• DJ Vital Viper: the new site's first change merged 40 minutes after its repository was created.
• LateOne: 119 updates shipped in its first five days in its own repository.
• Fair Go Finds: more than 250 changes merged in its first 16 days.
• SmokoShare: its first real listing went live the day after the repository was created.

Where AI gets things wrong

AI can sound sure of itself and still be wrong. So facts get checked against their source, not taken on trust. Plan Tally's prices, for example, were checked against each provider's own plan pages, and their Critical Information Summaries where they publish one. SmokoShare's state-by-state lease guide links each state's rules to the page they came from.

The automatic checks catch the rest of the usual slips: a broken link, an image with no description, a page without a title. They run on every change, whether a person or an AI made it.

What it means for you

You get a site that's live within a couple of days once I have what I need from you, sometimes the same day, at a price that reflects the time it actually takes. You deal with one person, from the first chat to keeping it running. And you get a site that's been checked by someone who's accountable for it.