Every studio using AI right now is running some version of the same experiment: figure out which parts of the work it genuinely improves, and which parts it quietly makes worse.

We've been running that experiment at BrightWire since late 2025. This post is where we landed — what AI does in our workflow, what it doesn't, and why that distinction matters to you as the person paying for the project.

The honest version of "we use AI"

When a studio says "we use AI," it can mean almost anything. It can mean they paste prompts into ChatGPT and call it a process. It can mean they've replaced junior copywriters with generated text and nobody told the clients. It can mean they're genuinely using AI to catch problems faster and deliver more reliable work.

We want to be specific about what it means for us, because you deserve to know what's running on your project and why.

Where AI earns its place: catching problems before they reach you

The part of our work where AI has made the most concrete difference is pre-launch quality assurance.

When we finish building a Shopify store or a web project, there is a review pass that happens before any live URL goes to the client. That pass used to rely entirely on us manually going through the site — checking mobile breakpoints, testing the checkout path, reviewing alt text, reading the Liquid templates for logic problems.

We still do all of that manually. But we now also run the code and templates through Claude Opus 4.8 and ask it to find the things a tired human misses on pass four of a review.

The model goes through the templates systematically. It flags things like: a conditional block in a Shopify section that renders an empty div on certain product configurations, missing schema markup that a human reviewer would skip past, an alt attribute that's present in the template but gets overwritten to blank by a metafield that returns null, a mobile breakpoint where a sticky header overlaps the first content section by eight pixels.

None of these are dramatic failures. They're the kind of thing that ships on a Friday and becomes a support message on a Monday. AI finds them reliably because it does not get fatigued, does not assume something is fine because it looked fine in the last project, and has seen a large enough range of code patterns to recognize when something is unusual.

That's what AI earns in our workflow: a more thorough first pass, before anything reaches you.

Code review and documentation

The second place AI runs in our studio is code review — specifically reviewing our own work for logic consistency, unnecessary complexity, and anything that will be hard to maintain later.

This matters more than it sounds. A Shopify theme that's easy to hand over and extend is worth more to you than a clever build that only the person who wrote it can modify. When we use AI to check our own code, it is partly a quality check and partly a sanity check: does this make sense to something that has no context for why we made this decision? If the answer is no, that's a signal to clean it up or document it better.

Documentation — the handover notes, the Loom script for the walkthrough, the guide you get at the end of every project explaining how to make common edits — is also an area where AI assists us. Not writing it from scratch, but helping us structure it so it makes sense to someone who has never touched Liquid or the theme before.

The human judgment call is still ours: what to explain, what level of detail is right, what the client actually needs to know versus what would overwhelm them. AI helps us turn that judgment into clear written output faster.

Content drafting: assists, does not replace

For clients who ask us to help draft initial content — an About page, a short product description, a set of homepage headlines to react to — we sometimes use AI to generate a first draft.

That draft is always a starting point, never the output. The reason is simple: AI does not know your business, your customers, or the thing that makes your offer different. It knows what a reasonable version of your category looks like. That's useful for getting something on the page to react to. It is not useful as the final word on why someone should buy from you.

We tell clients this when it comes up. If you want content that sounds like you and converts, it comes from a conversation with you, not from a model guessing at your brand.

Where AI does not go: design and brand decisions

This is the part worth being direct about.

We do not use AI to design your store or your site. The visual decisions — layout, hierarchy, how your product pages are structured, how the navigation is organized, what the mobile experience feels like — are made by people.

There are two reasons for that.

The first is practical. AI image generation and layout tools are not at the stage where they produce work that's ready to ship without significant human correction. The output tends to look generic — technically competent but indistinct. For a store that needs to convert a specific type of buyer in a specific market, generic is the wrong result.

The second reason is more fundamental. Design decisions for a business are judgment calls that require understanding what the business is actually trying to do. How prominent should the price be on a product page? Where does the trust signal go — near the Add to Cart button, or near the top? Should this clinic's contact form be minimal or should it include an explanation of what happens after the form is submitted?

These decisions come from understanding your buyers, your category, and your specific conversion problem. That understanding comes from talking to you, reading your brief, looking at what your competitors are doing wrong, and applying experience across enough similar projects to have a real opinion.

AI has none of that context. We do.

Client communication stays human

We also do not use AI to write messages to you.

Every update you receive during a project — the daily message, the design preview note, the clarifying question about something in your brief — is written by us directly. When something goes wrong on a build, the message explaining what happened and what we're doing about it is not generated. It is written by the person who found the problem and is fixing it.

This matters because the main failure mode in agency-client relationships is not technical — it's communication. You do not know what's happening. You are not sure if your feedback was understood. You feel like you are chasing updates. We have built our whole process around eliminating those problems, and using AI to simulate human updates would undermine that directly.

What this means for your project

When you hire BrightWire, here is what AI is and isn't doing on your project:

AI is running on:

Pre-launch code review and QA passes
Finding bugs, logic gaps, and accessibility misses in the build
Helping structure documentation and handover materials

AI is not running on:

Your store or site design
Design decisions about layout, hierarchy, or conversion structure
Any message you receive from us during the project
Brand calls about your positioning, copy direction, or offer framing

The work you are paying for — the judgment, the design, the communication — is still done by people. AI is a QA layer and a documentation assist. That's the honest version.

Why we're telling you this

A lot of studios are vague about AI use because they're not sure how clients will react. We'd rather be specific because you're making a real decision about who builds something important to your business.

If you have questions about how our process works — what we use AI for, how a Shopify build actually runs from brief to handover, or whether our packages are the right fit for what you're building — start a conversation at brightwire.studio or reach us on WhatsApp.

The first conversation is always free and always honest.

— Muhammad Tahir, BrightWire
brightwire.studio

Why We Use AI for QA but Not for Your Store Design